From f83557a2eece7917a70e30ee7f617cc7983ce676 Mon Sep 17 00:00:00 2001
From: Andrewwango <andrew.wang27@gmail.com>
Date: Fri, 19 Jul 2024 12:53:06 +0100
Subject: [PATCH] add tomography splitting loss example

---
 demo_splitting_loss_tomography.ipynb          | 379 ++++++++++++
 docs/demo_splitting_loss_tomography.html      | 540 ++++++++++++++++++
 .../figure-html/cell-10-output-1.png          | Bin 0 -> 10068 bytes
 .../figure-html/cell-11-output-1.png          | Bin 0 -> 9727 bytes
 .../figure-html/cell-9-output-1.png           | Bin 0 -> 9916 bytes
 docs/search.json                              |  26 +-
 6 files changed, 939 insertions(+), 6 deletions(-)
 create mode 100644 demo_splitting_loss_tomography.ipynb
 create mode 100644 docs/demo_splitting_loss_tomography.html
 create mode 100644 docs/demo_splitting_loss_tomography_files/figure-html/cell-10-output-1.png
 create mode 100644 docs/demo_splitting_loss_tomography_files/figure-html/cell-11-output-1.png
 create mode 100644 docs/demo_splitting_loss_tomography_files/figure-html/cell-9-output-1.png

diff --git a/demo_splitting_loss_tomography.ipynb b/demo_splitting_loss_tomography.ipynb
new file mode 100644
index 0000000..d44585b
--- /dev/null
+++ b/demo_splitting_loss_tomography.ipynb
@@ -0,0 +1,379 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Self-supervised learning with measurement splitting\n",
+    "\n",
+    "We demonstrate self-supervised learning with measurement splitting, to train a denoiser network on the MNIST dataset. The physics here is noisy computed tomography, as is the case in [Noise2Inverse](https://arxiv.org/abs/2001.11801). Note this example can also be easily applied to undersampled multicoil MRI as is the case in [SSDU](https://pubmed.ncbi.nlm.nih.gov/32614100/).\n",
+    "\n",
+    "Measurement splitting constructs a ground-truth free loss $\\frac{m}{m_2}\\| y_2 - A_2 \\inversef{y_1}{A_1}\\|^2$ by splitting the measurement and the forward operator using a randomly generated mask.\n",
+    "\n",
+    "See :class:`deepinv.loss.SplittingLoss` for full details."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "c:\\Users\\s2558406\\Documents\\Repos\\deepinv\\venv\\Lib\\site-packages\\tqdm\\auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
+      "  from .autonotebook import tqdm as notebook_tqdm\n"
+     ]
+    }
+   ],
+   "source": [
+    "import deepinv as dinv\n",
+    "from torch.utils.data import DataLoader\n",
+    "import torch\n",
+    "from torchvision import transforms, datasets\n",
+    "from deepinv.models.utils import get_weights_url"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "torch.manual_seed(0)\n",
+    "device = dinv.utils.get_freer_gpu() if torch.cuda.is_available() else \"cpu\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Define loss\n",
+    "\n",
+    "Our implementation has multiple optional parameters that control how the splitting is to be achieved. For example, you can:\n",
+    "\n",
+    "- Use `split_ratio` to set the ratio of pixels used in the forward pass vs the loss;\n",
+    "- Define custom masking methods using a `mask_generator` such as :class:`deepinv.physics.generator.BernoulliSplittingMaskGenerator`;\n",
+    "- Use `eval_n_samples` to set how many realisations of the random mask is used at evaluation time;\n",
+    "- Optionally disable measurement splitting at evaluation time using `eval_split_input` (as is the case in [SSDU](https://pubmed.ncbi.nlm.nih.gov/32614100/)).\n",
+    "- Average over both input and output masks at evaluation time using `eval_split_output`. See :class:`deepinv.loss.SplittingLoss` for details.\n",
+    "\n",
+    "Note that after the model has been defined, the loss must also \"adapt\" the model."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "loss = dinv.loss.SplittingLoss(split_ratio=0.6, eval_split_input=True, eval_n_samples=5)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Prepare data\n",
+    "\n",
+    "We use the `torchvision` MNIST dataset, and use noisy tomography physics (with number of angles equal to the image size) for the forward operator.\n",
+    "\n",
+    ".. note::\n",
+    "\n",
+    "      We use a subset of the whole training set to reduce the computational load of the example.\n",
+    "      We recommend to use the whole set by setting ``train_datapoints=test_datapoints=None`` to get the best results.\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "transform = transforms.Compose([transforms.ToTensor()])\n",
+    "\n",
+    "train_dataset = datasets.MNIST(\n",
+    "    root=\".\", train=True, transform=transform, download=True\n",
+    ")\n",
+    "test_dataset = datasets.MNIST(\n",
+    "    root=\".\", train=False, transform=transform, download=True\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Dataset has been saved in MNIST\n"
+     ]
+    }
+   ],
+   "source": [
+    "physics = dinv.physics.Tomography(\n",
+    "    angles=28, \n",
+    "    img_width=28, \n",
+    "    noise_model=dinv.physics.noise.GaussianNoise(0.1)\n",
+    ")\n",
+    "\n",
+    "deepinv_datasets_path = dinv.datasets.generate_dataset(\n",
+    "    train_dataset=train_dataset,\n",
+    "    test_dataset=test_dataset,\n",
+    "    physics=physics,\n",
+    "    device=device,\n",
+    "    save_dir=\"MNIST\",\n",
+    "    train_datapoints=100,\n",
+    "    test_datapoints=10,\n",
+    ")\n",
+    "\n",
+    "train_dataset = dinv.datasets.HDF5Dataset(path=deepinv_datasets_path, train=True)\n",
+    "test_dataset = dinv.datasets.HDF5Dataset(path=deepinv_datasets_path, train=False)\n",
+    "\n",
+    "train_dataloader = DataLoader(train_dataset, shuffle=True)\n",
+    "test_dataloader = DataLoader(test_dataset,  shuffle=False)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Define model\n",
+    "\n",
+    "We use a simple U-Net architecture with 2 scales as the denoiser network. \n",
+    "\n",
+    "To reduce training time, we use a pretrained model. Here we demonstrate training with 100 images for 1 epoch, after having loaded a pretrained model trained that was with 1000 images for 20 epochs.\n",
+    "\n",
+    ".. note::\n",
+    "\n",
+    "      When using the splitting loss, the model must be \"adapted\" by the loss, as its forward pass takes only a subset of the pixels, not the full image."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "model = dinv.models.ArtifactRemoval(\n",
+    "    dinv.models.UNet(in_channels=1, out_channels=1, scales=2).to(device), pinv=True\n",
+    ")\n",
+    "model = loss.adapt_model(model)\n",
+    "\n",
+    "optimizer = torch.optim.Adam(model.parameters(), lr=1e-3, weight_decay=1e-8)\n",
+    "\n",
+    "# Load pretrained model\n",
+    "file_name = \"demo_measplit_mnist_tomography.pth\"\n",
+    "url = get_weights_url(model_name=\"measplit\", file_name=file_name)\n",
+    "ckpt = torch.hub.load_state_dict_from_url(\n",
+    "    url, map_location=lambda storage, loc: storage, file_name=file_name\n",
+    ")\n",
+    "\n",
+    "model.load_state_dict(ckpt[\"state_dict\"])\n",
+    "optimizer.load_state_dict(ckpt[\"optimizer\"])"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Train and test network"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The model has 444737 trainable parameters\n",
+      "Train epoch 0: TotalLoss=0.032, PSNR=29.155\n",
+      "Train epoch 1: TotalLoss=0.035, PSNR=28.895\n",
+      "Train epoch 2: TotalLoss=0.035, PSNR=28.837\n"
+     ]
+    }
+   ],
+   "source": [
+    "trainer = dinv.Trainer(\n",
+    "    model=model,\n",
+    "    physics=physics,\n",
+    "    epochs=3,\n",
+    "    losses=loss,\n",
+    "    optimizer=optimizer,\n",
+    "    device=device,\n",
+    "    train_dataloader=train_dataloader,\n",
+    "    plot_images=False,\n",
+    "    save_path=None,\n",
+    "    verbose=True,\n",
+    "    show_progress_bar=False,\n",
+    "    wandb_vis=False,\n",
+    ")\n",
+    "\n",
+    "model = trainer.train()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Test and visualise the model outputs using a small test set. We set the output to average over 5 iterations of random mask realisations. The trained model improves on the no-learning reconstruction by ~7dB."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmsAAAC+CAYAAABqIEhRAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/H5lhTAAAACXBIWXMAAA9hAAAPYQGoP6dpAAAmKUlEQVR4nO3de4xdVfXA8VUVodDSOy0gFCn2DiagEWRGCCIYkakxvsEpxMTEaNKZPxBNTOhYEzUhxmbmH6N/mEyb+J9/lBk1RJRKpxACKlqmyEN5zm1py6M8Ope2lIeP+f1B7v2du/ZaZ5/de2fmzPT7SZr07Hv2ed1z75y791p7L5mZmZkRAAAAlNK75vsAAAAA4ONhDQAAoMR4WAMAACgxHtYAAABKjIc1AACAEuNhDQAAoMR4WAMAACgxHtYAAABKbNE8rHV3d8v4+Ph8HwYAAEBHLZqHtdlWq9Xm+xAAAMAJiIe1goaGhmT37t3zfRjokN7eXlmyZInU6/Vm2e7du6W3tzd5W0uWLOngkdnq9XpwvACAE8OifVjr6uqSkZER6e3tle7u7paWse7ubhkaGpLu7m7p6upqeQjL/uEdGRmRwcFBGRwclPHxcVm/fv1x/TFHOfX19cnQ0NB8H0YhlUpFJicnpVKpzPehYJY1vrPWrVsn69atk97eXkI8UEr1el3Wr1/f/Fva29srExMTIvLO38/u7u7mv66urpZlGj/SLNqHtXq9LtVqVSYnJ6Wnp6fly65Wq8m6detkampKNm3aJOvXr8/d1ujoqPT19cnY2JhMTk7O9qFjjgwNDcltt922YFqrenp65vsQMEeGh4dlx44dsmPHDtm5cyct+yidWq0ma9eubf4tnZ6elq1btzZf37hxo0xNTcnU1JSMjY1JvV5vLk9NTfF9lmjRPqyJvNNyIiJy2WWXydTUlPnaxo0bpVarLZg/2OiclStXysDAgNu6Nj4+Lr29vdLb2yuDg4OFtjk+Pt785Zj9gdD49dnd3d385SkizeXe3l6p1Wq5LcLZVt+89bZs2dI87kZrMPf3wlWpVKSvr6/lvgHm29DQkAwMDMjAwECzrKenp/m3FZ21qB/WUrqM6F46MW3atMlsXdu9e7ds3rxZdu7c2ex+jHWZ7t69W4aGhmRyclKmpqZkdHS0+drg4GDzF2a2JbdWq8nw8LDs3LlTqtVqbotwlrdevV5vHsPWrVulXq/TfbrATUxMyG233Sb9/f0i4v8gmJiYaJZnH9C9Hx3eA3+tVmuW0V0Fz/j4eOEfsWjfon5Yy9P4AhoZGQmaYxtfcjt27GiWVSoVWuAWoUqlIgMDA7J58+aW8m3btsmNN97YfMjZtGmTbNmyJXdb27ZtE5F3WtHWrVsntVqt2RrS+LXZ09Mj9Xq95T4aHR1teZjKaxHO8tZbuXJl8/+HDh3KPWaU04YNG6S3t1e6urpkeHhY9uzZI9Vq1f1BUKvVZP369c3ysbExqVQquT86vAf+8fFx6evro7sKrsaDfbVanecjOXG8Z74PYL6Mjo7Kgw8+KCIiY2NjzfKBgQG59tprpVqtSrVabf4RXbdunWzYsEGq1ars3LmTlopFZNOmTc3Yi3asWrVK+vv7ZXh4OHhtZGREdu3aZdbTX3hF7y1rvUqlIpVKpZkIk40hwcKxdetW6e/vl4mJCRkcHGy+19kfBCLS/EGwe/duueGGG5rrNe4p60fH2rVrm/eo9cDf09Mj69evl1qtJoODg3RrIdC4v2q1WvP/je+cvr4+8zsQ7Vk0D2u6BWJmZqb5/40bNwbrZ7uoipTrvnksHpVKRW644YaWL5gbb7xR1q9fLwMDA83WiBtuuCF3O/39/bJu3TrZtGlTsyV25cqVMjEx0QwWr9frs5rZ12gtyf4AwcLV19cn1WpVRkZGZOPGje4PguPtqrQe+Pv6+mRycrKZAd94cASy+vr6ZHR0tHkvTk5OypYtW1p6pNA5J2w3KJA1PDzcEsDd09Mjw8PDzdgdEf9BvqFarcrw8LBce+21LfFBfX19zTigRuvsbKlUKjI+Pi5LlixpptITc7SwjY6OyubNm6Ver0t/f7+Mj483u9EboRn9/f0tsZeN8htvvFFGR0eb5UV+dDRaSzZu3CibNm1yW4RxYhsdHZUtW7ZEw0PQITMnoBP0tHEC2Lhx48zw8HBzeceOHTPVanUejwipenp6ZsbGxlrKBgYGZgYGBmZmZmZmxsbGZnp6emZ6enpm+vr6Zqanp5vl1Wp1plqtzvT19c1MTU0F5Y1tzMy0fg8ODw+3bN/aTk9Pz8yOHTtm7byx8ExPT8/09/c37xfr3p2ZmZmZnJzk726blszMZPoLASxoQ0NDsmrVqmbXf3ZMQQDAwsTDGrDIDA4OysTERDMeaevWrWT0AcACxsMaAABAiZFgAAAAUGI8rAEAAJRY4XHWTjrppNzXdW9qdh5D/dq73/3uluX//e9/ucvZbVnbyy7Htv2ud72rrdfz1o31KOtj++9//5u7fh69b72tdrYNAADKg5Y1AACAEls0MxhgfmXno2w45ZRTCte3WgJ162HDsWPHgrKTTz45KPNaRa0WUK9V9D3vCT8ip556alA2PT1t1l+2bFlQZp2rd6y6VdnjHb9V3zqn//znP2Z9q9yqL2K/By+//HJQ9t73vtesv3z58qDswIED5roiIkuXLnVf06xrrlu68xTNwyr6fon477l137ebB2Ydl7dN67isY0o5V+s+8nprrO3++9//Ntctelwp5/r222+b63rbBtpV5PNNyxoAAECJHXfLWuxJMPuLR/+q1b9m9Ov6l7tePxaXVvQ1kfA8YvFyKevG4sr0vvPi5fJiAkXSWgkAAMDCQcsaAABAifGwBgAAUGIkGKAjUoYKsYKNrTKvC9sLcNe8rmHrWL0AZivY2Nq/l0xhbTcl2N0qt47pzTffNOtbQftFj0nEThrwjtVKsrASLLz6sZAFLSWQfDYSNazj9fZjbbfdYHWvftHj8upbQftWWUpSjPWepwT9e8dqbde6v717zktiAsqm8MNabKw0/cc27493bFu6buxLLfvh1h8+vW29Lf3FEIsjyzuv1C/fWBxadt/6AUH/wSZLCQCAxYmfFQAAACXGwxoAAECJ8bAGAABQYscds5YSDKzjvHQcWGx8Mi1vPs9YTFps2zrGLS+GLRZrF5vzVMeh5cXDeQHwAABgcSMbFB2Rku1VdABf7+HV2q7148HLGrWSMaysR5H2H5KtqZWOHj0alHk/IqxrUDTbTsS+BlbmqHetrMxTa7otEftcrePyzvXw4cNmuSflB6N1HClZh7MxBVO7SUEpn6+U6aqKTnflHX/RqeO8Y7LKUzKIrXVJwMJCRzcoAABAiRVuWYt1VeZNvRSrq19PGbNL7zs2dIceYiS1yzW7vdh0U/o8YseS14Ubm16K6aYAAFicaFkDAAAoMR7WAAAASowEA3SEFVz+1ltvmetaQcFW0HpKsLe1L+uYPN6+rO0uXbo0KLOmWvK2awXzewH+VkhA0TIRexovK2nCSxqwtvvyyy+b665YscIs11KmOcqTErReVLtB+977UDTBQaT4eXnXq+g1mK1rZb2/1rF61yrlPiiaYODti8QDLBTH/bAWi9fKfpBjUzbFPpwnnXRSy7L+A5QXr5WyrnVseUN5xKamSonrs7aX3XdsSBK+dAAAWJzoBgUAACgxHtYAAABKjIc1AACAEutYgkE745HpWC29rOPI8uKzdJyYHpleH4sOIPcCva36sVg7/XosoDfv9di0Walj082FlFHxrUB4HavYcOTIkaDMuie8a2Ltf/ny5ea6lmPHjgVl1vGL2IH71v69Y7USHKy4Sy+ZwrpHi86q4NX3rpV1rJVKJSh79dVXzfpW4kaelAD5orMVpIyq3+7+2+Vtc67iV73zT0kmsKS8L0Xf13aTMYD5RssaAABAifGwBgAAUGI8rAEAAJTYcc8NGosjy8YCpIwnJhIfG+2UU05pWV62bJm7ro7PicXHvf766y3Lhw8fblnOxkjErkG7552N+YptCwAALE60rAEAAJQY002hI958882gzJvCyMoStbK1vNkmrKzBotM6idjHqltQ844rJfO0aP2UllLdsiziZ6Na27WmhfKmBqvX60HZueeea6576NChoMzKnPWyPt944w2zvBOKZkimvA+zNV1TUd45Fd2ul01qfW6s+9vLsLSOq2iZx5oiTcT+jrCOazaycYG5RMsaAABAiRVuWYuN86Xlxazpuvp13XLQ1dXVspyNUdPbi83Pqel96TGkzjzzzJblgwcPNv+vWx1060Zs3ynjz8XmBk2dBBsAACwM/IUHAAAoMR7WAAAASowEA3SEntZLxA+6t8qtoP+UoP3Vq1cHZdZURyJht7qIn8zw4osvBmVWgLwXoG8Fe1vBzl4AdNHr6u3fCuK2EgG8bnQrmWHv3r3muqeddlpQZr2vXuKHl5Disa5tStB6Sn3r/bGumTdFmvf+WKxgeuu4vHu2aEhEyhROKYrW946/6PmL2OeakrgxV1NzAe067nHWYnNT5s0NqtfVX9L6j6n+46i/7LPHEvui0HVjY7rpL9/zzz+/+f/TTz+95bXnnnsud9uxL5G8uDR9vfVxku0EAMDiRDcoAABAifGwBgAAUGKFu0F112Vs6qTs+rp7T3cf6m5PHSejp4zS3YvegKAiYffg22+/3bKsj013k+pu0Ozr73vf+1pe0921zz77bMuynspKd23mTYXF9FIAAJyYSDBAR1hB695DtH749ugH64YPfOADQdnHP/7xoOzyyy8361tB89bxi4g88sgjQdnzzz8flHmBylYQdUpQsxW0bwXoe6P/63EDRexZCbwZHCxesHzRwHDvffXKPSmj4heN6fTqW/fM1772taDsG9/4hlnfmsnh4YcfNte98847g7IHH3wwKDt69KhZ3zqHop85ETuBxfqxmPID0vocpMy64SVuFK3vHetCSzDo7+8PyjZs2GCua31PWQk/IiK//vWvgzIrueqZZ56JHSJmCd2gAAAAJcbDGgAAQIkV7gZNncYp+7qOUdPjX+lYr9iwFHndJXpdveyN8dSgm8tjw4xk6fPS5zE1NdWyrJuk8+Ln9HHlTU0FAAAWD1rWAAAASoyHNQAAgBJbMlOw/0x36elu0LxuuU53g3oZLda6qd2gse7ebDeozhDT6x45cqRlOdYNqmWvaWo3qJWZN5suvPDCoMzL9rKyzaxsNa9+d3d3UHbzzTcHZVdccYVZ38pmtKZ1EgmHWxGxMwtfeukls751DlZmnBdWYL2P1jRa3v6tbNInn3wyKPvzn/9s1r/33nsLHZOI/b5aU3N5XzlW/aefftpcV8TO0PSy+4qGCXj1ly1bFpTdfffdQdnatWsL79+6NiL2PfPaa68FZS+//LJZ37qO1vF7GcRWZrB1rF4GsfVZ0rO7iIj85Cc/Mes//vjjQZk3NVXRqdtSsj7zvpfnO3u0VqsFZVZ2fCfov18iIv/85z9nZV+z4cCBA0HZyMiIua6VbT2Xinw/FY5ZW7VqVcuyThvXcWTZPxKp00fpPwb6j6v+ksn+oYt9mPQHObZ+7KE0S6eXn3XWWS3L+g3RHzz9YJDdlz4OfR7eFz8AAFjY6AYFAAAoMR7WAAAASoyHNQAAgBIrHLOmA5V1fJYOis0GtOrgVh33pevq2C0dMKtj3LLb04GoOk4sNu1KLMEguy99DXRgrU6s0NdBn8cTTzzRspyNA9Tr6jhAvTzXrGBUL2jSCha2gnqtAHIRkX/9619B2e9+97tC64nYwdorVqww161Wq0GZlUyxevVqs74V7GwluXjnagWbv/LKK0HZ2WefbdY/55xzgjI9p62IH6x+//33B2VeYLp1rtY0S944id40Vh4rKcObVsqKNbXW9e5Z6/7+9re/HZRddtllZn0rwP7MM8801/3MZz4TlF1yySVBmZVoI2KfQ9HptkTiY2g2WNOWidgJMNbxP/XUU2b9zZs3B2XePVN0aqn5TgzoFGtqqYsvvthc10rUuOiii8x1e3p6grJPfepTQZmXtLV///6g7LzzzjPXLcr67vO+p6zvOcu+ffvM8vlOMCiCljUAAIAS42ENAACgxHhYAwAAKLHCMWt67DM9R+YZZ5zRspyNwdFjsun4JD24YixuTMf8ZGPadGyCjnfT+9ZxZnr9vPHNdJ+6rqtjs/TghTrmRMew7Nmzp/n/lStXtrymx72LDfYLAAAWJlrWAAAASozmGHTEoUOHgjKdLdtQNFvLysATsVsRb7/99qBMZ+M2WJllXsuklaVpZX5a0/mI2NNVWVNAeddq+fLlQZl1/FaGp4jITTfdFJRZs1288MILZn3dKi7ivy/WNFTW1Ede1qd3DTztTitkZSV72aBWhuQDDzwQlP3tb38rXN+6NiIio6OjQZk1hZNuXW+w7mWr/po1a8z61r1sZVZbWZ8iIr/5zW8K7d+bSsy6Lu2+r56i05CVxc6dOwuVebZv3154XWuEgY9+9KPmupOTk0GZlxldlDVCgJdBbGW+6t4okXDKx4WEljUAAIASK9yypn+t6adWHW+V/SWpf73o8Xl064VuOYjFoWW3r+PIYuOu6daA2PhxWbF5Q/UYU3rbV155Zcvy4OBgy3J27Jrdu3e3vPbqq6+2LFtjbwEAgIWPljUAAIAS42ENAACgxEgwQEdYgfheULAV1GsFvVsB6159K4BZDwmTWl/Eni5Jd0GLhEPZNBQNdvaC7q1gcWsaF29qFyuwXU9rJiLy8MMPm/WtqaW898Ways1KAvCm1koJDPek3HNFy0TscAhrXW+qppRAdmsb1v5ffPHFwtu09u8FW1vX0EpqGRgYMOtbyS5WgoIX7G6da0qCQcr7Ct/09HRQds899xSun5L4UNRXv/pVs9xKhnj00UeDsm3btnX8mOZK4Yc1nZWm5yLUX8DZL3n9h1CPyebN9+XRH7zsB1n/cdDLsTHc9B8iXT97njpbRW9Lf8HpL1cdZ/bJT36yZfmCCy4Qz3333ee+BgAAFg+6QQEAAEqMhzUAAIASK9wNqqdG8uJOGrLdiXqoDR2fpAd3fOmll4oeVrB9HdsQG/ZDx0foGKG8aZx0jE5syic9mOidd97Zsqz7+D/xiU80/68HqtTn5cURAQCAhY0EA3RESoKAFSxslXlBwVbQulXmjYhvBTBbAdQidrC3da5WIL7HeqhPedi2Rub2Am+t2RKseMe9e/ea9a3zT5mZwjpXL5HAG9Hfk3LPtBtg3u49W3SbIvZ1sK6jdc+nHFdKfT2vsYjID37wA7O+lSzzs5/9LCjzYpWtc/XuDete9M7LkpK4gLl11llnBWW//OUvzXWt76lbb701KLNm2lko6AYFAAAoMR7WAAAASqxwN+iKFStalnVXhp4IOtvVFGtq1sOC6G4q3Vye19WgX4sN1aGPTTfB67i0bLO7rqvX1UN76HG/9NAduqn/ySefbP7/mmuuaXlNXzOa8wEAWJxoWQMAACgxHtYAAABKjGxQdERKJp+VDWjV97q7rWwxK5vTG0rF2q6XoWhtN6W+laVkbdPLsDz33HODsquvvjoou/DCC836Tz/9dFD2yCOPBGW6y74hJbPOmm7K6p73tulN1dQJsxEmYN0HndiPdR1SpraaDTfddFNQduaZZ5rrWpnFf/jDH4Iy7/itYaE6cX9iYUm556ypsbJhRItB4Yc1PVzB888/37Ks/9hmP4j6A6X/sOkPbWxMMf3GZIc9SJ1fMHYsejlvTDc9j+Trr7/esqxj8fRwDXpfBw8ebP7//vvvb3ntkksuaVn2bmIAALCw0Q0KAABQYjysAQAAlBgPawAAACVWOGZNx4npKUV0XFpeHFlsWU+Ro4OvdaxYvV53jyNG7zvvPPT6el09jpoOcNdBwjogO2/KoQMHDrQs62tywQUXuHXnghU07wUQW8kE1rl79a3polKmqLFYU+SI2DGQVpn33p188slBmRWE7p3rBz/4waDs+uuvD8r2799v1r/jjjuCsn/84x9BmZ63tsGarsr7jFlB8N51LVq/U4pu20sQKJpMkFI/RcrUVta9ZH0WvHvu8ssvD8q++c1vBmXeFG0//vGPgzI9pqSIyKmnnmrWT5l6rd33FeWQnQe74fvf/37h+l/5yleCsscee6ydQyodWtYAAABKjIc1AACAEuNhDQAAoMQKx6zp2BMdt5MXE6BjjGJxYjqWoVKptCzrcdiygyjqOUo1fZyxuCb9era+fk0fd97YcyJhvIU+tux10XEczz77bMuyNSApAABY+JjBAB1hBaJ7AcjWaORWgoA3wLEV4J4SwG0FVutBn/NYx+UF3Vs/BqxzXbVqlVm/p6cnKLMCs5966imzvpVgYAX9W0kbIv57aLECw60EC483Sr0nZQaBlJkU2tGJJImiyQTevqzzSvnMXHfddUGZdc/dc889Zn1rtgLrnLx7K+UaFk0c8LZJ4kE5fO5znwvKrFlddu7cadb/61//2vFjKhu6QQEAAEqMhzUAAIASO+65QXWzsu4CyXb16CZ43eUSm2NTdzudccYZLcsrV65s/n/58uUtr+nxp44ePdqyrM9Dn6eeVDjbHaCbaXU3TqwLKTZWXfa66Jg0fQ11DBsAAFgcaFkDAAAoMR7WAAAASoxsUHSE7roW8bM5rQwsb92i9a0MTy/T6/XXXy+8L90NLmJnTnpT91hZaFZmnh6OpuGKK64Iyl566aWgbGxszKxvnasOBRDxr7+VuepNB2RdK+u+8FgZxXnazeRLqV80m9S7D+ZS0Xuuq6vLrH/NNdcEZdb7+NOf/tSsb2X1WtmkXphI0emyRIpnzpL1WQ5e1v1nP/vZoMy6P6ypzETSpihbqAo/rOmbXX+564uVvdD6C1//sdN1dXzWG2+80bKsv9TPOuus5v/PPvvsltfe//73tyzv3bu3Zfm1115rWc7Gv1n7yp6XngtUn1csVk9/qeSNwxYbZiDlAQQAACwc8/8zEAAAAC4e1gAAAEqscDeojlmIdfllh7XQ8Qa6C1Uv6+5A3Y2qux+zr3d3d7e89pGPfKRlWccG6fgdPdSH7jbdt29f8/+6+1afp74msemltLypsGJdqAAAYHEgwQAdYU0r5AUQW+taQfte0LsVQGwFwntTKFnBzt6+rMBVa1/eg7cVa7h69eqg7Etf+pJZ31r3scceC8p27dpl1j9y5EhQpsci9I7TK/eulfXj4vTTTy+0nkja1FYiszMtUcq+ZusHUtHzavecvvvd75rla9asCcomJyeDsoceeqjwvlKmu7LuD+9cOzG9F+bOLbfcYpZfeumlQdn27duDsr/85S8dP6aFguYYAACAEuNhDQAAoMQKd4PqJv9YPFb2dd2srdfV3R+6m0V3Z+lutOzwG3raJb0v3cSvu2nq9XrL8gsvvODuKza9lO5C08uxa3q86wIAgMWDljUAAIASI8EAHaEHLhZJC1rXAyGLhBPdN1gByCmzCljrWqPvi9hB71aAvZU0ISJSqVSCsquuuioo+/KXv2zWt0aD//3vfx+UWYkEIvY1LDrqu0hasLdVbl0/L/EjZRYLj/eeFz2PlKD9lJHyU9Ytegxeooa1LyuB5Vvf+pZZ3/p8/vznPw/KvIG4syMBNFiJOrM1q0C77ws64/Of/3xQ9sMf/tBcV4/wICJy6623dvyYFjJa1gAAAEqscMvac88917KsW0LaiamyhkLI0r9+dCtOdlnPm6hTzvW4avpXvp6uSre4ZH8h6lYDvS39ay72Ky5v7DT9mm6FyBuTDQAALFy0rAEAAJQYD2sAAAAlxsMaAABAiRWOWZuammpZPu2001qW9Zyb2XgrHd8WG6MtNs6aXs7GjumsOD1uWmxOU11/1apVLctLly5t/t/KYMzbthY77+zrsW11IpOuHda18GIRrcwsK+bOOyer3MpA87IOrWO16ovY2ZhW7KF3L+i5aUVEvvCFLwRl1rRMIiJ/+tOfgrJHH300KPMy86zr6mWuWqwsPmu6LpHimZ9eZl4npm/yPiftZn4WXdeLS52tzEfL5ZdfHpQNDw8HZVamsojI7bffHpTdddddQVnKubabdZmyr9l4X5FP/50UEfnFL34RlHnf6X/84x+DsgceeKD9A1tEaFkDAAAoMR7WAAAASoyHNQAAgBIrHLM2PT3dsqxjebJzZoq0xvDo2CXdbx2LYdNxN3qctewo7V6ckkcfix6nTZ/X8uXLzf+LxGPz9HWIxeLlbUvHB+nx4wAAwOLAdFPoCCuw2wrOF7GD3q2g9dhgybFtevWtdb1jtQJi9UO6iMjKlSvN+taUPh/72MeCsmeeecasv3379qDs4MGDQZk3KLKVOGG9V15wv1XuTW1VNHHDm0YsZSBtkfaDy1P2ZwW4W9em3emqvO1ax5pNdsqykgnWrFkTlO3Zs8esb03zY71nXlJNyv1lSZkuql0kGKTxEgSs76m1a9cGZTpRscGbhgr/j25QAACAEuNhDQAAoMSOuxtUj+ukm0ezzd66CVw3c8fm0NSxXHp72dd111csbkyPI6WPRceGHTp0qPn/w4cP525bzyvqjeXVoLuysuelj0N329GcDwDA4kTLGgAAQImRYICOsAJPvaB3K9g4JQDZCmBOSTBoNwjdCqy+9NJLzfof/vCHgzIrQWDbtm1m/b///e9BmZUM4WUSW0HoRWcaELHfFyvBQsTOSLaulfdee/dLJ7QbtG7VT9mmdR9561rvhbWv3t5es/6HPvShoMx6z3/0ox+Z9Z944omgzEoA8t6vdmd7mEtlOIaFpLu72yz37kXte9/7nlnuJR7g/xV+WNMfQN2lp7+Mssv6A6HXjXWTxpazxxLrMo3tO3Ysea8dO3asZVlnUMW+xPKOVR+3/qKc7+mmAADA7KAbFAAAoMR4WAMAACgxHtYAAABKrHDMmo6n0jFTOh4ruxwbKTwWWKzjsWLTUWXFhu6IBZjq17PXQW9bH6c+rlg8XF7cXzuxdQAAYOEiGxQdYT0setPh6LHtvDI9ll/DaaedFpRZWYfeQ3zRzD4RO6P04osvDsq+/vWvm/WXLVsWlP32t78Nyu677z6zvpX5aSWT6DH9GqwfSta5elMH6aQZET9R5uSTTw7KrGmKvPEGUzKCRezz8H4YFk3ASZ3yqijrmnn3nLXuihUrgrKRkRGzvvX5+NWvfhWUWfehSPHPUsrxp2i3/lxOV7WYnX/++UHZXXfdVbj+LbfcEpTdcccdbR3TiYxuUAAAgBLjYQ0AAKDECneD6u4B3aycN0aYfk3XjcW/pYy7prs7YvvS8mLvRPLHj4s1taeOjZZ9PXb9iVkDAGBxomUNAACgxEgwQEdY09l4LY1W0LkVyG9NcSNit45arZResHjK1D9r1qwJyj796U8HZRdddJFZ32rxfOihh4Kyffv2Fa5vJR14wfnWuXpTS1m8ZABL0cSHw4cPm/U7EQTutVYXTSrxrmPRluuUBIWUAP3h4eGgzJrKzKs/NjYWlHm9DFaCQUrLfdEAf+/9TkkQKHpcKdOA4R0DAwNBmfV96Ln33nuDMnqAjh8tawAAACVWuGUtdTyy7HLeWGXHI2UOTU234MTm3NRSfhnExoeLxbxljyU2rhqp6QAALE60rAEAAJQYD2sAAAAlxsMaAABAiRWOWdPxVzqjLC+WKy/GzDwoFVdmTUWUt/2849JZi7pubN/Z2LDYWGepY6Hl1dfXTGfppWT4zQYrZs5736zMT6vMO6eiMY9eHJ+VuXreeeeZ615wwQVB2Re/+MWg7JxzzjHr12q1oMzKkPSyGK3rYmXTvvHGG2b95cuXB2XW9bMyTEXsY/XiOo8cORKUWVOOeVNbpcZdWufhfcbmKgvNuzeta+ad75VXXhmUXXfddYW2KWJP+2XdB977MBvXKiXDM+U+sK43GZ7prrrqqqDs5ptvnocjgYeWNQAAgBLjYQ0AAKDEeFgDAAAoseMeZ03H2OTN7xmL3dLb0rFOKeOy6W3r+IXYtnSclD72bAxRbE5T/XpsrLQ8etvzHaMGAADmBtNNoSOsoGDvwdgqLzpVkYidIGAFyFvB7d6+vGPt7e0NyqwAbi/Af+/evUHZ448/HpR5D+7Hjh0Lyqxg8aNHj5r1retiTSFlnZPHC0y3rmFKYL33fnva+bHj8YLTrfrW/lPqe66//vpC+/KmAnvssceCsl27dhXapkjx9yxlirOU85+Nqa2894Xpj95x9dVXB2XLli0rXH9qaioo876TcHzoBgUAACgxHtYAAABK7Lhj1mJjimWbnWNzg8bGxWmnqVofV0qsnSUbKxY7r1iMWuw6ZLs5dIyarhsbiw4AACxMtKwBAACUGAkG6AgrOLzdjFUvaN8qtwLkvVZSK4DaCuQXEXn11VeDsv379wdlExMTZv277747KDtw4EBQZiVNeOVWYHlKMoV1XVJme/BYsy2kBOF773cK7zyKts5766Vcs6L1vRkIDh48WOi4Hn74YbP+d77znaDs0KFDQZn1fnn7Klomkj4TRRHtvq8pIwrA591z1157bVBm3XM4foUf1lI/gNn19Qel09M05Q0TEutq1K/HhtvIO69Yd643pZC1betY8vblffECAICFjZ8bAAAAJcbDGgAAQInxsAYAAFBiHRu6Qwc9Z+OtYsNjxOLGYrFe2WOJxY3FAk1Tji0Wxxc7r1hsXra+XlcP1RG7RgAAYGFaMlMwnUYHsJf1YS12OrGHNf2wlzcuW2rygn49dc7UrNjD2lzPHdrV1RWUee+FVe5NYWSp1+tBmTUF06mnnmrWf+WVV4KyFStWmOtaU66sXr06KDt8+LBZ38ocLTpFjoid+WllTXr3tZUle+TIkUL7EbF//HjJLEUzHr37wjoH671usM7Nuw5Fp4tKyXBMeR+t65jyA8v6fLz22mvmutY1KDrFm0j752qZrWzSosfl7cu67/OmSZqNTFegyGMY3aAAAAAlxsMaAABAiXVsnDU9eGbeuF+x7kHdPB8bhy1vv3ljlVn0vvK6NmPdmvr1WFOnPu9s/dg1isXqAQCAhYmWNQAAgBJj2Ht0xPT09HwfAk4wb7311nwfAk4wRae3AjqNljUAAIASKzx0BwAAAOYeLWsAAAAlxsMaAABAifGwBgAAUGI8rAEAAJQYD2sAAAAlxsMaAABAifGwBgAAUGI8rAEAAJQYD2sAAAAl9n9pb7GNPeGrtwAAAABJRU5ErkJggg==",
+      "text/plain": [
+       "<Figure size 800x200 with 4 Axes>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Test PSNR: No learning rec.: 24.549+-1.052 | Model: 31.911+-2.220. \n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "(31.911066627502443, 2.219884675922773, 24.548791694641114, 1.0523162766060832)"
+      ]
+     },
+     "execution_count": 8,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "trainer.plot_images = True\n",
+    "trainer.test(test_dataloader, pinv=True)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Demonstrate the effect of not averaging over multiple realisations of the splitting mask at evaluation time, by setting `eval_n_samples=1`. We have a worse performance:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "image/png": "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",
+      "text/plain": [
+       "<Figure size 800x200 with 4 Axes>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Test PSNR: No learning rec.: 24.549+-1.052 | Model: 30.434+-2.487. \n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "(30.434445762634276, 2.486670644991658, 24.548791694641114, 1.0523162766060832)"
+      ]
+     },
+     "execution_count": 9,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "model.eval_n_samples = 1\n",
+    "trainer.test(test_dataloader, pinv=True)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Furthermore, we can disable measurement splitting at evaluation altogether by setting `eval_split_input` to False (this is done in [SSDU](https://pubmed.ncbi.nlm.nih.gov/32614100/)). This generally is worse than MC averaging:"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmsAAAC+CAYAAABqIEhRAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/H5lhTAAAACXBIWXMAAA9hAAAPYQGoP6dpAAAlbElEQVR4nO3dX2wcV/XA8ROgpPnjZu00aZvSlKyDaBVR6LqpSmkRkI2E+E+xkxeeELElqr5Uqo0fEFIFCjbipUJITl4LUmoLRFtBwDYlaguI1Kb/oLSNN2mSNk3SxFvnX9MC/j1Uu7/Zc8+dO5PdtcfO9yNFytydOzs7O969e+859y6ZnZ2dFQAAAGTSB+b7BAAAAOBHYw0AACDDaKwBAABkGI01AACADKOxBgAAkGE01gAAADKMxhoAAECG0VgDAADIsEXTWGtvb5eRkZH5Pg0AAICGWjSNtWYrlUrzfQoAAOAyRGMtob6+PpmcnJzv00CDdHR0yJIlS6RcLlfLJicnpaOjI/WxlixZ0sAzs5XLZed8AQCXh0XbWGttbZXBwUHp6OiQ9vb2mp6x9vZ26evrk/b2dmltba1phEW/eAcHB6Wnp0d6enpkZGREurq6LunLHNlULBalr69vvk8jkVwuJxMTE5LL5eb7VNBklc+srVu3ytatW6Wjo4MQD2RSuVyWrq6u6ndpR0eHjI2Nicj735/t7e3Vf62trTXbdH6ks2gba+VyWfL5vExMTEihUKj5sCuVSrJ161aZmpqS/v5+6erqij3W0NCQFItFGR4elomJiWafOuZIX1+fPPLIIwumt6pQKMz3KWCODAwMyOjoqIyOjsr4+Dg9+8icUqkkGzZsqH6XTk9Py+7du6uP9/b2ytTUlExNTcnw8LCUy+Xq9tTUFJ9nKS3axprI+z0nIiKbN2+Wqakp87He3l4plUoL5gsbjdPW1ibd3d3e3rWRkRHp6OiQjo4O6enpSXTMkZGR6i/H6A+Eyq/P9vb26i9PEalud3R0SKlUiu0Rjvb6xu23a9eu6nlXeoO5vxeuXC4nxWKx5r4B5ltfX590d3dLd3d3taxQKFS/W9FYi7qxlmbIiOGly1N/f7/ZuzY5OSk7d+6U8fHx6vBjaMh0cnJS+vr6ZGJiQqampmRoaKj6WE9PT/UXZrQnt1QqycDAgIyPj0s+n4/tEY7y7Vcul6vnsHv3bimXywyfLnBjY2PyyCOPSGdnp4j4fxCMjY1Vy6MNdN+PDl+Dv1QqVcsYroLPyMhI4h+xqN+ibqzFqXwADQ4OOt2xlQ+50dHRalkul6MHbhHK5XLS3d0tO3furCnfs2ePbN++vdrI6e/vl127dsUea8+ePSLyfi/a1q1bpVQqVXtDKr82C4WClMvlmvtoaGiopjEV1yMc5duvra2t+v/Tp0/HnjOyaceOHdLR0SGtra0yMDAgBw8elHw+7/1BUCqVpKurq1o+PDwsuVwu9keHr8E/MjIixWKR4Sp4VRr2+Xx+ns/k8vGh+T6B+TI0NCTPPPOMiIgMDw9Xy7u7u2XLli2Sz+cln89Xv0S3bt0qO3bskHw+L+Pj4/RULCL9/f3V2It6rF69Wjo7O2VgYMB5bHBwUPbv32/W0x94Se8ta79cLie5XK6aCBONIcHCsXv3buns7JSxsTHp6empvtfRHwQiUv1BMDk5Kdu2bavuV7mnrB8dGzZsqN6jVoO/UChIV1eXlEol6enpYVgLjsr9VSqVqv+vfOYUi0XzMxD1WTSNNd0DMTs7W/1/b2+vs390iCpJuR6bx+KRy+Vk27ZtNR8w27dvl66uLunu7q72Rmzbti32OJ2dnbJ161bp7++v9sS2tbXJ2NhYNVi8XC43NbOv0lsS/QGChatYLEo+n5fBwUHp7e31/iC41KFKq8FfLBZlYmKimgFfaTgCUcViUYaGhqr34sTEhOzatatmRAqNc9kOgwJRAwMDNQHchUJBBgYGqrE7Iv6GfEU+n5eBgQHZsmVLTXxQsVisxgFVemebJZfLycjIiCxZsqSaSk/M0cI2NDQkO3fulHK5LJ2dnTIyMlIdRq+EZnR2dtbEXlbKt2/fLkNDQ9XyJD86Kr0lvb290t/f7+0RxuVtaGhIdu3aFQwPQYPMXoYu05eNy0Bvb+/swMBAdXt0dHQ2n8/P4xkhrUKhMDs8PFxT1t3dPdvd3T07Ozs7Ozw8PFsoFGYLhcJssVicnZ6erpbn8/nZfD4/WywWZ6emppzyyjFmZ2s/BwcGBmqObx2nUCjMjo6ONu11Y+GZnp6e7ezsrN4v1r07Ozs7OzExwfdunZbMzkbGCwEsaH19fbJ69erq0H90TkEAwMJEYw1YZHp6emRsbKwaj7R7924y+gBgAaOxBgAAkGEkGAAAAGQYjTUAAIAMSzzP2hVXXBH7uB5Nja5jqB/74Ac/WLP9v//9L3Y7eizreNHt0LE/8IEP1PV43L6hEWV9bv/9739j94+jn1sfq55jAwCA7KBnDQAAIMMWzQoGmF/R9SgrrrzyysT1rZ5A3XtYcf78eads6dKlTpmvV9TqAfX1in7oQ+6fyPLly52y6elps/7KlSudMuu1+s5V9yr7+M7fqm+9pv/85z9mfavcqi9ivwcnT550yj784Q+b9VtaWpyyo0ePmvuKiCxbtswp812vevOorOM245g+1v3h+/uwzmsuz3++r7VVP825Wp8voeMA9Uhyf9OzBgAAkGGX3LMWaglGf/XpngT9K1E/rn+56/1DcWlJHxNxX0coXi7NvqG4Mv3ccfFycTGBIu41AQAAiwM9awAAABlGYw0AACDDSDBAQ6SZKsQKWrfKfEPYvgB3zTc0bJ3re++9Z+777rvvJnp+XzKFdVzr+X3napVb5/TOO++Y9a2g/aTnJGInDfjO1UqysBIsfPVDIQtaM4LmfZpx3Lk81zSB8Umvq+/8633+ZlyXtPcWkDWJG2uhudL0l23cl3foWLpu6A89+oeo4770sfWx9B9xKI4s7nWlzRQKxaFFn1s3EPQXNllKAAAsTgyDAgAAZBiNNQAAgAyjsQYAAJBhlxyzliZgU8d56Tiw0PxkWtx6nqGYtNCxdYxbXAxbKNYutOapjkOLi4fzBcADAIDFjWxQNISV4efL6ko6ga+v8Wod1/rx4MsatZIxrKxHkfobydbSSmfPnnXKfD8irGtgnb/vmlrXwMoc9V0rK/PUWm5LxH6t1nn5XuvMzIxZ7pMm6zBphmOzllBKwzoH6/72nat1fetNQKp3Cad6n8sn6XOFfqQDWccdDAAAkGGJe9ZCQ5VxSy+F6urH08zZpZ87NHWHnmIk7ZBr9Hih5ab06widS9wQbmh5KZabAgBgcaJnDQAAIMNorAEAAGQYCQZoCCu4/OLFi+a+VrC0FbTuCzS26lvPZZ2Tj++5rOMuW7bMKbOWWvId1wrm9wX4WyEBSctE7GW8rKQJX9KAddyTJ0+a+65atcos19IExl/KcSxplkta6JK+Lt9+Sa9rvctNpbn+ac51sb6vuLxdcmMtFK8V/YMJLdkU+pC+4oorarb1F1BcvFaafa1zi5vKI7Q0VZq4Put40ecOTUnCclMAACxODIMCAABkGI01AACADKOxBgAAkGENSzCoZz4yHault3UcWVx8lo4T0zPT63PRAeS+QG+rfijWTj8eCnyNezy0bFbauenmQppZ8a1AeB2rWHHmzBmnzLonfNfEev6WlhZzX8v58+edMuv8RezAfev5fedqJThYcZe+ZArrHk26qoKvvu9aWeeay+WcslOnTpn1rcSNZqo3zjNNIHuaQPhmxJ82I8A/zWoRzWKdl3XPplkeEcgietYAAAAyjMYaAABAhtFYAwAAyLBLXhs0FEcWjSVIM5+YSHhutCuvvLJme+XKld59dXxOKD7u3LlzNdszMzM129HYh9A1qPd1R2O+QscCAACLEz1rAAAAGcZyU2iId955xynzLWFkZYlaGVy+1SasrMGkyzqJ2Oeqe1DjzitN5mnS+ml6SnXPsog/G9U6rrUslG9psHK57JRdf/315r6nT592yqzMWV/W54ULF8xynzQZivUcU8TOJpzL50qzXFYzet2bca3TZJP6sjmtz4g0mZ+s/IKFgp41AACADEvcsxaa50uLi1nTdfXjuuegtbW1Zjsao6aPF1qfU9PPpeeQWrNmTc328ePHq//XvQ66dyP03GnmnwutDZp2EWwAALAw8A0PAACQYTTWAAAAMowEAzSEXtZLxB90b5VbQf9pgvbXrVvnlFlLHYm4w+oi/mSGN9980ymzAuR9AfpWYLYVAO0Lik56XX3PbwVQW4kAvmF0K5nh0KFD5r4rVqxwyqz31Zf44UtI8bFemy9gPGmAvi8437o+aQL5kyYN+Mqtpdd89esNurekef3NWNrKd382I/EByKJLnmcttDZl3Nqgel/9Ia2/TPWXo/6wj55L6ENJ1w3N6aY/JG+88cbq/6+66qqax15//fXYY4c+sOLi0vT11ufJ2ncAACxODIMCAABkGI01AACADEs8DKqHLkNLJ0X318N7evhQD3vqOBm9ZJQeXvRNCCriDg++++67Ndv63PQwqR4GjT5+zTXX1Dymh2tfe+21mm29lJUe2oxbCovlpQAAuDyRYICGsILWfY1o3fj20Q3rio9+9KNO2ac//Wmn7PbbbzfrW0Hz1vmLiDz//PNO2RtvvOGU+YKarWDvNAHQVtC+FaDvm/1fzxsoYq9K4FvBweJLZnjvvfecMuu1+t5XX7lPvcHlafa17tk777zTKevq6jLr6x+oIu6PuYonnnjCKfvnP//plJ04ccKsnyaZwZI0caPepIE0Kxj4JN03zQoIWdbZ2emU7dixw9zX+pyyEn5ERH75y186ZVZy1YEDB0KniCZhGBQAACDDaKwBAABkWOJh0LTLOEUf10MAev4rHesVmpYibrhE76u3fXM8Veju+tA0I1H6denXMTU1VbOtu6Tj4uf0ecUtTQUAABYPetYAAAAyjMYaAABAhi2ZTTh+pof09DBo3LBco4dBfRkt1r5ph0FDw73RYVCdVaj3PXPmTM12aBhUi17TtMOgVmZeM910001OmS8b1Fouycq289Vvb293yu677z6n7I477jDrW9mM1rJOIu50KyJ2ZpkvM896DVYGmi+swHofrWW0fM9vZZO+/PLLTtnTTz9t1t+3b1+icxKx31draS7fR45V/9VXXzX3FbGzetMsN2Vdc1/91atXO2VWBt3HP/5xs77Fd89Z94x1z/oymK1s3zRZydb9bYV/+LKCret67Ngxp+znP/+5Wf+pp55KdE4i6ZYcs1jHjftcnu+lrEqlklNmZcc3gv7+ErGzkrPq6NGjTtng4KC57zPPPNPs04mVpBmWOGZNf1idPXu2ZlvHkUW/JNIuH6W/DPSHgp6mIPrhEPpj0n+cof1DjdIoPSfb2rVra7b1G6L/8HTDIPpc+jz067C+FAEAwMLHMCgAAECG0VgDAADIMBprAAAAGZY4Zk0HKuv4LB0Uu3LlSvP/Im7cl66rY7d04LGOcYseTwfS6jix0FJHoQSD6HPpa6ADhnVihb4O+nX8+9//rtmOxgHqfXUcoN6ea1Ywqi9o0gp2toJ6rQByEZF//etfTtlvfvObRPuJiLz99ttO2apVq8x98/m8U2YlU6xbt86sbwUwW0kuvtdqBZu/9dZbTtm1115r1r/uuuucMr2mrYjIyZMnzfpWsLdvaSvrtZ4/f94p882T6AtY96l3WaI0cxOeOnXKKfvpT3/qlN1yyy1mfSsZwEoUERHZvHmzU7Zp0yanzPeeW/diaF7M0L7Wezs9PW3Wt17Xpz71KafslVdeMev/+c9/dsrSvNf1Lq2VZdbSUr577qWXXnLKbr75ZnPfQqHglH3uc59zynxJW0eOHHHKbrjhBnPfpKzPPt/nlPU5Zzl8+LBZPt8JBknQswYAAJBhNNYAAAAyjMYaAABAhiWOWdNzn+lJEq+++uqa7WgMjp6TTccnzczM1GyH4sZ0zE80pk3HIOh4N/3cOs5M7x83v5keU9d1dWyWnrxQT+66Zs2amu2DBw9W/9/W1lbzmJ73LjTZLwAAWJjoWQMAAMgwumPQEFa2m86WrbCytaysLCvDVMTuRfztb3/rlOls3AorG9HXM2llaVrZdjrTt8JarspaAsp3rVpaWpwy6/ytDE8RkXvvvdcps1a7sJYDEnF7xUX874u1DJW1hJQv69N3DXzSLDWUJkPQYmVIjo6OOmXj4+NmfSuzLU2GppXtbS2B5Xsu6z72/X1YGfNWBvKGDRvM+g899JBTZl3//fv3m/Wtv0Xf+5o0mzPNclVZZt1fvnvOsnfv3sT7WvecldUrIjIxMeGUWVnNaVgzBPgyiK3MVz0aJeIu+biQ0LMGAACQYYl71vSvHd1q1fFW0V+N+leVXmhY/+rTPQehOLTo8fWvytC8a7o3IDR/XFRo3VA9x5Q+9p133lmz3dPTU7MdnbtmcnKy5jE975P1yxcAACx89KwBAABkGI01AACADCPBAA1hBTCnCfa2gt6tgHVffT0Fi4g7JUza+iL2cknW0kN6KpsKa2ktiy/o3gq2tpZx8S3tYgWx62XNRESee+45s761tJTvfbEC063Abt/SWkmvVUW9SQNJE11E7HAIq74vkD3Na7POQYeOiNjLpvnOIc1yW9Y9Y/0d3H777WZ9axmsZ5991ikbGxsz61vn70vGSHoP1HuvXI6s5cSeeOKJxPXTJD4k9a1vfcsst5IhXnjhBadsz549DT+nuZK4saaz0vQfpP4Ajn7I6y9CPSebb70vH/2HF/0g0n/oejs0h5v+ItL1o69TZ6voY+nYuzfffLNmW8eZffazn63Z3rhxo/g8+eST3scAAMDiwTAoAABAhtFYAwAAyLDEw6B6aSRf3ElFdDhRT7Wh45P05I4nTpxIelrO8XVsRmjaDx2LomOE4pZx0jE6oSWf9GSiv//972u29Rj/Zz7zmer/9aSr+nX54ogAAMDCRoIBGiJNgkDSmed9QcFWALJV5psR3woWt1YFELEDm63XagXi+1iN+jSNbWtmbl/grbVaghXveOjQIbO+9frTrExhvVZfsL212kGcNPdMmtUO6nl+H+u80pxrGkmfyxe0b92LN910k1N2//33m/Wt9/zhhx92ynwJElaiSpoEgWa815h7a9eudcp+8YtfmPta9/KDDz7olFkr7SwUDIMCAABkGI01AACADEs8DLpq1aqabT2UoReCjg41hbqg9bQgephKT+0R1yWuHwtN1aHPTXfh6y756BCQrqv31VN76Hm/9NQdehjo5Zdfrv7/85//fM1j+prRzQ8AwOJEzxoAAECG0VgDAADIMLJB0RBpMvmsbECrvm+428o2s7I5fVOpWMf1ZShax01T38pSso7py7C8/vrrnbK7777bKbOy9UREXn31Vafs+eefd8r0kH2Fb/kki5XFZw3P+47py05MI80SZ/Vmk9ar3mzSNMdN81qtaZl6enqcsmuuucasb91zjz32mFMWmuooiXpfK8tQZde9997rlK1Zs8bc11oaKxpGtBgk/mvR0xW88cYbNdv6yzb6was/nPUXm/6QDs0ppt+YaKp52vUFQ+eit+PmdNPr5507d65mW8fi6RR5/VzHjx+v/v+pp56qeeyTn/xkzbbvJgYAAAsbw6AAAAAZRmMNAAAgw2isAQAAZFjimDUdJ3bx4sWabR2XFhdHFtrWS+To4GsdK1Yul73nEaKfO+516P31vnoeNR1Aq4NZdUB23JJDR48erdnW12Tjxo3eunPBCpr3BYxbyQTWa/fVt5aLsoKV0yQ96Pu5woqBtMp8793SpUudMisA2vdaP/axjzll99xzj1N25MgRs/7jjz/ulD377LNOmV63tsJarsr3N2YFa/uua9L6aff3Be2nCTrPonqXUEpzz23atMkps5Yz88UH/+xnP3PKrPvTl2Bg3V9pkk/qTcbA3Iuug13x/e9/P3H9b3zjG07Ziy++WM8pZQ49awAAABlGYw0AACDDaKwBAABkWOKYNR17ouMV4sb+dYxRKE5s+fLlNdu5XK5mW8/DFp3EUa9RqunzDMU16cej9fVj+rzj5p4TcWMr9LlFr4uOiXrttddqtq0JSQEAwMLHCgZoCCsQ3Uo6ELFny7eCjX0BzFYAcpqgYitYWU/6HMc6L1/QvfVjwHqtq1evNusXCgWnTP8oEBF55ZVXzPpWgoEV9G8lbYj430OLlWRhJVj4+FZR8GnGCgD1HnMuA9bT3PPWvr76VrC2/oEsIrJ//36z/sMPP+yU1ZsM4atv/d3N98oUSO9LX/qSU2at6jI+Pm7W/+tf/9rwc8oahkEBAAAyjMYaAABAhl3y2qC6W1kPgUSHenRXtR5yCa2xqYedrr766prttra26v9bWlpqHtPz+5w9e7ZmW78O/Tr1osbRISjdTauHcUJDSKG56qLXRcek6WuoY9gAAMDiQM8aAABAhtFYAwAAyDCyQdEQeuhaxJ/NaWVg+fZNWt/K8PRlep07dy7xc+lhcBE7c9K3HI6VhWZlsFnZdiIid9xxh1N24sQJp2x4eNisb71WHQog4r/+Vuaqb2kt61pZ94WPlVEcpxnZmM1agiqLSyD57rkvfvGLTpmV1fyjH/3IrG9l9VpTC6XJ0Kw3m5Osz2zwZd1b95wVRvTDH/7QrB+3VONikbixpm92/eGuL1b0QusPfP1lp+vqP+wLFy7UbOsP9bVr11b/f+2119Y89pGPfKRm+9ChQzXbb7/9ds12NP7Neq7o69JrgerXFYrV0x9AcfOwhdY8TdMAAQAACwfDoAAAABlGYw0AACDDEg+D6tiT0JBfdFoLHe+gh1D1th4O1MOoevgx+nh7e3vNY5/4xCdqtnWcho7f0VN96GHTw4cPV/+vh2/169TXJLS8lBa3FFZoCBUAACwOJBigIaxlhXzzzFn7WkH7vqB3K9jYCoT3LaFkLdfkey4rcNV6rjTL4axbt84p+9rXvmbWt/Z98cUXnTLf0j9nzpxxyvRchL7z9JX7rpX14+Kqq65KtJ9IuqWtROY/aLzeJZSaxfr7sN6z733ve2b99evXO2VPP/20U+Zb4sf6+2jW0lzzveQX0nnggQfM8ltvvdUp27t3r1P2l7/8peHntFDQHQMAAJBhNNYAAAAyLPEwqI6JCsVjRR/Xwx56Xz38obvs9XCWHkaLTr+hl13Sz6W7+PUwTblcrtk+duyY97lCy0vpITS9Hbqml7ovAABYPOhZAwAAyDASDNAQeuJikXRB69YM53qh+worQD3NqgLWvtbs+yJ20LsVrG0lTYiI5HI5p+yuu+5yyr7+9a+b9a3Z4B977DGnzEokELGvYZoZ4q1r7Qvgtsqt6+dL/EizioVI/cHlaa5DM57f91zWfWv9zaR5ri1btjhl3/3ud819rff8oYceSrSfiJ1gUK96VzBIe1zU78tf/rJT9oMf/MDcV8/wICLy4IMPNvycFjJ61gAAADIs8U+g119/vWZb94TUE1MV+iWmf/3oXpzotl43cWJiomZbz6umf+Xr5ap0j0s07kz3Guhj6V99oV9xcXOn6cd0L0TcnGwAAGDhomcNAAAgw2isAQAAZBiNNQAAgAxLHLM2NTVVs71ixYqabb3mZjTeSse3heZoC82zprejsWM6K07PmxZa01TXX716dc32smXLqv+3Mhjjjq2FXnf08dCx0mbSNZp1LXyxiFYGlxVz53tNVnl0LdoKX9ahda5WfRE7G9OKPfTdC3ptWhGRr3zlK06ZtSyTiMgf/vAHp+yFF15wys6dO2fWt66rL3PVYi23ZS3XJZI889OXwZd2fdtmZH7O5VJFvueqN0NRr48sYmfWWZnKIiKPPvqoU7Zv3z6nzPd+JT3/Zi3XNd/v6+VIf0+K2BnEvs/03/3ud07Z3/72t/pPbBGhZw0AACDDaKwBAABkGI01AACADEscszY9PV2zrWN5omtmitTG8OjYJT1uHYph03E3ep616CztvjglH30uep42/bpaWlrM/4uEY/P0dQjF4sUdS8cH6fnjAADA4sByU2gIKwHCCs4XsYPeraD1NMvWWMf01bf29Z2rFRCrG+kiIm1tbWb973znO07Zbbfd5pQdOHDArL93716n7Pjx406Zb1JkK3HCeq98weJWuW9pq6SJG75lxNJMpC2SLpB8IS0rZJ2r9T747u+f/OQnTtmGDRucssOHD5v1rWQE67qmSTBo1rJQ1nGblbgAf4KA9Tll3XM6UbHCtwwV/h/DoAAAABlGYw0AACDDLnkYVM/rpLtHo13kurtcd2mH1tDUsVz6eNHH9dBAKG5MzyOlz0XHhp0+fbr6/5mZmdhj63VFfXN5VeihrOjr0uehh+3ozgcAYHGiZw0AACDDSDBAQ1iBp76gdyswOc3s9VYgepoEgzS9kNZzWYH0t956q1l/06ZNTpmVILBnzx6z/t///nenzEqG8GUSR1fcqEi60oCI/b5YCRYidkayda1877XvfvFJuipBI8xl0LzFujbFYtHcd/PmzU6Z9Z5biQQiIqVSySmz3rM0Qf+WNK+/GatVID1rdQwRkY6OjkT177//frPcl3iA/5e4sab/APSQnv5Si27rPxS9b2iYNLQdPZfQkGnouUPnEvfY+fPna7Z11lvoAyfuXPV56w/v+V5uCgAANAfDoAAAABlGYw0AACDDaKwBAABkWOKYNR1PpWOmdDxWdDs0M3kosFjHY4WWo4oKTd0RCjzVj0evgz62Pk99XqF4uLi4v3pi6wAAwMJFNigawmosWpmIIu7cdr4yPZdfxYoVK5wyK+vQ14i3ztXX2LUySm+55Ran7Nvf/rZZf+XKlU7Zr3/9a6fsySefNOtbmZ9WMome06/C+qFkvVbr+om4STMi/kSZpUuXOmXW0lK++QbTZASL1J+hmeY+aIY091xra6tT9uMf/9isb/3d/epXv3LKfBnIvntprqTJ5iTzs3luvPFGp+yPf/xj4voPPPCAU/b444/XdU6XM4ZBAQAAMozGGgAAQIYlHgbVwym6+zlujjD9mK4bin9LM++aHiIKPZcWF3snEj9/XGgIJu3caNHHQ9efrn8AABYnetYAAAAyjAQDNIS1nI2vp9EKOreCqpcvX27Wt3pHrV5KXxayVe471/Xr1ztlX/jCF5yym2++2axv9Xj+4x//cMoOHz6cuL6VdOALzrdeq29pKYsvGcCSNPFhZmbGrF/v8k2NUO85+Oqnuees97y/v98pu+GGGxLXf/TRR819Lc1YRqtZ17VejEj4dXd3O2XW56HPvn37nDKu96WjZw0AACDDEvespZ2PLLodN1fZpUizhqame3BCa25qaX4ZhOaHC8W8Rc8lNK9aFnolAABA49GzBgAAkGE01gAAADKMxhoAAECGJY5Z0/FXOqMsLpYrLsbMPCkVV2YtRRR3/Ljz0lmLum7ouaOxYaG5ztLOhRZXX18znaWXJsOvGayYOd/7ZmV+WmW+15Q05tEXx2dlrvoy6zZu3OiUffWrX3XKrrvuOrN+qVRyyqwMSd+ce9Z1sbJpL1y4YNZvaWlxyqzrZ2WYitjn6ovrPHPmjFNmLX3kW9qqmXGXSWNN0yxrVG9933182223OWX33HOPU2bdxyL2NbeWbrOWBxNJfq18519vxl8zskl9iPV931133eWU3XffffNwJvChZw0AACDDaKwBAABkGI01AACADLvkedZ0jE3c+p6h2C19LB3rlGZeNn1sHesVOpaOk9LnHo0hCq1pqh8PzZUWRx97vmPUAADA3GC5KTREmgBqqzzpUkUidmC1FSBvBVr7nst3rh0dHU6ZFZjtC/A/dOiQU/bSSy85Zb6G+/nz550yK2ng7NmzZn3rulhLSPmCzS2+BAHrGlrJCL6gbt/73SxpgsuTBr373sc0SR3f/OY3nTLrnrWST0REpqamnDLrnqs3uD6UKHapmhH0zzJH8e6++26nbOXKlYnrW/ec7zMJl4ZhUAAAgAyjsQYAAJBhlxyzFppTLNpFHlobNNSdXk8Xtj6vNLF2lmisWOh1hWLUQtchOlylY9R03dBcdAAAYGGiZw0AACDDSDBAQ1jB4fVmrPqC9q1yK0De10tqBb1bgfwiIqdOnXLKjhw54pSNjY2Z9f/0pz85ZUePHnXKfLPRW+VWgkCaZArruqRZ7cHHCni3esV9Pem+99snzetIqt5A9DTPb72PIiLHjh1zyqwVCA4ePGjW7+/vd8qmp6cTP/9cXYO5XD3A91ysYJDOc889Z5Zv2bLFKTt9+nSzT+eykrixlvamju6vh+wavUxT3DQhoaFG/Xhouo241xUazvUtKWQd2zqXuOfyZYYBAICFjWFQAACADKOxBgAAkGE01gAAADKsYVN36GDVaLxVaHqMUNxYKNYrei6huLHQclNpzi0Uxxd6XaHYvGh9va+eqiN0jQAAwMK0ZDZh6o8OYM9qYy30ckKNNd3Yi5uXLW3ygn487ZqpUaHG2lyvHdra2uqU+d4Lq9y3hJGlXC47ZdYSTMuXLzfrv/XWW07ZqlWrzH2tJVfWrVvnlM3MzJj1rczRpEsXidgZe1bWpO++trJkz5w5k+h5ROwfP75kFus1WEkyaZZkst7rCuu11bOOsEi6rMFmLWFkvRfr1693yg4cOGDWt96zNMvBJZXm9VvP5fthXe+1TpMlbH1OxmVAkz2KZkhyfzMMCgAAkGE01gAAADKsYfOs6a7juHm/QsODekgvNA9b3PPGzVVm0c8VN7QZGtbUj4e6OvXrjtYPXaNQrB4AAFiY6FkDAADIMKa9R0NYy9kAzXTx4sX5PgVcZpqVVAKE0LMGAACQYYmn7gAAAMDco2cNAAAgw2isAQAAZBiNNQAAgAyjsQYAAJBhNNYAAAAyjMYaAABAhtFYAwAAyDAaawAAABlGYw0AACDD/g9ELvyyc2Id3wAAAABJRU5ErkJggg==",
+      "text/plain": [
+       "<Figure size 800x200 with 4 Axes>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Test PSNR: No learning rec.: 24.549+-1.052 | Model: 31.003+-2.107. \n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "(31.002875900268556, 2.106733038650352, 24.548791694641114, 1.0523162766060832)"
+      ]
+     },
+     "execution_count": 10,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "model.eval_split_input = False\n",
+    "trainer.test(test_dataloader, pinv=True)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.5"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 4
+}
diff --git a/docs/demo_splitting_loss_tomography.html b/docs/demo_splitting_loss_tomography.html
new file mode 100644
index 0000000..bccb9d3
--- /dev/null
+++ b/docs/demo_splitting_loss_tomography.html
@@ -0,0 +1,540 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"><head>
+
+<meta charset="utf-8">
+<meta name="generator" content="quarto-1.3.450">
+
+<meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes">
+
+
+<title>Self-supervised learning with measurement splitting</title>
+<style>
+code{white-space: pre-wrap;}
+span.smallcaps{font-variant: small-caps;}
+div.columns{display: flex; gap: min(4vw, 1.5em);}
+div.column{flex: auto; overflow-x: auto;}
+div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;}
+ul.task-list{list-style: none;}
+ul.task-list li input[type="checkbox"] {
+  width: 0.8em;
+  margin: 0 0.8em 0.2em -1em; /* quarto-specific, see https://github.com/quarto-dev/quarto-cli/issues/4556 */ 
+  vertical-align: middle;
+}
+/* CSS for syntax highlighting */
+pre > code.sourceCode { white-space: pre; position: relative; }
+pre > code.sourceCode > span { display: inline-block; line-height: 1.25; }
+pre > code.sourceCode > span:empty { height: 1.2em; }
+.sourceCode { overflow: visible; }
+code.sourceCode > span { color: inherit; text-decoration: inherit; }
+div.sourceCode { margin: 1em 0; }
+pre.sourceCode { margin: 0; }
+@media screen {
+div.sourceCode { overflow: auto; }
+}
+@media print {
+pre > code.sourceCode { white-space: pre-wrap; }
+pre > code.sourceCode > span { text-indent: -5em; padding-left: 5em; }
+}
+pre.numberSource code
+  { counter-reset: source-line 0; }
+pre.numberSource code > span
+  { position: relative; left: -4em; counter-increment: source-line; }
+pre.numberSource code > span > a:first-child::before
+  { content: counter(source-line);
+    position: relative; left: -1em; text-align: right; vertical-align: baseline;
+    border: none; display: inline-block;
+    -webkit-touch-callout: none; -webkit-user-select: none;
+    -khtml-user-select: none; -moz-user-select: none;
+    -ms-user-select: none; user-select: none;
+    padding: 0 4px; width: 4em;
+  }
+pre.numberSource { margin-left: 3em;  padding-left: 4px; }
+div.sourceCode
+  {   }
+@media screen {
+pre > code.sourceCode > span > a:first-child::before { text-decoration: underline; }
+}
+</style>
+
+
+<script src="site_libs/quarto-nav/quarto-nav.js"></script>
+<script src="site_libs/clipboard/clipboard.min.js"></script>
+<script src="site_libs/quarto-search/autocomplete.umd.js"></script>
+<script src="site_libs/quarto-search/fuse.min.js"></script>
+<script src="site_libs/quarto-search/quarto-search.js"></script>
+<meta name="quarto:offset" content="./">
+<script src="site_libs/quarto-html/quarto.js"></script>
+<script src="site_libs/quarto-html/popper.min.js"></script>
+<script src="site_libs/quarto-html/tippy.umd.min.js"></script>
+<script src="site_libs/quarto-html/anchor.min.js"></script>
+<link href="site_libs/quarto-html/tippy.css" rel="stylesheet">
+<link href="site_libs/quarto-html/quarto-syntax-highlighting.css" rel="stylesheet" id="quarto-text-highlighting-styles">
+<script src="site_libs/bootstrap/bootstrap.min.js"></script>
+<link href="site_libs/bootstrap/bootstrap-icons.css" rel="stylesheet">
+<link href="site_libs/bootstrap/bootstrap.min.css" rel="stylesheet" id="quarto-bootstrap" data-mode="light">
+<script id="quarto-search-options" type="application/json">{
+  "location": "sidebar",
+  "copy-button": false,
+  "collapse-after": 3,
+  "panel-placement": "start",
+  "type": "textbox",
+  "limit": 20,
+  "language": {
+    "search-no-results-text": "No results",
+    "search-matching-documents-text": "matching documents",
+    "search-copy-link-title": "Copy link to search",
+    "search-hide-matches-text": "Hide additional matches",
+    "search-more-match-text": "more match in this document",
+    "search-more-matches-text": "more matches in this document",
+    "search-clear-button-title": "Clear",
+    "search-detached-cancel-button-title": "Cancel",
+    "search-submit-button-title": "Submit",
+    "search-label": "Search"
+  }
+}</script>
+
+  <script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
+  <script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-chtml-full.js" type="text/javascript"></script>
+
+</head>
+
+<body class="fullcontent">
+
+<div id="quarto-search-results"></div>
+<!-- content -->
+<div id="quarto-content" class="quarto-container page-columns page-rows-contents page-layout-article">
+<!-- sidebar -->
+<!-- margin-sidebar -->
+    
+<!-- main -->
+<main class="content" id="quarto-document-content">
+
+<header id="title-block-header" class="quarto-title-block default">
+<div class="quarto-title">
+<h1 class="title">Self-supervised learning with measurement splitting</h1>
+</div>
+
+
+
+<div class="quarto-title-meta">
+
+    
+  
+    
+  </div>
+  
+
+</header>
+
+<p>We demonstrate self-supervised learning with measurement splitting, to train a denoiser network on the MNIST dataset. The physics here is noisy computed tomography, as is the case in <a href="https://arxiv.org/abs/2001.11801">Noise2Inverse</a>. Note this example can also be easily applied to undersampled multicoil MRI as is the case in <a href="https://pubmed.ncbi.nlm.nih.gov/32614100/">SSDU</a>.</p>
+<p>Measurement splitting constructs a ground-truth free loss <span class="math inline">\(\frac{m}{m_2}\| y_2 - A_2 \inversef{y_1}{A_1}\|^2\)</span> by splitting the measurement and the forward operator using a randomly generated mask.</p>
+<p>See :class:<code>deepinv.loss.SplittingLoss</code> for full details.</p>
+<div class="cell" data-execution_count="1">
+<div class="sourceCode cell-code" id="cb1"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb1-1"><a href="#cb1-1" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> deepinv <span class="im">as</span> dinv</span>
+<span id="cb1-2"><a href="#cb1-2" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> torch.utils.data <span class="im">import</span> DataLoader</span>
+<span id="cb1-3"><a href="#cb1-3" aria-hidden="true" tabindex="-1"></a><span class="im">import</span> torch</span>
+<span id="cb1-4"><a href="#cb1-4" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> torchvision <span class="im">import</span> transforms, datasets</span>
+<span id="cb1-5"><a href="#cb1-5" aria-hidden="true" tabindex="-1"></a><span class="im">from</span> deepinv.models.utils <span class="im">import</span> get_weights_url</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="cell-output cell-output-stderr">
+<pre><code>c:\Users\s2558406\Documents\Repos\deepinv\venv\Lib\site-packages\tqdm\auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html
+  from .autonotebook import tqdm as notebook_tqdm</code></pre>
+</div>
+</div>
+<div class="cell" data-execution_count="2">
+<div class="sourceCode cell-code" id="cb3"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb3-1"><a href="#cb3-1" aria-hidden="true" tabindex="-1"></a>torch.manual_seed(<span class="dv">0</span>)</span>
+<span id="cb3-2"><a href="#cb3-2" aria-hidden="true" tabindex="-1"></a>device <span class="op">=</span> dinv.utils.get_freer_gpu() <span class="cf">if</span> torch.cuda.is_available() <span class="cf">else</span> <span class="st">"cpu"</span></span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</div>
+<section id="define-loss" class="level3">
+<h3 class="anchored" data-anchor-id="define-loss">Define loss</h3>
+<p>Our implementation has multiple optional parameters that control how the splitting is to be achieved. For example, you can:</p>
+<ul>
+<li>Use <code>split_ratio</code> to set the ratio of pixels used in the forward pass vs the loss;</li>
+<li>Define custom masking methods using a <code>mask_generator</code> such as :class:<code>deepinv.physics.generator.BernoulliSplittingMaskGenerator</code>;</li>
+<li>Use <code>eval_n_samples</code> to set how many realisations of the random mask is used at evaluation time;</li>
+<li>Optionally disable measurement splitting at evaluation time using <code>eval_split_input</code> (as is the case in <a href="https://pubmed.ncbi.nlm.nih.gov/32614100/">SSDU</a>).</li>
+<li>Average over both input and output masks at evaluation time using <code>eval_split_output</code>. See :class:<code>deepinv.loss.SplittingLoss</code> for details.</li>
+</ul>
+<p>Note that after the model has been defined, the loss must also “adapt” the model.</p>
+<div class="cell" data-execution_count="3">
+<div class="sourceCode cell-code" id="cb4"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb4-1"><a href="#cb4-1" aria-hidden="true" tabindex="-1"></a>loss <span class="op">=</span> dinv.loss.SplittingLoss(split_ratio<span class="op">=</span><span class="fl">0.6</span>, eval_split_input<span class="op">=</span><span class="va">True</span>, eval_n_samples<span class="op">=</span><span class="dv">5</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</div>
+</section>
+<section id="prepare-data" class="level3">
+<h3 class="anchored" data-anchor-id="prepare-data">Prepare data</h3>
+<p>We use the <code>torchvision</code> MNIST dataset, and use noisy tomography physics (with number of angles equal to the image size) for the forward operator.</p>
+<p>.. note::</p>
+<pre><code>  We use a subset of the whole training set to reduce the computational load of the example.
+  We recommend to use the whole set by setting ``train_datapoints=test_datapoints=None`` to get the best results.</code></pre>
+<div class="cell" data-execution_count="4">
+<div class="sourceCode cell-code" id="cb6"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb6-1"><a href="#cb6-1" aria-hidden="true" tabindex="-1"></a>transform <span class="op">=</span> transforms.Compose([transforms.ToTensor()])</span>
+<span id="cb6-2"><a href="#cb6-2" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb6-3"><a href="#cb6-3" aria-hidden="true" tabindex="-1"></a>train_dataset <span class="op">=</span> datasets.MNIST(</span>
+<span id="cb6-4"><a href="#cb6-4" aria-hidden="true" tabindex="-1"></a>    root<span class="op">=</span><span class="st">"."</span>, train<span class="op">=</span><span class="va">True</span>, transform<span class="op">=</span>transform, download<span class="op">=</span><span class="va">True</span></span>
+<span id="cb6-5"><a href="#cb6-5" aria-hidden="true" tabindex="-1"></a>)</span>
+<span id="cb6-6"><a href="#cb6-6" aria-hidden="true" tabindex="-1"></a>test_dataset <span class="op">=</span> datasets.MNIST(</span>
+<span id="cb6-7"><a href="#cb6-7" aria-hidden="true" tabindex="-1"></a>    root<span class="op">=</span><span class="st">"."</span>, train<span class="op">=</span><span class="va">False</span>, transform<span class="op">=</span>transform, download<span class="op">=</span><span class="va">True</span></span>
+<span id="cb6-8"><a href="#cb6-8" aria-hidden="true" tabindex="-1"></a>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</div>
+<div class="cell" data-execution_count="5">
+<div class="sourceCode cell-code" id="cb7"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb7-1"><a href="#cb7-1" aria-hidden="true" tabindex="-1"></a>physics <span class="op">=</span> dinv.physics.Tomography(</span>
+<span id="cb7-2"><a href="#cb7-2" aria-hidden="true" tabindex="-1"></a>    angles<span class="op">=</span><span class="dv">28</span>, </span>
+<span id="cb7-3"><a href="#cb7-3" aria-hidden="true" tabindex="-1"></a>    img_width<span class="op">=</span><span class="dv">28</span>, </span>
+<span id="cb7-4"><a href="#cb7-4" aria-hidden="true" tabindex="-1"></a>    noise_model<span class="op">=</span>dinv.physics.noise.GaussianNoise(<span class="fl">0.1</span>)</span>
+<span id="cb7-5"><a href="#cb7-5" aria-hidden="true" tabindex="-1"></a>)</span>
+<span id="cb7-6"><a href="#cb7-6" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-7"><a href="#cb7-7" aria-hidden="true" tabindex="-1"></a>deepinv_datasets_path <span class="op">=</span> dinv.datasets.generate_dataset(</span>
+<span id="cb7-8"><a href="#cb7-8" aria-hidden="true" tabindex="-1"></a>    train_dataset<span class="op">=</span>train_dataset,</span>
+<span id="cb7-9"><a href="#cb7-9" aria-hidden="true" tabindex="-1"></a>    test_dataset<span class="op">=</span>test_dataset,</span>
+<span id="cb7-10"><a href="#cb7-10" aria-hidden="true" tabindex="-1"></a>    physics<span class="op">=</span>physics,</span>
+<span id="cb7-11"><a href="#cb7-11" aria-hidden="true" tabindex="-1"></a>    device<span class="op">=</span>device,</span>
+<span id="cb7-12"><a href="#cb7-12" aria-hidden="true" tabindex="-1"></a>    save_dir<span class="op">=</span><span class="st">"MNIST"</span>,</span>
+<span id="cb7-13"><a href="#cb7-13" aria-hidden="true" tabindex="-1"></a>    train_datapoints<span class="op">=</span><span class="dv">100</span>,</span>
+<span id="cb7-14"><a href="#cb7-14" aria-hidden="true" tabindex="-1"></a>    test_datapoints<span class="op">=</span><span class="dv">10</span>,</span>
+<span id="cb7-15"><a href="#cb7-15" aria-hidden="true" tabindex="-1"></a>)</span>
+<span id="cb7-16"><a href="#cb7-16" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-17"><a href="#cb7-17" aria-hidden="true" tabindex="-1"></a>train_dataset <span class="op">=</span> dinv.datasets.HDF5Dataset(path<span class="op">=</span>deepinv_datasets_path, train<span class="op">=</span><span class="va">True</span>)</span>
+<span id="cb7-18"><a href="#cb7-18" aria-hidden="true" tabindex="-1"></a>test_dataset <span class="op">=</span> dinv.datasets.HDF5Dataset(path<span class="op">=</span>deepinv_datasets_path, train<span class="op">=</span><span class="va">False</span>)</span>
+<span id="cb7-19"><a href="#cb7-19" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb7-20"><a href="#cb7-20" aria-hidden="true" tabindex="-1"></a>train_dataloader <span class="op">=</span> DataLoader(train_dataset, shuffle<span class="op">=</span><span class="va">True</span>)</span>
+<span id="cb7-21"><a href="#cb7-21" aria-hidden="true" tabindex="-1"></a>test_dataloader <span class="op">=</span> DataLoader(test_dataset,  shuffle<span class="op">=</span><span class="va">False</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="cell-output cell-output-stdout">
+<pre><code>Dataset has been saved in MNIST</code></pre>
+</div>
+</div>
+</section>
+<section id="define-model" class="level3">
+<h3 class="anchored" data-anchor-id="define-model">Define model</h3>
+<p>We use a simple U-Net architecture with 2 scales as the denoiser network.</p>
+<p>To reduce training time, we use a pretrained model. Here we demonstrate training with 100 images for 1 epoch, after having loaded a pretrained model trained that was with 1000 images for 20 epochs.</p>
+<p>.. note::</p>
+<pre><code>  When using the splitting loss, the model must be "adapted" by the loss, as its forward pass takes only a subset of the pixels, not the full image.</code></pre>
+<div class="cell" data-execution_count="6">
+<div class="sourceCode cell-code" id="cb10"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb10-1"><a href="#cb10-1" aria-hidden="true" tabindex="-1"></a>model <span class="op">=</span> dinv.models.ArtifactRemoval(</span>
+<span id="cb10-2"><a href="#cb10-2" aria-hidden="true" tabindex="-1"></a>    dinv.models.UNet(in_channels<span class="op">=</span><span class="dv">1</span>, out_channels<span class="op">=</span><span class="dv">1</span>, scales<span class="op">=</span><span class="dv">2</span>).to(device), pinv<span class="op">=</span><span class="va">True</span></span>
+<span id="cb10-3"><a href="#cb10-3" aria-hidden="true" tabindex="-1"></a>)</span>
+<span id="cb10-4"><a href="#cb10-4" aria-hidden="true" tabindex="-1"></a>model <span class="op">=</span> loss.adapt_model(model)</span>
+<span id="cb10-5"><a href="#cb10-5" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb10-6"><a href="#cb10-6" aria-hidden="true" tabindex="-1"></a>optimizer <span class="op">=</span> torch.optim.Adam(model.parameters(), lr<span class="op">=</span><span class="fl">1e-3</span>, weight_decay<span class="op">=</span><span class="fl">1e-8</span>)</span>
+<span id="cb10-7"><a href="#cb10-7" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb10-8"><a href="#cb10-8" aria-hidden="true" tabindex="-1"></a><span class="co"># Load pretrained model</span></span>
+<span id="cb10-9"><a href="#cb10-9" aria-hidden="true" tabindex="-1"></a>file_name <span class="op">=</span> <span class="st">"demo_measplit_mnist_tomography.pth"</span></span>
+<span id="cb10-10"><a href="#cb10-10" aria-hidden="true" tabindex="-1"></a>url <span class="op">=</span> get_weights_url(model_name<span class="op">=</span><span class="st">"measplit"</span>, file_name<span class="op">=</span>file_name)</span>
+<span id="cb10-11"><a href="#cb10-11" aria-hidden="true" tabindex="-1"></a>ckpt <span class="op">=</span> torch.hub.load_state_dict_from_url(</span>
+<span id="cb10-12"><a href="#cb10-12" aria-hidden="true" tabindex="-1"></a>    url, map_location<span class="op">=</span><span class="kw">lambda</span> storage, loc: storage, file_name<span class="op">=</span>file_name</span>
+<span id="cb10-13"><a href="#cb10-13" aria-hidden="true" tabindex="-1"></a>)</span>
+<span id="cb10-14"><a href="#cb10-14" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb10-15"><a href="#cb10-15" aria-hidden="true" tabindex="-1"></a>model.load_state_dict(ckpt[<span class="st">"state_dict"</span>])</span>
+<span id="cb10-16"><a href="#cb10-16" aria-hidden="true" tabindex="-1"></a>optimizer.load_state_dict(ckpt[<span class="st">"optimizer"</span>])</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+</div>
+</section>
+<section id="train-and-test-network" class="level2">
+<h2 class="anchored" data-anchor-id="train-and-test-network">Train and test network</h2>
+<div class="cell" data-execution_count="7">
+<div class="sourceCode cell-code" id="cb11"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb11-1"><a href="#cb11-1" aria-hidden="true" tabindex="-1"></a>trainer <span class="op">=</span> dinv.Trainer(</span>
+<span id="cb11-2"><a href="#cb11-2" aria-hidden="true" tabindex="-1"></a>    model<span class="op">=</span>model,</span>
+<span id="cb11-3"><a href="#cb11-3" aria-hidden="true" tabindex="-1"></a>    physics<span class="op">=</span>physics,</span>
+<span id="cb11-4"><a href="#cb11-4" aria-hidden="true" tabindex="-1"></a>    epochs<span class="op">=</span><span class="dv">3</span>,</span>
+<span id="cb11-5"><a href="#cb11-5" aria-hidden="true" tabindex="-1"></a>    losses<span class="op">=</span>loss,</span>
+<span id="cb11-6"><a href="#cb11-6" aria-hidden="true" tabindex="-1"></a>    optimizer<span class="op">=</span>optimizer,</span>
+<span id="cb11-7"><a href="#cb11-7" aria-hidden="true" tabindex="-1"></a>    device<span class="op">=</span>device,</span>
+<span id="cb11-8"><a href="#cb11-8" aria-hidden="true" tabindex="-1"></a>    train_dataloader<span class="op">=</span>train_dataloader,</span>
+<span id="cb11-9"><a href="#cb11-9" aria-hidden="true" tabindex="-1"></a>    plot_images<span class="op">=</span><span class="va">False</span>,</span>
+<span id="cb11-10"><a href="#cb11-10" aria-hidden="true" tabindex="-1"></a>    save_path<span class="op">=</span><span class="va">None</span>,</span>
+<span id="cb11-11"><a href="#cb11-11" aria-hidden="true" tabindex="-1"></a>    verbose<span class="op">=</span><span class="va">True</span>,</span>
+<span id="cb11-12"><a href="#cb11-12" aria-hidden="true" tabindex="-1"></a>    show_progress_bar<span class="op">=</span><span class="va">False</span>,</span>
+<span id="cb11-13"><a href="#cb11-13" aria-hidden="true" tabindex="-1"></a>    wandb_vis<span class="op">=</span><span class="va">False</span>,</span>
+<span id="cb11-14"><a href="#cb11-14" aria-hidden="true" tabindex="-1"></a>)</span>
+<span id="cb11-15"><a href="#cb11-15" aria-hidden="true" tabindex="-1"></a></span>
+<span id="cb11-16"><a href="#cb11-16" aria-hidden="true" tabindex="-1"></a>model <span class="op">=</span> trainer.train()</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="cell-output cell-output-stdout">
+<pre><code>The model has 444737 trainable parameters
+Train epoch 0: TotalLoss=0.032, PSNR=29.155
+Train epoch 1: TotalLoss=0.035, PSNR=28.895
+Train epoch 2: TotalLoss=0.035, PSNR=28.837</code></pre>
+</div>
+</div>
+<p>Test and visualise the model outputs using a small test set. We set the output to average over 5 iterations of random mask realisations. The trained model improves on the no-learning reconstruction by ~7dB.</p>
+<div class="cell" data-execution_count="8">
+<div class="sourceCode cell-code" id="cb13"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb13-1"><a href="#cb13-1" aria-hidden="true" tabindex="-1"></a>trainer.plot_images <span class="op">=</span> <span class="va">True</span></span>
+<span id="cb13-2"><a href="#cb13-2" aria-hidden="true" tabindex="-1"></a>trainer.test(test_dataloader, pinv<span class="op">=</span><span class="va">True</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="cell-output cell-output-display">
+<p><img src="demo_splitting_loss_tomography_files/figure-html/cell-9-output-1.png" class="img-fluid"></p>
+</div>
+<div class="cell-output cell-output-stdout">
+<pre><code>Test PSNR: No learning rec.: 24.549+-1.052 | Model: 31.911+-2.220. </code></pre>
+</div>
+<div class="cell-output cell-output-display" data-execution_count="8">
+<pre><code>(31.911066627502443, 2.219884675922773, 24.548791694641114, 1.0523162766060832)</code></pre>
+</div>
+</div>
+<p>Demonstrate the effect of not averaging over multiple realisations of the splitting mask at evaluation time, by setting <code>eval_n_samples=1</code>. We have a worse performance:</p>
+<div class="cell" data-execution_count="9">
+<div class="sourceCode cell-code" id="cb16"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb16-1"><a href="#cb16-1" aria-hidden="true" tabindex="-1"></a>model.eval_n_samples <span class="op">=</span> <span class="dv">1</span></span>
+<span id="cb16-2"><a href="#cb16-2" aria-hidden="true" tabindex="-1"></a>trainer.test(test_dataloader, pinv<span class="op">=</span><span class="va">True</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="cell-output cell-output-display">
+<p><img src="demo_splitting_loss_tomography_files/figure-html/cell-10-output-1.png" class="img-fluid"></p>
+</div>
+<div class="cell-output cell-output-stdout">
+<pre><code>Test PSNR: No learning rec.: 24.549+-1.052 | Model: 30.434+-2.487. </code></pre>
+</div>
+<div class="cell-output cell-output-display" data-execution_count="9">
+<pre><code>(30.434445762634276, 2.486670644991658, 24.548791694641114, 1.0523162766060832)</code></pre>
+</div>
+</div>
+<p>Furthermore, we can disable measurement splitting at evaluation altogether by setting <code>eval_split_input</code> to False (this is done in <a href="https://pubmed.ncbi.nlm.nih.gov/32614100/">SSDU</a>). This generally is worse than MC averaging:</p>
+<div class="cell" data-execution_count="10">
+<div class="sourceCode cell-code" id="cb19"><pre class="sourceCode python code-with-copy"><code class="sourceCode python"><span id="cb19-1"><a href="#cb19-1" aria-hidden="true" tabindex="-1"></a>model.eval_split_input <span class="op">=</span> <span class="va">False</span></span>
+<span id="cb19-2"><a href="#cb19-2" aria-hidden="true" tabindex="-1"></a>trainer.test(test_dataloader, pinv<span class="op">=</span><span class="va">True</span>)</span></code><button title="Copy to Clipboard" class="code-copy-button"><i class="bi"></i></button></pre></div>
+<div class="cell-output cell-output-display">
+<p><img src="demo_splitting_loss_tomography_files/figure-html/cell-11-output-1.png" class="img-fluid"></p>
+</div>
+<div class="cell-output cell-output-stdout">
+<pre><code>Test PSNR: No learning rec.: 24.549+-1.052 | Model: 31.003+-2.107. </code></pre>
+</div>
+<div class="cell-output cell-output-display" data-execution_count="10">
+<pre><code>(31.002875900268556, 2.106733038650352, 24.548791694641114, 1.0523162766060832)</code></pre>
+</div>
+</div>
+
+
+</section>
+
+</main> <!-- /main -->
+<script id="quarto-html-after-body" type="application/javascript">
+window.document.addEventListener("DOMContentLoaded", function (event) {
+  const toggleBodyColorMode = (bsSheetEl) => {
+    const mode = bsSheetEl.getAttribute("data-mode");
+    const bodyEl = window.document.querySelector("body");
+    if (mode === "dark") {
+      bodyEl.classList.add("quarto-dark");
+      bodyEl.classList.remove("quarto-light");
+    } else {
+      bodyEl.classList.add("quarto-light");
+      bodyEl.classList.remove("quarto-dark");
+    }
+  }
+  const toggleBodyColorPrimary = () => {
+    const bsSheetEl = window.document.querySelector("link#quarto-bootstrap");
+    if (bsSheetEl) {
+      toggleBodyColorMode(bsSheetEl);
+    }
+  }
+  toggleBodyColorPrimary();  
+  const icon = "";
+  const anchorJS = new window.AnchorJS();
+  anchorJS.options = {
+    placement: 'right',
+    icon: icon
+  };
+  anchorJS.add('.anchored');
+  const isCodeAnnotation = (el) => {
+    for (const clz of el.classList) {
+      if (clz.startsWith('code-annotation-')) {                     
+        return true;
+      }
+    }
+    return false;
+  }
+  const clipboard = new window.ClipboardJS('.code-copy-button', {
+    text: function(trigger) {
+      const codeEl = trigger.previousElementSibling.cloneNode(true);
+      for (const childEl of codeEl.children) {
+        if (isCodeAnnotation(childEl)) {
+          childEl.remove();
+        }
+      }
+      return codeEl.innerText;
+    }
+  });
+  clipboard.on('success', function(e) {
+    // button target
+    const button = e.trigger;
+    // don't keep focus
+    button.blur();
+    // flash "checked"
+    button.classList.add('code-copy-button-checked');
+    var currentTitle = button.getAttribute("title");
+    button.setAttribute("title", "Copied!");
+    let tooltip;
+    if (window.bootstrap) {
+      button.setAttribute("data-bs-toggle", "tooltip");
+      button.setAttribute("data-bs-placement", "left");
+      button.setAttribute("data-bs-title", "Copied!");
+      tooltip = new bootstrap.Tooltip(button, 
+        { trigger: "manual", 
+          customClass: "code-copy-button-tooltip",
+          offset: [0, -8]});
+      tooltip.show();    
+    }
+    setTimeout(function() {
+      if (tooltip) {
+        tooltip.hide();
+        button.removeAttribute("data-bs-title");
+        button.removeAttribute("data-bs-toggle");
+        button.removeAttribute("data-bs-placement");
+      }
+      button.setAttribute("title", currentTitle);
+      button.classList.remove('code-copy-button-checked');
+    }, 1000);
+    // clear code selection
+    e.clearSelection();
+  });
+  function tippyHover(el, contentFn) {
+    const config = {
+      allowHTML: true,
+      content: contentFn,
+      maxWidth: 500,
+      delay: 100,
+      arrow: false,
+      appendTo: function(el) {
+          return el.parentElement;
+      },
+      interactive: true,
+      interactiveBorder: 10,
+      theme: 'quarto',
+      placement: 'bottom-start'
+    };
+    window.tippy(el, config); 
+  }
+  const noterefs = window.document.querySelectorAll('a[role="doc-noteref"]');
+  for (var i=0; i<noterefs.length; i++) {
+    const ref = noterefs[i];
+    tippyHover(ref, function() {
+      // use id or data attribute instead here
+      let href = ref.getAttribute('data-footnote-href') || ref.getAttribute('href');
+      try { href = new URL(href).hash; } catch {}
+      const id = href.replace(/^#\/?/, "");
+      const note = window.document.getElementById(id);
+      return note.innerHTML;
+    });
+  }
+      let selectedAnnoteEl;
+      const selectorForAnnotation = ( cell, annotation) => {
+        let cellAttr = 'data-code-cell="' + cell + '"';
+        let lineAttr = 'data-code-annotation="' +  annotation + '"';
+        const selector = 'span[' + cellAttr + '][' + lineAttr + ']';
+        return selector;
+      }
+      const selectCodeLines = (annoteEl) => {
+        const doc = window.document;
+        const targetCell = annoteEl.getAttribute("data-target-cell");
+        const targetAnnotation = annoteEl.getAttribute("data-target-annotation");
+        const annoteSpan = window.document.querySelector(selectorForAnnotation(targetCell, targetAnnotation));
+        const lines = annoteSpan.getAttribute("data-code-lines").split(",");
+        const lineIds = lines.map((line) => {
+          return targetCell + "-" + line;
+        })
+        let top = null;
+        let height = null;
+        let parent = null;
+        if (lineIds.length > 0) {
+            //compute the position of the single el (top and bottom and make a div)
+            const el = window.document.getElementById(lineIds[0]);
+            top = el.offsetTop;
+            height = el.offsetHeight;
+            parent = el.parentElement.parentElement;
+          if (lineIds.length > 1) {
+            const lastEl = window.document.getElementById(lineIds[lineIds.length - 1]);
+            const bottom = lastEl.offsetTop + lastEl.offsetHeight;
+            height = bottom - top;
+          }
+          if (top !== null && height !== null && parent !== null) {
+            // cook up a div (if necessary) and position it 
+            let div = window.document.getElementById("code-annotation-line-highlight");
+            if (div === null) {
+              div = window.document.createElement("div");
+              div.setAttribute("id", "code-annotation-line-highlight");
+              div.style.position = 'absolute';
+              parent.appendChild(div);
+            }
+            div.style.top = top - 2 + "px";
+            div.style.height = height + 4 + "px";
+            let gutterDiv = window.document.getElementById("code-annotation-line-highlight-gutter");
+            if (gutterDiv === null) {
+              gutterDiv = window.document.createElement("div");
+              gutterDiv.setAttribute("id", "code-annotation-line-highlight-gutter");
+              gutterDiv.style.position = 'absolute';
+              const codeCell = window.document.getElementById(targetCell);
+              const gutter = codeCell.querySelector('.code-annotation-gutter');
+              gutter.appendChild(gutterDiv);
+            }
+            gutterDiv.style.top = top - 2 + "px";
+            gutterDiv.style.height = height + 4 + "px";
+          }
+          selectedAnnoteEl = annoteEl;
+        }
+      };
+      const unselectCodeLines = () => {
+        const elementsIds = ["code-annotation-line-highlight", "code-annotation-line-highlight-gutter"];
+        elementsIds.forEach((elId) => {
+          const div = window.document.getElementById(elId);
+          if (div) {
+            div.remove();
+          }
+        });
+        selectedAnnoteEl = undefined;
+      };
+      // Attach click handler to the DT
+      const annoteDls = window.document.querySelectorAll('dt[data-target-cell]');
+      for (const annoteDlNode of annoteDls) {
+        annoteDlNode.addEventListener('click', (event) => {
+          const clickedEl = event.target;
+          if (clickedEl !== selectedAnnoteEl) {
+            unselectCodeLines();
+            const activeEl = window.document.querySelector('dt[data-target-cell].code-annotation-active');
+            if (activeEl) {
+              activeEl.classList.remove('code-annotation-active');
+            }
+            selectCodeLines(clickedEl);
+            clickedEl.classList.add('code-annotation-active');
+          } else {
+            // Unselect the line
+            unselectCodeLines();
+            clickedEl.classList.remove('code-annotation-active');
+          }
+        });
+      }
+  const findCites = (el) => {
+    const parentEl = el.parentElement;
+    if (parentEl) {
+      const cites = parentEl.dataset.cites;
+      if (cites) {
+        return {
+          el,
+          cites: cites.split(' ')
+        };
+      } else {
+        return findCites(el.parentElement)
+      }
+    } else {
+      return undefined;
+    }
+  };
+  var bibliorefs = window.document.querySelectorAll('a[role="doc-biblioref"]');
+  for (var i=0; i<bibliorefs.length; i++) {
+    const ref = bibliorefs[i];
+    const citeInfo = findCites(ref);
+    if (citeInfo) {
+      tippyHover(citeInfo.el, function() {
+        var popup = window.document.createElement('div');
+        citeInfo.cites.forEach(function(cite) {
+          var citeDiv = window.document.createElement('div');
+          citeDiv.classList.add('hanging-indent');
+          citeDiv.classList.add('csl-entry');
+          var biblioDiv = window.document.getElementById('ref-' + cite);
+          if (biblioDiv) {
+            citeDiv.innerHTML = biblioDiv.innerHTML;
+          }
+          popup.appendChild(citeDiv);
+        });
+        return popup.innerHTML;
+      });
+    }
+  }
+});
+</script>
+</div> <!-- /content -->
+
+
+
+</body></html>
\ No newline at end of file
diff --git a/docs/demo_splitting_loss_tomography_files/figure-html/cell-10-output-1.png b/docs/demo_splitting_loss_tomography_files/figure-html/cell-10-output-1.png
new file mode 100644
index 0000000000000000000000000000000000000000..262169ba9fff2c0538abf363dcb97ead5b411d57
GIT binary patch
literal 10068
zcmZ{K3p|tk`~P&%0ZD~QC?({uBIKM(PKl^7hbhcySPpZTMo*q1LNbILa;(ROZJ0KP
zba0q6*_<{>j&rtAj{n<Jzwh^X{=ff!uh-nO``&in*XMd)*Zcjx?xz+v4F&ce-wy(T
z1dNTYS%N@25Wwr2y}ZC*&yz;CfQ?3gflYuF(lsFHo*x`!b}!(*7c#)h!|7xo+|S<w
z>7$}_Nl8WFq<cWXeSb|b*!w>RC?Wmazzt(IR=}6+yKe;b2Z03dZ9jS5=oNZ^KxgZW
zuj$_oel^1k^@88wzF$y1xA^kdSRn5~ULKEx%Td}s`nT%dls!d0t2lVb=-Iv*$v?wy
zyzDx7@x@EyqOA$Zw6-&wk(y-*N1x_+-`XX~WBdJeV{Qu1lkxW)W$?i-o0})AYyBB8
z_B3gn)z#g$l6XVeZHUy~H$Lc(Y}X0JS2GlV$6ny<OeS@jnZOqGqY}Ck1d_erL*IV)
zS-L6+^zhJ#s{ja;{ptU;%T84t;o#t)eBlCNuod`-d(j<#QpFd~@_s*c9uVlnlS-((
zySuxQtD9R2^_Y9&?5q!z<Qn^h&Ar|?J&g+p3X(x0TGLd$y;AY8l07rKK_?_2Gp)9U
zhLO+W;!J{q$N?yns?IU?)YOaDuTM<^o4B~Rg7ng9Nmgg)tz#!nVC(AiQ(wM(zPY)1
zb=5h}Z)$Qf;q~j+qdwP9C7Bqwy1M3BzEudaD*Yr<OG1Z+=99_R6e@M!!ezN0HfnNi
z4!N#4Gdr97_N|;y$oQO_wYBw#qoJW;x3ZvK7p_0O99^xHk&&USqH?;&vwX6-7*b@6
zM3%I)wl>YrmvUBC#(fZ8UN_Kaw9A7B_Q202QznB`_yq)-yvfW8r%cG5qMlKo9W3e7
zr%x+uXvifdCMxUbv?gM(H}R0fZgair1#&=G7_0(0#0thVCo3wetIPQN`z!0}Dsj=(
z1Z{Qoh81Yeolw%*#kR!5LJ2WS6r-`Ru@**!SD#e5t@r2ky6bCRgO#^<d3gs55r+>S
z?qV_(nM|hgrAs#M>A?T7j&XgGdt+*KsT9ibT5W?o!XR)&xvk>L)BDSBku3B;g*T!s
zbiw01V}*9-?d`n6!oo=kPx3%V$8&yu{>Jz3FX!TH@oJkDIn@wutF1|j5YsfWaCrFM
z(quodh+a&}or1zbF=jkLsk^TadhGb|7pbYoQq$8D2?RntA|fIp*70SZ%)4@1v+LKd
zyP(mwnp#@^S$o|=*gZfZhym9QTx-T`x3G_=r-=`BMs1SKJ}k7gr|!#_+qv4I5wjZ;
z^FBWi<+i!cX3qtmJ$rWlR+IsI`&=?VgR?t3JHPr)e@hws{@o=PXZd!1myhsYx6dh`
zKY!AW?BjiIh~OM)A2OO0oBj$o=e`PWyV;7s>E2UPbJ_;%nVBS8JU#e0@dturi#Mr;
zAt`b7_4R2!1&f*6C+Iy`Nm$s@30?kPacneTrkwG2P6h-tHa13SZ*aQ1`uZj}Prm8#
z>>3&xl9iN{OfkuSv6lOhI-CV0WoD)XrRb||6=80A+I}qggf5FbC}5-H?CiYwey3nm
zRO62y!MP9a-o30uomQTlp2oJdwbe^@vRJTp%l(La63Ge(Y=|WxnqN>*-3e?<@aR4Z
z#3nHzA>nH|kpJ(tHrHCwt9&IB<#4<;uymGhFCMx_yl}=wV>_rA3`SX1ReE~?Ry5Zl
zE?+yYw--R-XtaE>JB}P|2G!HNskgzYSM_AhTNm8{(2!z_KMNclfH+kei~$n9eoM60
zA4u+&xyxm)J%hMOnS*?Me6d@|KBBaZ-hjcKJo}YaXYQNZ=lnhkTVKS~Z3h|%1$gJq
zow0>N*7m*zvw^Ftt9bzIo<v59nLDZb49Z$rS;gL6Kv&n~rD9^tO^JG!OZ~VUj***N
zA%J>XvZCFf<gUp6J08sW6%MDdrRB!yonZ_0LXF$t-p;A=O8_3X({?K1%G!wvX7*)z
zW22E3D5%^v35d=0>A0xy)3tAa_yGy`o!G2|h!zAs>+y+4Q8&Zk98eg5Bys`jPV8;8
zQyip%K)BFHK^1)gDgqGQr%#_=S6_8<zwYShXf-|R^fSpCCI_rp?4p#Orbt$@S0B*;
zxEQ9)hf7tlv9V1R-UB(qr`!`Cf9$x2?!Ckh?1$fTamgcSYiPXd=<9p;O!3NJp5>JD
zy1J~}YUj_Fo(G=W58#a&3p_;hub!zZ=H~4m*4H28R*^{lueYCtg>LtJe3WDz2kSpv
zSz7rDTwwPye&%*j&*P@1CTT4noc7Yv5+AU}vFy*3Kv5A95up71pbq4y^6{l^UIEGX
z0|55wFMHd`%@Vgd2r5<fCGp~vf#n5~q9@9*zXqdq3<D~C1iTFbSvZ|J3KVecQT6Qx
zC=CgIRen0|lGF-{nVy!>U{SlwTB#v>{0!e(bDZ@Hw4*N*n+e{v<2rL5liVtlvkr~o
zy|pnH?$2n?UBQ!kXcau5%OUeFe|HW@D?hIdtna6+s5t7&t;cLT^Bvsoj|O`LCHQG^
z+KY|ZEurT;b$Ga$(KU0JVZ-k)o#Xb=#>U56SJ`wt9}VZs?f)z5Txiv@9T)Gs+8(~?
zOefRut5VCMrJ*ac+0(%9)<VNKR=N6p#SYXKl)<z=n-KTWGcrg5ztUpM&{C(b6=juy
z(QG{Vkun>@L2Kaxp$&5mb`UDgyTG0Vffg~UqBg|uB|*o(*sZ+(W9RIqWk2tcJ7?Mx
zW&VywV52$RDd<t3qr$57+f9RfM@H>U?kb_OET`n@pAAAWDW*_$5$l4L3{h~3=waD!
zTPAxzAbBzKfA-PSbs+Bq<=auh(_Tp-4!y90<S~?hoi}D}hTMw6zjr}w!IU4DI{+Iy
zykC(SvgT(2bZCD+xc{t>mdATUQIg9&X`W!N^C`YG;S;mcCN^`a<$B<%+=**oxu+5=
z4RTvN3dgF!jS#0oPf0*Fu)Ch-jMGd{nxx)t@U%7|9vVJvBe#DD_qpDr+wnz6c$M=&
z7s~nAr9MI2rp1&u+j~+lR3T1buF_KFv6HWRABL@`wkoIwH35NeHv=Me{5L!5TaSZ4
zm!FFJwb+`RFO6I3^1Vw~Xez7vsqK8hgW%0RO0GIuVS{t<YO@lLOcoU!-~E0k=wVN+
zmqnRDr<`XMN|Duy=zNQ;!rhWT_K}7@;<0hIFRjacYjZ<Cb>;L2%{`E{YJbm+o$~%)
zw5%axZ&jJeQ5c+oikQK*<yj^oIn<KK8*?)VtGqk0+{Yx{l6cBW`9#dk6>N?>@3Q4W
zT=_)kOG`=F(-eJxg_z*x+w*O#;jAyF1j;HGdx6TSalRTfaq7Ay)t)pEW2T-Tf0~|L
zr8L;fv3w@F3-o^KQ{q3FkJ0(db~bgRTbnDjz(Qri<lS}PnkNvBkaTSz+%pmwPN#mV
zYR?O@gS^cE-`9#OJED`H_I87z#)n~D56hWMYEU}cr?<3Dfl=iz&S%^aeY5Dhu`zGy
z2Y%er#BQ9V?<b+h48^p|-!d9<?wygdLO{n=^Ec%79gwrK6UNk9T}3=3YJ5*Gx2b|G
z@03g@$}RuV?HI5sXaTE&&?u0B#V3nVW@br;7;fHR>Rj*hGdI+C16OfPU3Pn!e&c%(
z2gCwvGjNFC_E`Ll2G2>jxM8wrR{qE*P9uk+sQFM^%`%Af(uYv-X-C=;EZ!{rPCcj}
zdusuIIaprS?wZR`4Msl|WfnaCsefyg%Wg#}{*7SVa!l_ZYxK?P3fp}LAcOfSH|{y)
z8o1q=p9&qPc_Q*H_t9XXGpeMTq{;*tI4@7ppk_R*<3T68+w?Evr*ev#S0f82UdFIg
z5e||DZe2#h;)&eq;qvl{Ksz%FP_mfyx&L($y{Fm*j>nvCu6M|LCUf?Xq}a@x+&E8L
z7}+O-S{zp!^|?X((*10YkU3VhS#w`-RIi3Rw~|B<BY5)wY54RPr0WB$++);-9;nDJ
z6`Us$t8q95a6~ihxPm4=F^UP<*ri(HAeUNbz%JC5>oHF@SohLEQ0ZkBFcIf!m>}(@
z?oStgCcq4K!69R~6z<P@G33ip3AT2@jX$t2?m;;^D?~mk)=~}|y=J0OoWr#up0%s4
zv1?T>Ck9v*6z`-({W=fGk0x8MS71f3s~MxGge|j2TwEMD>!YgG)xGPl(pV`HSEqiA
z#3OCPevYPyZ86NnIo+|nHCFXmF|7i$VEt`$u0ICK&3|b7uAOC%GwG(E1Svq7%-+Ig
zZ?vYX?}5{A8%o_HQ=+n3=DSS6yWx<BRifFB0aGWz9q%SW%u>f1O?J&wI`J+~GPHuy
z*gz>&lO?gN!3!TvjTXv|??oE#`;*MtGuO*`epzZWz|(-7-UUH8u=@`Y67S%Y=j`p)
zOpHOqyfZ#|XA+;^eVFv^c>^@2)htzA{SC}I+?cIbuo|Dc`P^;b(AxYYg>Zq%5uNfV
z%%K0+A~Z-0aQc_p^|Gily(tKoeVXnNCF8*)2vn-sDFDbb3#Z@dJM3e@L>)`ll%v+n
zvbvd(lmV-TrU|*+fR!BelW&ZAb$-z8kgAI8ib)?!ra&5IW2fYg%iRbHbve~7)5DhT
zD^EeFAkQfjr0f8Foxk>or~ku#+ZCT2ZY$Yv$OgRH5vh9wi&-C$6r|~{cE+c1@!26o
z8&Xn&dh7QsmOt=;cfnbECn4Xi(Du$iI%<Sp&ctq(Gzc6axJl2u-Y*Sq;z!nQ1X{R?
z7p>HwM>BKkMx~a^-K*7E@lr)qwY7S()Uh-~%PK_q53X)h)U){bJ#au1097wjSYV9I
z0t>Azd5G7hfLsW1dT_cx37#JMa<QiPBGBjc`^p(GL@@$m4&A3GB}HbT5f-k});ioT
zR)n)oPdomvICXsScttE>A63=aOp4=13z#M_9C8>v%>)95GqTuw{xo#8lH&^~86aAQ
zT_Fqp6o(Kit8IWp{07LcDvyWC_J$WwIRWbKFtR<Bs$cy~1i}2`(_s^YtPs2HCOm~&
znHgY|onZ6!5XUc;i+yR%v_0pRfr40!!9$@?7YJ{fb!}0RyexHdl^sEh0oU~6kZ={W
z)tpWDE-1;MYLO=Ix{(8og4qB;vF&ybStApTCBxSjvZsM&Pl9gEQs<PuoE2KbloyOG
z*7)I8&2@j~y(^Atq@dV?jOgiZS{kR{6|S<m5FgG@!`*-jt-T8AHiNQlQ?tM|KXRas
z?RipN+DkMkWV9;6kv{%$)hntuYt(@nxErqbqfW48l^!~a(fw7>s{FmHhgiN(^Wc8q
z$F_J8YzV`?h%*EhN(cw?!jxE&f7ii3H#c{TBB(w8-cnu0I$@yn!v#dTnzyGdt2R8m
zn;3Vxc{9N%L)Ro6I;mHTe)(uxlF}P#y!AR3FG-ba?Q+~J^XmSUZerL~&K~n{W0#fu
zf!FaLS91X|4$=M9wEiu=_c!1sFT=&N@^`)^)zxYqv}wRhQZxfi_ecQEc6vFW0~dEV
z#te1D^nBy2Y1U5hT}6lGIUyYl3nk$jMIIVrd)_d5d>^8JGF^A)qV*FYcS%WJ_ssaD
zr51q*tH@F}mIwIoGE$4H?2<ijF*UFAjG5_aAe1iAK<P;5;$3#hH+|NE$m<XHSUU>P
zZfY-FBW)VXWE{axYIqx53!Gzl(`#2)p%KbRPkFWt{w@;v>KgKR(GixaoAiB~eFCr@
zFUn_I)rU@CQ9(iJT*(LXb=lKZOKmEQnVA^@TARlBtL$m4)y)fRI~-e9nx}EKrQn9|
zG9ezLAWJH^dPoQ$=m?8sW#Lq=1bUw~UDeK&{>5@UGqm{KtaH4M2sQ(=+&4z)@l=&v
z;}q(NV?&JX+I3d!Et35(wm5e8S|+h(T%lce4F|0m+{bJ1o0|U1)WF?uNt0N+(Vh@8
z=-Z&G=;nN%^U_OG*I4kw(HODr3;X(x@`(!;j(|Bes8^{lJ>&xwW??aV>XM*N@N0=d
z?yn-fI$0>THB{;A+>JmFQhV>#f<SlS-<tj=ObeHIy$NpZnb`{PXi<dgUipu=KZQmA
zD0195C)?du8G~z4o@>fW%J+bWp{*pLRty`f%OT@58jjSDga=1Q*#|LE0Mi(Ftd2?u
zJE!!m7EHfr3~_y(!kx^B2$V&)hCm}BYQr2(fJGp~At;rL$9-&SGWKUmZq7#ny<6Ot
z@gyqaod-sw3IS=al9E~k8i6otYk6iMLYZC7CF)x#>)}^A&Mg%f*?H(3qN}<kh0Fr0
z^nq=An*UY!=Fc&Tii(PUpMj{u?O;a{#GemkGJ`+ZGiVQbs&7dRW8>xb{i3`8<=Vf&
zcWtPAD!6DGd~2hq0#>4~$aaifO}V3RK#__OHxv;IGTZtV9V?k4oUVR5{z5)^<Ei<B
z9FLIu_e(uAv;teFr@gunCDqHe-9*)85fI4cXE5<^rShv4^wOe>uX#XuIw094QR36y
zyQ@R>>mV4NT`@+N&fM_C2!&G4Fn22+Ishj%?q9A5)?a~u%2Z=x)XQUpkCqgw9u^9j
zcpV8KAA5zh=43YNx|<$qg%H-UHf=&gpd|21UmXk~IgizQ5>OahV{v-<WO*|hiBo&p
zn`KvoPjL);2OUaVHH`F)idz3BT6>;tx8}+1pPw3k{O-gopk5+~dh5f2kJ*K1bdPYk
zCE-j3@{3avA>R|Sy&B@gy^-D2{Q1_jw)IYbp(*g!>6EI?9o^2vkR?nACk49PY!KeT
z_4DFa-!Ld-yTx0%6-{_g_G)F^${%Ub<-RpnRm}3OKlbe%fPg^;2q1g<{~~*jsruY6
zMzF!A9Iv39xY&*bWAs#>H7q3Go%Cpi9URkJDyiwJ->`R7Annb_Wv87AQ<aHBo6BUg
zW=i~4V%ECL$QBx8im`{C5CQ_<x6Od_&g?xRjvCxwCPLWx9IuSL_CWs4CHQwVfcDD|
z;!##vekAYMmYLlwTavWRLY!96>_xU+Lzc42V-yZnN*0};B`l9SsL*|nFbkqE80f%_
zeKsqH?&A=zC;0iLme-s!2C%y#RNA#wWY++ujii%bz{6I%vv}dGct7^isPRu$DX#1o
z{{b2LjpKSisRy`t`!_}M;2|tjWd?WTUkU~C<JkNY0kO1Xl=YL$y_y5!Uv-Kd^CEY5
zI_yKc$5%BBu9o*&h*P`)2^VMD9C%@Zf_Bg;i<DR~6=c4qN=pFEr+!OsDYgA#bwN9H
z`UD_flKzJ`{%VT}`y(ohn!b~WJor^u5ILz14);f=!svn3{00cB92c)3jYx+*8Eulu
zI)QaSPv-0QLESRo^zg+dnFwN}WY{IyZinTVD!Bv&I`+Cjq0%9_Ou{?aQH*SwM%tm9
z9a9uJMv#eJOqfWuKfGE`ofYhCF=j?^HH(GAPuekJj266nh1RNU7`IH4!L(rtyO@s_
zr?(mG?||^cc^&QawD{p_R%lf_|AEh}d6iFU?T=2$y&KZ6_3Q~3K6Atp%~0kUSC!}@
z)vweG(ld+Z0MTpKtSMq%IKjqLuP-*oZViRbRhyM$1hGQQnpp<?Kx<O_|Fxz|7j*KY
zfI!Baj!|UG`320E#|alP*gjfLoSLP$t0g6jo{=|`m+y&x|7)k$Qo)J-a)fm^QCG(8
zmG2X~73SUDt)V<406bCCFbu*(QeIX|da^D}%J5U%A7!+vmD$GA@jx@W>oI4ujj26w
z$F+XOBO;LdZMg6T@t|FBz?T?`P?ndq|CCuwTYCo8A7q6n+Umul+ALp)S{4+YOKW-V
z$Iqe|Bq~>BujTNpvQ$Gie~#~gXKk&IZ#OMG^1+ChexH@O&Yd@AnUHn_Z7n*;F;HUK
z0Ad+44peAQi{**j_=>K)uHj$B{NJ)__@%0NXwHc0;y1wnud50-ZwOY{yuF>4%)R7g
zD2}=fz$wnsjI1KU$a8F`GfE}{$76M54%M4ZRKVL~;=mzescvb>)jVf;K<{bw{Jidb
z@&}nFng<Dn&E!M7EU(*`?>kcQsFj=D9)b%D`D2D2&%?G+_wIM0mq3bumf*L3*j)P_
zoqUKNmmLna84MLvguJID-5p+Vy)~crY|8QC#GolhHZA-4i?3Mao%*!t0|tkWRC%)k
zy3K&b(7HKk9<I!`8>OSZJBXPs)S!n1-;!MKdOR5BbKFbAK^*`ivUN3lgY7uLDj*J<
z?EbIZn0`La+Y_IJ5;a;TybIe^LbZ4D_nPu49EsB^_6QlNkB!|bajdPG1bFC|&5gD6
zWUlYTNK5_4rx%0tCmYHrVlwo{N5qPhLSX?n(sUy=p8%c_?tu2c1`&X9@OYLYvOv<n
zw<sZ*IY47m0!K^(Is14epIT=y-C_(b!nxOHOT@W$ein5+fvpBerGDQJV4yLi2@&6a
z@>^E8JYPq8L8GQM9W_i)j$S&_iatUZ%EIDUIo&$>H+V+jvbHmTb@K55)VrLfsi_IH
zh5#!F?*h4Y!<)1zhO838=HF<&b8(5DxLEU*ISOZn&-!Np{?z%Lxd-D~t{YWvk6Xul
zP)DpfF(h>{M52ss_%K^9kWw7i-X$5fkstP9pUlP0@NZ9L)T)>31j9-_LdM%N%$iGT
zH__dGYMZV7*2|R~`|Q<9Cj#K=$|6rx^gaj4sh<QW#AvqIk^fF%2p^c)n#*VCqbcI_
zD-DndrQ6jw>Q0pUU$bE?HLRYV9!|fZ?qWg#TwE+~B#2`nNg3euxn+5;I|dC+jeHqD
zqgIVgRu_ig_=LBEedcJkBJQnomG2l1@Ha5v3FdyaiXFq%a#-ACS``NRM-jcIN(?Da
zzrQrrs^4c_vnWieDWA=a*FBD9hHq`SZeZ4ycf-*NS4uKq?}CbPlA;X~p<b`K*SZQS
z4jHE>zj5DyFw9a#<fP@K;rsJ*ysM}FDr)i5iOY{_#H=tyu_4Y{wv5O-^r?lIBj(D+
z#_R99-VHwyYgyz|syr!6U24;cAl9yrf`PVaQa1uPmH!z_?GK8WGNcoNano$dz}1o>
zsvhUGKQkkr+DZI~oG@HMI~Hc5J-dms1#cN_v*v9%KrI1`9txisxS?(Gbuk}M*aS2G
zHb;)`1AvG4vWU$_=^P|R>3CA2#IrM%iyoF2tFG2WiF15*o7x#4mXb;`k(Q-?e_{HF
z*bJ^6#Fb>RqNCPV8MOy!xNAFv)=m^WD?nzjv3I@M>mg4i<w!Qcmi@_oUi5P8p+=tt
zT}gw(GC)|zSEh<e4$uHExfReKYMPqIx1~nlnTG~)<M};+a6fi%QF*pA4IH)*qu&RR
zXRqQS!{CEL??#jD`dI}@&;DZ|Y)j{%GF;Z5WrwWb0*{~PhYL>@Uh0GEAin|bc{fqQ
z+H7loPxqZk8Ku5Tl?}ezNYkw=&+@;F$7U2bcAKC2@p$k5a=m_0V*(&-fO(WzGlEg<
zZA<s}aUH=2Jh1b6R?&M2fz}YyDp0|4Dm@4qW~dFbbTw5)Nx%(@-LgCOEM$Q0)R2ME
zslW%ibd`V56on}`b+%~Fy~O#+?`&<>EbI5|5hai=;Crvm*>?DMM@K!FG<Doh!+E4$
zyB4-W9oh|#*IqQT2>kk<hg5@tRkz%U0enDJhZLiQmUis3W^Fkr9`X()9zw{X*Wgy=
ztZYJ}UBl4cc%UNo1=X%Gvs<XH;Y;a5yWrj3-GIbSCFriuSFE)P)=Zk#6>KFZIYY)m
z36dKVhQQ7Y9|!;NwSgg9Wbe5TN&m1M{>Kr&<mw$3R^tI}Sacb0IpPr)*>QoNGNdW;
zvA|h{a|zLRtH1h^bi7ZDM4_1^W7{9wiuvTc#0iP|{B6|Rv|-QW!pJsHdk@%7eDV)_
z_qT}>cIM)<xxIIr2;OKe9t9X^JcQM=4%b|OqUpMU3|mUq?|Uo?JBr<w_^umTb5$y`
zp@0b#p$zUF+`og&k_UJ6pIa$7b}p@Q*z9z&^7FvIfYv6*Gj34$_>c$Y`8@Vof*rD2
zRhB(5v6lvWuwX#ij8$jZxNpH8>34#{=j;n#F(8LEjm{g5Nbj{7vMS69Lt7La78VWZ
zwiw>>&Hsnz8`#LqM%&A51YcYWhU$6J<BOgB-KT7ns;dFu@tqP;FK9^U^6n(u3tgV}
zSPuuRw|4{drD!yhHzvUP>T+d(RscDJVFnBrAZq_bTMV}ID?Xcrz3eD&qa|H;a#Sv_
z94!BQ^AVrYd3Kh6n}&XB(#cZih6AL{zv~3F=^*T^Z(x0_fq~ocgM=Yb`MN8g+KejN
zl2%}$j2Ic|gVk@a*FcCZKzHKV8GG0pgtpkAR%nwF-4`<#I;6t3d%8!C1hcVpO*cx#
zzn<_CG)>CBUkOZ>m7A$lDqqhQom>M?4t=`kNBCnTo%6Yv%HgEw2SUS7awr6!u3o<i
zMPSa^TjT8A6gu;@E&?D;Y{lisGbMNmbF~@qt}q=4nrVcO#XoA)f1#wAY0_|G5-3)F
z3hwG6Hn<UjQ}-?S=O?pKyI)Q*NaOlQJF*p7UPLfObcJE!B@U49ca6LTb~frUkh3{P
zmnt%>3mO*f$^vKlvs-AI9G`Fa3bZ;E5;z1X0WquthTG%1fnw8l!?&wt1Tl2>T?1gY
zgxNNoLEoq~pbJU^bbU$Y(An0gA5WDb_1UWBSasIjsV{nR&~SQr=@I9{XuTVF=cQ$2
zQr7(s`}F(F!xj$%@N3;%AJ-G4g)Nh^09onmu9~hQ?N9FUc&}ipk<{mj>h~xFeiIl0
z@iA(F!u^P-HCvgRuO{9$!TS%cfV;2$ChQqb^4@r?5Kwqf&I?Gnk>W|EoZgzUs$Q*x
z^@R~9=DD}dxp2q$jBqDldSa9DTlD?iqvZUmWMUy5?}U$FS}rua<l+l+=~gD-H<j+7
z@~FnQg_*~(7~l*6KS6iRi&Vq?v{bi8lAo3dcmp4}_`IE9(10cc=_wbJ!hcsT?MG?Q
z9E=p^`99nv$EZku4iaM05lXvlOLv<rNcY)KYtvq8`LIUW_D1H0O!!Uqt(qHOFU`AQ
zXw9iYuOU@^@2QF%JqpmHxvTBWpdoQ==gxqMO3Pm(hs5$P3ys9)5a%TEXspIX)s0;;
zYU2gTi_It1bnO+!6bF)fD?Lpw@B>HNg#DZC{JQNz{sgIDq$J=v6sBMedzRWKD{Sp*
zW`HGcR0U4nkatk8nDg?o@xqb55IPp8*!~41j-~Q?@*`oD@oFSwtnkj5xK-y%LvUO>
ztwL)QYt@2kzclhJd%D?i)6XKX%UC9r3ry54T;WD&=UM3ZXIcSfBv!p;mCYy-q~Vs_
z;4h<ly}92s0dG$f3r&5@`T}|Dn?m;u9#S+QXLLQm%006zsHXt(B7%r0es@-OiM3>s
zZ|UyHT1h|w8siiR*N#Vt7gky0U4>q7Q|DfoD(#~c*KV%D-S=tR*KJQ3f6tMAGwDd}
zwzRl=jp;)wBGi0xzH^4<Hvvus%Kn;{ohIuFSw}2XD82i3t%R7oVmxcvui*{gRR-<p
zHhZmTSujGN#oWAe#z|(c4bLLvK9vkE!2>rz==UWUN*u$6)Ug0L%1@IBohATH!B;PM
z{w`ph3QP~x4&jg)qKN%+?3dcGM-Nd40H+IF$_lO3!QT{3vaEGi{=>R<qU@ejAYj;z
zw50E!?lhDXpaD~KKy6G6IZN~9+wYOHswz+W`ux#kJj-`7KZO?#j4kEu!?&lmOEro2
z-TwuF>F3ULC+BFmB)nmaHx_lFZ1IgPSzemn_F^V3O59rEx?-RNk$Hk5c8w7mYXI2#
zkhA|SmwfV<s(PN?0lN>KBk%ZLU6=>ypb9sFr55`UCA;B&`(0|Ps&INOyO{cj__w_Y
z>H5>!XLoYp8LPZ!1#{)iPP=h-=AWx&CRuA4Qqt~BPdgg!7(nlwNmHTwcF;BS#$Gx<
zB5vC{xbU7k*1O@`?9?Lp1^+z<L)HxugG7xgiNkFB@)sI&NvD`s1$A7K9RvH|`hA%?
zp~)1D$RvOOzZDG)#B2;BRxX-_eEjnh<v5p37*<=1lc#5hHM>U(Y$T9Nuq{e}yYKfe
z*ZOy>rB}Sx3MlnbW1UmezOLY-Ai_@aHU+k7%-bddPQDKaBhjD7;A}REYawo``avB)
z5DX!*zo2gdv`l2-J|B<?Bo!c+7wh4sX0VQu!nV{jVJb9~JbqwiyAM{ad;pqeF#)&>
zDZ3@@Q4=j2u!6Wf))7PjF}0U$TdQI(R+WbA05zxtIz&6MJs6=KEhC<-X7|%xFv&>A
z*Bep2dSO*ABJ88MGpSgG%j{~HOLeUr;5b)R2xr_>l)l>Kj1j@D_^!+{T`@%h0C!Rx
zEXc9C8lv|sCR4s7%w+6MIhY+vnh3ZjCAH_kiTjSHi;kGSZuCCN%>D;3%DrEy>O6O5
zRXP%8^*4@csU5Img)P;eBTZmNQQq|wRUod08LO2fuJgF;=mk9>5E{SBC4Xpq#mlTL
z$Jw)A&ySc1@hT)?JxU%1{x%|oB`CmsUZ6SM->LO$=>Kmx{&LxVFUkLFSML^&Rr-$h
VZ;T=@;JZM^1~;!2U%C6}{{Vwe`9S~x

literal 0
HcmV?d00001

diff --git a/docs/demo_splitting_loss_tomography_files/figure-html/cell-11-output-1.png b/docs/demo_splitting_loss_tomography_files/figure-html/cell-11-output-1.png
new file mode 100644
index 0000000000000000000000000000000000000000..6be1d5f5123e6a21a263ce7d3a684a84683756eb
GIT binary patch
literal 9727
zcmaia3pkVi`~Q48Q%QwNC?&};hH|WsQx2t^CUe?Y4$C=XloBOWPB|xsm2HMO$EX}?
z4#U`xLy_aK#Edcik3N0>-_Pgz{r=bAb<K7?*R%JY=YHS!>wdjn_w&TU%wQkyF<uY|
zw9m*;?*<6;mk02fw0kG;|IP`+>%iMZe|;N&OC;Ps$jKK5GIjF5>xuODL^z)ag!%d*
zklxDj7vz;?Pq_R0-}O^hQ1JS{C&(jx-4t3TY%GCA_S`kJ^#g(Uowi;)Z*?#T5J=L~
zNbmB^dl_>>P|xGmtcHa{uk%*N;B#O^dUqo3p}XMqshW4j2!ZFZk5<e}5`?o3lt>nm
zZWhLhAGooDS9LW1ilLbU-;FzbdXUQvs*-p3qI(K9Uzf$IAzrGQb~i#Mq13-Nv~Iqy
z@Aso*Y9tJmw{);i>5~UDHGse2x)A0?wtQ-4W~Lx;eXiZRCf2qCAHEs5>;r*Ld~(O}
z07s9+s(?WE_vV%W@p$e8rU54?J!?J&0$qCCWwQ+g`dq9}wedhAQ(nAyv8oMx@&02w
zaBMkTCdqga4Mg>ulAeoP4vD0lnvs#gq+WXH@9+QlrKCxoy}Whn87Zm!+}u;Fl4M!i
z?gnLIQY&mSln0a&rT^8&!^<n_!-vb$Q&Z2EmzUGBN|u)crx}a{;0@vBg-1NnufMlz
z*RFgF#sGyvxw^XQ)zs9;Pq!ZLzbPw}Paq&k=BHlR%S#<Pbm(R4_{7@HPS+#~4iBj$
z)9?`BS05hWcfC<<pIlT_G!uj{Fo=2<8~fz*+DP`?++1%s9*<Y~>K~RzbQkd$_qF}C
zT6^u<HNzZ>x2wg#os|YKtWwj`UT$oJP0!3c)uMHFK0h3ze^g8?`N@+f@ND%7Lnt&`
z^VjE7d<B(-7H8Yo(vER^_wM~XHTCvcTpW32<$~7Q=ZqWgPK&5|z70z;GdFLYA9|0C
zXD=_4dwLSDnVNq2`Lj5&b$n*Ir{{3Co3z|ObK2)MEh!0!bTwr1yLZxyCBwrG(n-dA
z6MQ1ER2uEf(9qEEaQTZD0$UM_i`zdvJ#FOQ2JFg=Ldz8YBQk*kgmFf}T;QE|kiWm-
zyPHqv1mc!C&2yC=^rD$yYt+N)df_4<U`@^?6C|Dq2yh4fK|v-#K|$p>i?cWDAzd~x
zH6_XL&(YuWTP?%G>-*oUUc4x+5j3B;u&^-m)5ZO>zpw8N4u|9F?0lu5pupAL{Vm7F
z+}xZc3@?+aym{_RL&N=N@$n|w+3~<{|C+OwPr*(Whp@e+Q(E_C%inczaREXWZ|LTR
ziCzgm0Bp#u*4%ymBk}v859@E9yVZK6D1!3&v*D<C$=i`P-ngcBp`d;H_p1+;B=J{w
zb#>jbl~2jt<t;!S_w61Uk^`;`Cp|)Ri}xHM)eU@}W-?Q5zYlB@^PpsDN&E^^7}!b#
zVA&AebM>lxkr6<+)7IR9i!L}XNI!La{3`*-0?sBCW8RXbT#9Ik6YcKmItd(8QdB&#
zi(f1;RGZ1YY;i6r4~xaxX=`fAR@=)j8$=O4=qM>kG&#Tce(v6_W6P_dVOf+PzZP;<
zwI8f1fx)LKnv2h(qxbUjTerS{{~p$wEccQ!DVa^NuGg#d9y+HHIQu+QTVZT$Z0V+~
z1eeS0?TbO-9O~@UH8fz3jt#+{ckfy(Ge?j#s`x!4rkqDTuxZuyMQOVa#h(1|u>^?c
z^z3Y6Ny&L21>TY9UE<p%Zg4Lv0(<qf!V5cZU5&wsi*6;}aIV$EdW*A$*48PB#*)C<
zPBwgTWbUnVNrjt)&<P-;7nKjNY{9WAL)*@4Xh68(NF;$6?jL>5s9G$&W%NbsxGS)v
zin!v}N@Pk=k^J=he9`h%&#x77%NGDx*(wtt>V)OSit_R=eSJ16HnsNhYAQrRH{>9f
zCn6GSBWtXkow8Ntb91bX6IDGq#@G)~Ss|FS^P4mtDB{9j*PEQLKYskUkUiSi_-J;Z
z91GXujrHr_|M0Kt+qP{};Tep)a(U;@orNVD5*v?0^?}tMJmBdsg_)R!T9nw}nt5)a
z+|ngeTEBe$94%{GnwMs5-q!U|71)NY#PzL~1+G&`N$G~@Rl*~<e_$Z3(ZPybi5!vF
z)z$6&by`RLT%BZHU0wIc2y;%LdECq0-7s)&KyWYr<%ew!kL9T;v8|VoNak!q!=-=_
z0NYfvDU`9JC-GNu6g4z8?Ja3TLt0;f*B0Ue2Q(pP55xH)KYEtfWMpP#6(vtDuJ(s&
z>p*0M+%N?N8NbNq;cgx}yLRo|c_$7jiUF`AE-p@WF2Ln`YF1X(-0UoFSb<jed$sfD
zKd`+sliND3HuLewx}(wPQWR@urZe2%>@uj~qn^t?8`mvp^z_6Xi!TGxLWu)%<oSc-
zr;lu5{vUKV(WDFrTH6b(Rf(C#7P-6?U;DO8$A-wQ-{dL;YM^8p_qygfOLTrd;O}AG
z8zkijz|h{CTxPfnJ3coXB96^+*#^2l``Yby-w<l&-T4aI#X2ifRQrqE$iin`8Va%T
z*;o(8cusfco23%0duDi8tfyg9y7Sqee~q_hmQGJ2h#TZ2nM@@{Ij7?m``p?}$4DWv
zu3&S0QU{L44i+jn*YA=t&+;d0uYc!HXF=@jGD>p|;?vfiM$1Af+N2k66Htq9abv1u
zue530?`vzt9Uk(Oa1x>XmQH)MUA98b7>H%)8%`zuZe%HgM4>x-u7<Owv$k_(uIm(q
zS2MV}J_sZ6Lq1T0vkvm@q?{@x?x4?tuh01h67woRFO1oywM7k%Uj4g28Y*7W%}YJk
z$ZJ<g4%O`=fr8rAz@C|m5N4g*{M!6`<{4BNR4FpRp0SFaypsf(T<XwZzGC5DbMq!o
zJ4@|b>F6~sEj}n4%F;c_XQv!7fll)otk$H1A1oLVu*`J^dIN5sQpOf~uJT00Bd5B1
zDaC6rC1}c~#(Roiixc&BjI>n3Dp?ZuIU6TCACTXEn71iMKO3TUU`bfj)))kmHQW2}
zum5iHpN&sBPE0=IXxS1xeWAE>-PLXom6h|Vtg|HG8N_}s<Sb3TY~o#<ZcI$n;*JJj
zSDRuzEx?bGzE8H!7^^=Z5Ekx;(m4y?T91{#JdeW}yx@p@p>-R(QL+Nap*j*c568@U
zW||$nfqa~GTSjKF$IjcNF$_-&UUgu!wLP<)PdLL+L_BsOCR-SPQs2ecSkA--pc$r@
z3WN4wP}!|g(|(a*dE0DTME7p-jVB?rE4oQOCJK==GcDA%{`_nWa6nX+^Xa+g9!vLf
z#tW7hyE9D&4|ux=ODCuQ1#<Lx^LG;dAuoG&y6={$S|4qSnIKsemkE~5^MNa7AQV_;
zLOin?HzvQ8I>cnVU!rv03HFW^$VncfR=rtREiP{lwEKR*V1kr++FFraIa!w8y}cPq
z_K`&-o7hdgB*~A284*0{T21q-HxL~GB_u*s?QLHdxx)lO?!!HD{f9ZvNuy6??o_aZ
zE$nNp($ILnm~`@~GVramDs7;iCij~fee?y0;s3r%`%Dzyx@jwkV)H8lj@#OYCdVD3
zDpMJLT|7TXmyp=7votduG2YPF5X~}AtAXn2i7RQg3;0X-M6otL6KzK^Z8cJ%^hVT1
za<Y&gv~_iH5~!?o{n-Ze?_`v`VB?cQa^+=+nbs6W5J*fr+sU-n@r+VSj9DrR-xv!o
zRO?VSJrR<2X$9Vxt$M0Kr~QsVyK`nr0HwfIOyIKw>VF}ECJrLt!v;dJl7`CJuMF>b
zm&p)nOW`5o<fLJAK&aVS?fJoKKb5b_%E~HVyZWL!Ro=1#3|vT8T7V>tIrMjJHTzn+
zM+XfDw@5F_Il01~*d4Ve^`61uKIFBkKk1t+46zvNkTIP=baN0W;{Zx&w2nxxdQrJ-
zA`(&TTRoe2JDkg!AYsh21If5HTPSTZl25qJHyeEPDeJgpW5=0{)=6Wcy_GU*DBTB7
zbxq)uUc^(P!KvA}^@zxbtlwPM1i*&0Nt2iuTWSDntY-3cd3iZp<5+G%ys%xiC6$st
z`P_~OH*H^D7;C|9((Cru1f5zML`-D!;rFEl3V4D*#fm@o@&AJ_Z{G%f<UliN6!%9a
zK`w71B9`fR$Yi3eiRhE6Pcsr5u8-A5LaEZkzkZ;3I984of}lCRr)H?zXbN(N1>^~c
z<m4}t`n@wpcY-8ig}Y0}v3pZizMB{>lsKd-IBFv9lN2tD)Mz9AT1_~!Z-tD{r-!l_
z`)Vrlg1PNB1LUA^)l&;RpWm2dxyhBO!UhLkqoCZA`1)(Cl-!Nt_$-qtT_gIw_UQth
z)5zL|=9x`G%}ilhivtkLK9?x%;?dFWB^GV+%RtcoNLLfhy@gTox+YTkZ2G4V)QSav
z$jpVKb*_nG4ojxH&yDWqCjn5X@|82Vr1(y3XRLNuy9&KWvmnX?8ZuX&=2e%NBsTC2
z!W%Q54+2$bzHq$%xHy7_j&&beS?=1$cp4wJlxx<H(cvO6v&&zRY7EHbMcQgiwC3je
zqFKKPvi5u8YD5~|kqPGMwCW@YzGkL#p%js`&IIFckPjb5P2l&Kb?l6-wlZuDtv0Xw
zFpgL*pYPXT)>5SzGc!sIhWJ~F`S|LmQd##tO0P}oY)&U;Uhw&d%RZC(;U(B7cR23>
zSmwye=^?dt-rcse{{HF2xxSK5gLF<t>A^5Py|C3*s5;EbxA(!eT~{#CXPk~4{;ztn
z8FlI`YrUW^PS*vYHe^H@f@#ST-c@=`Y$e)H^VZZnrxSeYIwxS7eqKf877_nGZ)|jO
zDY!|_bO#6|#Q(1<lS?(d5LOViw4~@hSZtpgbs2qBjz&U6UDo|n*X3l9dFw4(%WvXZ
zd!YM~n=ua5vQMwA;10?|{!~-5R0WY9JNYl<ip5?e5*b7Wnv*tWHcn$+kpby{Mk_Uw
znD>{F2!y99C{s}Ttahd;;2Kckcvx8|1qn+at^aB@>!+IQc+o~Gk<d4O2TMk*JrMFE
z264AGd##}=Y{3#DiYluHyX^<%_F}CyU{*VaT68vih@_?<iY@zD0EMIU3hy}rw*5lm
zdPRn>lt$ngsM)^rm{}m;)wY}OFj>a_W)joM$B8yAv1|5O%zT`2{!2?ZtVJziW8+{6
z@g;F)stZVkZ{J>5GnFZt)LH72q`afa7vQGhHkn!>6>vz)3;p`j-SXo3^?R41BC-lj
zaK0blxO9hS<ZyLkyU%J#eH|$G@Tg#kyxIKe|LFwBo};<MC?*}|0mxE|;@SU7TE%0(
z`~240yUsCIWJM3Z_Z>H0+?4PhOew;;RQi7_iSvmOGvC!cbD^a3hC$Buhf{vj9<R4O
z4sebn?dJrnuZ?UL^~W{y%%B5~zaGx=uu?2kaPQ^~tcXD2LB0Yn!nq8nSwAvtF<o!%
zNpvlD`Kwuf;Vo<%dU3VZ5-w?pgH(3lkUJD1M)D;3ogL}*EE>eDznt#iL93syVPq)P
z72u;ke@T#P=U5e2A`vo-@+_V3Vd=&8geg87_PZe4H(<X*S?U2~>9uL|@R`IR-&Q#|
zTIOhj@nHdh{@P%oWd(wI2GhpQg0P+49ojlcv=9v~EiYQ710AEylE%y?Sv#C(Y_3wm
z1IgXBbe23m2}bL_K?<4APjT=idy`h-ej`_Mz9pjP{j&YYp+*pcxZ^%#@2c-R{YLU|
z6z!48MAk*Oi=ivUw%Q|mb{cj5O}zf)AV|j9QvI^0jSdn>oSC#a_q}mh38pPl?QLy6
zo>8OC{%z*q0@IcgQ}@)kgX$gEnBr>nsbZG&fB?_rtXd26`5Jh8TH4&hUHn#?BQ%=T
z;;SsSJh)#01hRQGWxHigDAu@i9<ZuN??}cMmzLvjvNE4;HqPs>&mqGZiPN5gyfPwP
zbY%n#Vb>q1PObcCS46h^iyVZv1L!5tF!6iM1zrGrjxJ5KdS9Hp?VwLclm0q0rhIQw
z`J+Sy1~q73Xt%6N;u+meXP6;C)YV~tYy;v<lrw!?kTmBd_h39ko}s<WMcW)*wA%UW
z@jj6@4!_pYb4&Q`P3{^yorSO0A6ZKSrz^Mn-m--cK}jJUr9sT$#x*;3MpCbggwpx*
z-xF<Z+YvU(+gFsL@om;x$0SGPSF4`?2`E63mI$IlcX}7Bx}dQf>m&zvo6Q-j)~=ij
zWm5|*0=cu+qekrnINbiG`;diJOQQC2X3(%LafnK_B#LIfhx_`LPMp_bCa_wC0eBBx
z`ls!4>H}z>(+)+<p5=;&Vf_67C|YYrOtgmNwE>RAvab=XTNiaMo!0{)7o#vW%v({B
z?6V*@=e(nY0c+RO%L<vlNF<e)B{I<^e^K!b-Xq<1Muc1Xbw=Ba4&Bz9y}KJrQh2V_
z?m_Ih92FPc*IKaJakAYPLep~LiVlXY&B4(gM{#c1XqvQ$h=?)+GRZV)Hbn9Fgb?iG
zaN~_x_eP&IvZ$xEPqXkybgOLT%aacGzVE<z0{VWR4yO8LRY?J{dr3TR0~8nbnW1n+
zODY{%ry$Oa66T)h4xd74fTPkkB)VC-1WPyUl*@GAd!P*y0U$|5FB7xz{lis0xhtBh
zKN5ypK%j=v5C6o7;>J&}QHayfK_h+t3)judbH);ue#A+cWf-FDIu-&vueuNLi^jZh
zc{G~r(|uk&M>2cK^$nn>m;eAB*>uxhOr6<e<zYDitE17t;vcU@@EnscjkwgEI*Gpg
z!3tD*Ya(4>gjcB2YR7?+ZRvOb1xD9wVmaVgFAumk^c%k1E<%ubZ>sYok;LLqao3#c
ztf0kp<pY;9!0c;jz58c^!^9c2-5DOojIvvW%aB7cYD4}BmVukxt*m3Oe~%ImB!})f
zX!XW8ZTkawE&WvpU<3t_!xmzouYm0P^^jkCD$!PEPWJG>WCYOhYwq+Grg>>4<_Jf`
z?N^DcdCe+2fqG?ZJ|{qzEE4B4?aiT>27PD9uTC9!UW8@$T82yo*Dv*mqK-_Br$CD~
zHW#xuL)%?AX<hC7KwR}G|F9DkOD}gFKq#%^JTGBM$f~boRbuHywB<XS78dNRU5lyZ
z2>C{7+;nSk?FNUDS^Azy`YKH<W$EcTpjo#tv#XQYLFA?9=t$)1>6oM@w|$f~;AFFF
zr`iv?vET_b?67AyxW#nQ4v$C$v~fnMM^SY%4mso(ueJQjQYdZa>=P7Jrqu#QuVW80
z67jag=0UpMF_~jm%Z{Ah@aUon0t86};4`hNoDW!?43=>E#`FE{(gh>0+zW#TSUYql
z-&BrG_){5^nHoVwKrW3=9}ec)x5SGRHn!SaVeo81_PT3(1@jCbMEOKMz5a)!{9Coy
zjJ{G4l?q%KZA$YJO%W0{5E`JF{FwUYAu3}fpHep|@FrVhK46N=BB8`oL@!|<+6G(-
z9DJ#35HE)Br_o)<nq!A+{66rev3ER!eA!O(VE0xg$CO4SQBP}4tyO<~`k;&)${gy>
zH_23z0S&P2be2ovFT&V^e({!pKRyc(H^$<1<`cEnW{GC04wV&KsK{1>@9<bF@&I@@
zo4sQWvqzos29;uA>__n$DEk6PDCNbbU7NFRI(}r_8a2hyaR`cdX+C4x+%OnZ^zilq
zU>4~3aFL;7B4+(D3M3nfG;IIB)bu}YJ><=p;Arzkz_aUT%{*6sjjMbG+$}hS@<}MO
z@jBsp;q>C(CY9-yTo%iPQ#&@1b;!p>W~o}A5EJ(VHCIl*Q-R-O^t-3?ZmGWmL&oX{
zO59p2BVy4!ZB`)PwAaouHn!%R<1Ok6(9SxRO9KjzwI--Cgn4&hu5F$@AP_vBWZlNG
zig+Nngp~A(UK!DiNpf|Pv{_lPy8aW<%nM?P`N?miQ3n@kBofeoD>fd;N|c6gh7mVD
z@$1Yc_NMRsjzw~Z+S%m8hD;ME2Ou)|86|CPZ`vjUx;gOEk9_R$Wx!GRkCT4;*TCx<
zh%NsEy(pBwb-9@D9(>8WsLZ0@U9qB~VrCQYJMM^47uPa=GoAYVTy@r)=B|@xhfJ|i
zK38-NbE;?F(DhsLraj)e^lgFFrT<Y>s42U~>~;jqadp|$ZiM`E35EhvRDQv4+Uc%<
zhF^E(@wlW&IIP&VKTtv<XwWChGH|qU561J}y*?bW&PtT`#AyA7)kGl@A;++@Wk<#*
zT|bUF+W)HEF5+B_$%)nZY6Sp0%u<tj*K@_WpK(YS%g&N7|4S!MEj<B8@%#xJg0M3L
zuP%KO-{{1NSKb_RegPmfGLZYdNWp693WGnC3&SvGbY}U4<$F9C{H$^hy47Ylu%$H3
z<rndxdGIwr0;zmO)^h{f1Hl{ZsG=!Gx5tASy5Q4Zs`w?J#o|bsrYRK!M3!&YzY6Oh
z|HGNSgwE$se>*-==Iba6R)yI9o=<O4E9W`)LPA2?U9wjirS4P&WIT==>9RymIZ;1F
z@mp^aqL3^`5B+pmZK6=+VJr9Dz@$WW|LG&IPvT&ap^I&{Yj&5{NUwur`ucbuvE_1H
z<>)^`<Yn&mqKd$+-?vzt^x6QO1gBLlqZgWlGnGcVb=*2Ke8@}ok4TG4Noz$(tg+ET
zY>{zUwqZ>^UT;Fjy>?hKF|CQSRy<5(qM0FZE{j3efSK!X%Pc1#qb-~@cUrgoV%-nn
z?%;0^cD4P2AnSUizrGJ5ST3j3ujhZCEVJko1<22TrZut)if`kd@HV*sD&&(G3aYi9
z^M00fH(Bgzj9Gn7%I$7S{`(b{qA5^h{n~-Euh`XBT1}=;g6l+U3Q}se$du+EGIelo
z=K<8H^9%pW4BFmFRL&Z&%q6Y%!oO%F+U%M{VCQ>5oR_3&bo3A3i<sG>_w&AF(guqF
zbesNf;vuu=F>S`D50ns`(vFVeJ?8knRgm$FNt1O$;?qb&E}+Ybbd|G4*!DEE2e&;h
zxJV9CaRD~Moeye91MMm&N(_imbf%*P0cEXd7-HYUY=v%=S}s2nUvC(+ru;l1@$1_|
ze%191A*H`%lk9@|YdSQ%4h~^I>N#X7$dN1v=T13fOjS6j`2hYhfGF64@Nvd?p?|yy
zc>Wsu7(fA_=p3kzj~`~b#a}^hFayx?j^F=nDg8YUGdWIF-BN6I&P`7N6*~b34;!~s
zMozf%wG#^}b4y&*Wy~n)LUz=fbZ8wapnbU%)>^n(sdQv44j<_dw6xG!(hcucNehq*
zEs3_;nIh;n8DeS`>@PRK3#H`*6j`AKG(@ORwU&nXP+oQdF9@OIGG?jevI)_j#W#NL
z!puerJKQCAWNNt4DvPiRe!b3el`uKF<V(#(YRZR{K=fC*Ygotl4SmvNp_=Y>LcFZA
z8_hPAEFlrE&3S8CfgI5W$bXom)+Stzt}?=8j2niVKkYl(NTMul#oVlaWMjFKW9})C
z^Djf~zjv>tb(@HH34Y@Nf7;XhbE;2L%UkrR6V*ozl~q0^J4d=8DUOy$WvOy&05aXS
z8S}_xaVcPT(`b{ZNz1P3)v+y*Y5?4x-7-8Mc{_wl9uDJQv%sx1o0g^C%4Jq2R`38;
zWLiSP%gYPSWrj{_nDu*A4oxJAUbx$Us}Ev2U}jeq#>`TYH(X_WKB5SrA4aXJkqB(w
z6A==uvHMz{+)MfAp(o{JTr;sO?Ui|E_WBPgvwnbqXOsf94P>S3c1BHsk)k`T)Pm-d
zo>T=2kGRje2pjTKzh3gOa<AZq_`fhjKKRH@3M22mLGJAzHSg_%5K*+wo)UY`y{;Bk
z=cTjW*C?m*Df%ssGquq*oXynK)IJwRfIiwU_|{Yo4gjjjoT#~}j$QxtD7Oq`n+KRJ
z_)xsqKFhK?flAx{5M1}1=4bn0sd5;FwGF2h0O~#IM6nx1jJPqfxe){u(c_=6g9g@v
zZw+g@??-b<HhY950<V?^;6s_^%&-Ncnsb*=wZS@Tn`%^>oVx~t7AM*PQ!fkhWV?h!
z{tcj44->(cH<Q7H`lHxeC$u~Q_6ocJq>CPlSr2ru!IXRVH&Xyy-UKZ@&)T>V<IT`H
zT!(aj1uhfSO6h!#n5x#DJjm_?EJcrs02*X!o7B8LE11F0Ty^jU=Dpy51MffDx_Q1!
zfxo7ezXn)2_-wIFi%Vp3?o(dVApu^^-gpDTL+uJ}bF@Rwy{f9JdXzI`ixlgto;~*Q
zj;vj3w8Z>iAW2Cp1={pfP$ol3hS5onuI%Wg_mwLdBdf=SFtcArHY#kx0khGppAgRV
zz|4M;$_6taso7RaPqL_lkDMIWIx7a}NtKbZ#D-pKzDbJuZ$Z1dlj^!XrYAcmbwb<y
zF_>-d;eoYW9uwaLx*m6kJ7Zn%OAbBynG}L3%KI{LdXX6HUAu4<GaDbi?lQ{Z*Zg_k
za(QJeo*rP9qWrI>E-ALN_mgzZ+C}d;MeeI2k*DD%%FHw|30^IL(}hSK)CsqD-+g9o
z(7mfgm!4e|4j4XCe+-{3zpy+bmVoKvRj-UFOi_S((<)I0cn4#sQ0c9C4f^kv0A2Yg
zoKc#NB=zHvta}^4CEq3Q7uU|HA|?5M<Ljt&z#<XSF$zZ;UV;~#;P7S~l=;gv)`EF)
zWd>6{iDTE+*N;fQ|NdCyAU%Ak%b*=C*KTtaYfCS8cxJc~5EP^VqfO>eNa(c;r6P8I
z?%z0Ykd4dFjF+tuRkQBTHzYnTC<wWG#l!m!BJAj+i739NT78Nnm|!P_k_7yev9AAz
z=wGl_&y|%=mXU96)07sphRf+64qC5T?6!>oh8xwnB~Cfj9o-#~Le@~Ia%O(YnNNhH
zxK|fP*5I%(I~a+c2%hK?ZgGPZag-RU-h)$#XciV4Kn7ZlIx+0sO<yu@uhB?G2rZzb
z6ro^W<w!#!5I$ra!wvT8Y2V~b3`5gz;+Z8}4*df%P?+kIV&q~<mG!uZM310Dzj{^v
z;ll8t7Kvq*%4Bb1@~%ThVffHVD6$!<Txx`lW}!!J2MvuU4h4;%7w?|=`M-hW`yatH
z|7WIyoUEKaIkc=OBN6VPiuMFNK=tIfgzibP*ps|?M4>o;dy<fWcnHa*2T1ur$7*S?
zUK;2QR&)A>zC@s7x!?)p4!gCi)}SI$QBkv0!B=y9KXv@4o?9+=n<!j_ksh!8cK2lv
z-)8%Xa5SUF#>8+nqFs)(oriJE?#ph{YwwY@Az4I3Yr@XDwc!@9k?6?!gc>jp$SQwn
zuBQhK>H(`G3#(S8op?6zU7ElWMaf`cp0wtT@dT>r2?HQ)OyxZ(kDV<3@gN8=MG{K0
zlpW9)vl9XLWIC7&S7;N!C4tHAIGd?}Navm_^pooL`7S;FL!I{iKo@oT(?x%S-iPlj
zIF1vsjQ5qfG)y<Sv!f33U|m82W0vVljyryFG%+?dHkGxxgweSO15^z#S|WBq#;qbQ
zr2#XGnmpUN4=ud+Dp5^nt!Wbbt5iF~uyFX&5}1Oli<gi9=JbFF5!}6BOw*mlWHN)u
z1sKdBolX1O53B998rb$Y$LAe>6fL)8WPfRsv<8~vPc)w_1>Az<<$}0*Le%WBx@(q$
z;FCEySGo7xYhtVDg2y^oa?-A%U)58pCV#;m^MLM0#{MfFf7@C(C(EEI$6oi^Tkfc5
z7K1itX)-AJoT`#eUrR)X+~F2s)a{(UT6c6EA*=9|f9w{-mAR@q&BX6_q}*|lTk^dE
zjN1ER(o^KTeno_7&ZWJ(p3^*FPY@j#7???19p=LhRz7->Z!Zs&=D};NDGF+gzn#I~
z-XUJn8hD~%G7+%PfH7lRSDuORlfI)NR<t}OsiM@Cl5sI%Dp+q;T!-W;#89y@QbHuV
zmJTy`8RfmixMIT^+y^)?bEy+%&<R20{l_mj(kg*g(fUC$X@UizjITM}W|o^JNP42p
znqij`w4TC&ajNoO0Q3LDG;vOgxap#?s^I!7pB|zN6~O$E!NunGTktN5pjz3{W3-<5
zs2X(^BaX&@JJ$jy_79(~pw2+ornMi`?hnY`RqY-2@U*0xe?9t-$Ay^(RC2@M9}N04
l-24AMne@lT{WGrJ;MbF1FZFX49RZdC8R?tp6<u+R_&>G<p_l*w

literal 0
HcmV?d00001

diff --git a/docs/demo_splitting_loss_tomography_files/figure-html/cell-9-output-1.png b/docs/demo_splitting_loss_tomography_files/figure-html/cell-9-output-1.png
new file mode 100644
index 0000000000000000000000000000000000000000..a1cfd47680259492eb11dd53390d98d8b5a5c947
GIT binary patch
literal 9916
zcmaKS2{>Epw|+XP(?NBjgQ7=&TCJ&rp*1#bb)be*M@ZCAgqEOMF~rnz6jfAfp0!0)
z5)yMrP-A1B5}~0e6+sX&Hr&tYx&P<%-uwLXJjqV7lfCy^-&*fl@4I%Gk-<%&UB`Ao
zAP}M3x31rXK(;!8_mtmufLF&8xBdW|%kDa6?#8b6?w<D_BO&_t-5;S`-BAy1PIw?6
zKY8fta#7)u!bQ0g58T}!JyBCsbpC%oP;h<hpx8KWW(*Fp^U*E%69`23{^oPb2ko4P
z5Xh-Bx35G0^h%j#`n<F-;Z@I>h-7wjsmq@@bY^kysf_)0H-3{k5}kXVQgf!RuI@os
zqV_)hI}>IHm49!rAD4U=kZ>n?IXxiesPgp~-JE?}&myv$u6egPKWu)dp%z~f6B4z;
zhtBX*D;5zB=}aGo@s@?P+er&kK8xPWC!CauC4`S+B6#cvmlxAjF}?=2kgucng&>fO
zIxn+r5Qxlq2^#oL_hpDO1oHI2|FsL%sewg%dNwaFFZ)UYr(L-zpA=HefaUd$ukC|C
zs&|Hl6bC+k{@iY9hRO9iGVy9~&=O9x3+rL?>=`r6imcL7Rc~{>I62nWXvQQD^QaQ?
z^kD5<KQs2^<m>3@{jBNf*qND`xH!VhjLXqu$6kXCYjV=j`?u!xVez9!qibrQM~@$m
znVWNiSNZeZ_~&n4=J9x+J32aiC2#y5tE+SB^y#d-xh+eX)^W;|LK@EeG@FKd11{sL
zhLJ80V?oQe8v0@0KQJJB@ZdqI`L>~UaK7;>JRTp1Jf__x*B?@hAuIGcnrt2@CnsmD
z84t_*OhU8m936F4ocsU7rsuaaX%iD47QlJ?`V#IK7_=>B+D7gX68h|1$}F))<Y6!v
z;Er=QpNE7X8zaSuwvBrk<xft_ugRn$aHpK;c;|Kb)U>oXoO!8=)liw+yF2>&pL=>@
zUPeURCR@5H80zcK`Ncix)WU{Uc0ZPqddxt^Z|+~|Z-@}fg5}94X!lK|#LAjdtE-<5
zkW{p^6v$+97qBTPC@8q(vX<81kFf;)vyNpRgUMw67&{}?ItE;j+v`{^|A_$`s^pcR
z7+o4OH=NEbEqyY{{quc(e*PF`%lJS?$E!Vi_B2&jUn#@_>sLxeQj0%G((aO8PFSj)
zckVCxjItOY`;?*vE8!k9Ul>#VFVTrJgJko~MSuMGZgzIIox6L{O8?id*4)GPCDNt9
zd$Y3zQ(wxBKlVDhgg=TawFEXw)PkkSvdr)HBUS6)xzhwZcZ$#FOO0%tNJvO{5a%G-
zL6#IB+!la?ImLBa2(&LRdksqx9t^$pH*4-*-g@ll(Va=QVpfmUJ*FPSnVc0td=<^q
z&63)-Z5z>>J03UhJW%pELUpjuw~0~o@C}MVl=d_qC~-a*?TP5zjEZ6ej&ZDOxtP)2
zeUgVEk7*qf7kBnfpgQeuo2QwVKSpvlLiUPNHSJ@%p9_h-FD*4Ckw|u4US^9^jFY{N
zy?m%qUauX0mDf5?Ae_(r^K7}cTBYDR2uv26{jOpcz9Go37?Jw@<Yl)?2a2#Y<z-|f
zHAq-XMoP-o*0y#2*|RO>i~UYn2%K&)!g+naMX5_Fs;>b0vG*hJ)tk5%I(FUJ`mKH8
z9#}-N?&{ha(#z|Tm$RMSb&u(OB(2-ZONS%xL<T0u>SiVW_9o)@nok5Y8^+>r;={sr
zf+Rsv=nTp2Qe8iw1wKnI<-^a#lGw^KXV^Cz`-`2Dmp<^Shk)%MoZ~@Ig1~Wi`n{v9
z%CQ*1-H70%P$*hl^?lQDC1EF$fQ`L<_R2<tnCcF;L&b*|F>H#roCHWVm$KpKaf%6I
zu9bg)yix?tZ>5Ebxs?6xsSXyC=)zP3&NnWU_86naUO0cgbB|~~on4s3pg)lRybh`P
z)JvzI2B~)A#tpmChKL#Az|rZCMH0(vtE(WEh>2T<ydG^Qu7Xb>q`Vak3=F7o^5zK4
z`;N<}QK~Yplr=QI9H?djD{5<NTl>WKZ$K%T2pko}+dZG@V);3!pPygrrHmWwDHK?t
z3m{Jb^gPV&I{2ihAOft%cxy6<LA`*0s+Unwsui7LnMc8O%M;=w<~MD+ar0*Kn2x`@
z5<*YK<%>+Rit}UhzuauZ#KmLf%}cJ=UhDY%;^~Zo3q@{Q`^wx#TV`h;J@Ez!f3epQ
zEfMuwF2Di%hK2?Nb;ph!_M6Ym2l&E%bI0Zrm}Z}ylao9K{*aBpwE;+J1y~?37UA5|
zQsghZS6H|dHPmllXxQEoDXwlmT}ha|13oM)Eva%2vj<Z*0X-$<8^<L%uBF9pb@vvz
z2iLA$BRckew5JDn+<`)>Kh@P9D`?uYcQ3WB?)pWEhX4)j7PeGxZKyp4;5#hZ8X&%r
zjRtlLs0wDoZZrcm*84AXw5y|s#5I0@#SnicZft|k2`F!7IKu~;6&g$Ucu#U3j?Se;
z@MPGwYn*_M<O+WYvMU+qQ38qyH7M@y@?oxY@rboQ^7}i~rLh>}UG!4KRy6y5Ey1=i
zkod+>>#__Np@S1O$~Rfi<m}Xz%1OgG8mVY%I@7F5P;evr)I3it$%HW~US%?*mb3BW
zySBYw-j|PxHdTU(RX-jZEe<YG=-4XXmHCXR>Sx==Nm5dJ@d{|GZmbjDoWwoUT6=cY
zKWv#(jvgva7PJ?SFV$z4VQqCUo3gd$F;3X&^|hf56}mj;-kI;KivzC7Y|Oxpev@xp
z{pP<NBktLZn7Vy#E&S%j$KERY1@>xug{8J=JZwQZdRD_`?(rFR!w7YO!W|i#x7n)G
zh1dzo-zVtLKhG6_+Gu3+pCcJ3?>^r}2#mm8$lpD%`w_hof1hnehS<yY_4QZ9-?{Is
zD&5CCTN=q~RUJ}dn?u)hqHfn29Gi{o)g=}jkgHrJyb_R?e&wrjKp8vATeJl*X^qbJ
z?;M?<1V4tsb??69rR8kr`!qT_(@9lipEf<4t<>z<S_BHK#BWsLM0ujjNiI`*@@vC1
zNiDK7S)O0NFZ^V)dVY`MklAsB$I$8)=tj*cy`7M!^<n8EAR46q35|b@XcRAuJFo{k
zTIqh=4CQ-4-c_sJ><$7opw=L-_*`Dp>2Nal&&JHL`daTTvk-{2+exuQ;?b|CP23??
zoZ;SnJa4q7`B?Gm^H}W7=tarbxZK!C3C&|M-r7=FCnKc+gF4C%*!Txe9Ja;Si6p;&
z?H+D)zKu6KTx(x`dQ9*g`uG`mNa1+_<BHqRGwJy;tdB1YJjcWXI&()WZHZLl3!yml
z?dnfgZ>q{{#<PeEJREH@u1`LcmuF>ov`~+uuHqgdVe_BjX5|WUpS$s@hE3yZdazis
z<qK9e1Y(x*0RHdRXZtqyzeMh>)688H*$8M2tOKK4P+F|WPS6;nV5b!<NJ=t=*@*cP
zCjqX{mUY;zE1ehi`y5l0efUZ5SV76!e6wOQ8*fJVBuBrSeVO{^K#}F270|W8H6x`O
z==)`FZ+hN@asti?Nv@(1F!ZxyS~5x2U#G+Cw#gl8<QIsB)2aTiY>idwwNg%}y5DUk
zu5XJuoF7ZXqfy@1yz8>Fb;H<&CGkv8)U5*)_sz?n*dg%0IybgiAXI%cnq?+W#*Q}4
zj}*G6BuFG$N`~(}v9y(pTP2vO^D%N6svKTl5j@wnyEQoGk2%~GFY&y-N8g&FGg?Te
zE1hhn16Jn8#z}FujgbOK+FW?wvo+c}*|zw=acr5MV&W&#&G5F5UwfDEv1n|puC$B~
zuh+D76*h7sEe5jkwov5#%a21d>jPuXPw*qAWUk@uQw8rU6?KN_9&gS+UAgf~ru@X8
zKDTzEV<0#iw!VLP`qTkQvFR9NgNJa-5>#RZKC(TmzLC7X%9d{q5tej;w>8|V!$T{+
zr$(&2aNZ4p$Qcy<4Pr;S^)pWn`f^({T9Q9CQ4bUoiz^wYu-KZn-D+WdjUUa8xRJ8b
z+@c5kn7F%4qTGnXJH$Y7cJO@mjyyUW$|(Py=7jf|=*Y}yp{DSNT0>r=LWPwFU)bNu
zt`sg*wlMx<@UjRKAFuD|=Ym}uXf7y<%zKd7m?WiNw;ov#>7+t*dscq^r*)gr7{SdN
ztH-iPt7zuqj^5s0VoJmx-2>sR?YS&J*7^5s8r+t+u^>TVm5qFJad-9&)X<sH_WJQ1
z`l;K+fK_{b1PlIafj8qbjiy&8Oe&?~GB#o|_`P|if=Jq*H>K{El0ws4W=FymcOYR;
zD-UP53iz1Uv{^#jJK2-dL%T*w4s5A(ahFZJ&lx0-lX6l$5JvXoe9MuL_*WwYyt6Kd
zbw4(rfKOD_Nm^Pr!&+n5WtBc94{lwj9x^!dg0>V{HL()j_dV%#*?CjrCtIe~X545t
zPmIS~g+9`It-4mDa@R*cR818&9$0(Zfwr|Ju@r~6?b=l<8{LN5<gw!KX+8>`0|dqB
z$}%NCGSBVx^ma(~$n~F@`w!d%S3uEpBx>^g)GEox&{J~K#Zdh8yPTI@%<#dd8B+I`
zz7DunF^SkHea|<OE*(>Bj?d5<!m^uL+^)5TY9^*pbv>QO^@Hmd0J}>`xOMKzO*1Ha
z@LG}o^5TRb`?;|1*KJTnqxSk-#uUg_RRFqN5Zs{}^GzqA3~;rQG2YLpuzhG<qe77)
zSAhwicN=WCZ8}r%os~zA%UKrqc2+;)`0=QkXWoq*j$kcJnlH7`DT~D!Q({BQ&*4K!
zpNEG-Bvt5SZku`#QBczNoQyhV;RAd~E#z(b<^Af<>Q@N~^V_;skH2pr4NJQm!g0je
zSM|<Ka5%zL6;z)!-pS7p4l1Hl)jBUf=z?tdWTHb_n7EYvPG-ix;8|n_gh<1gHB#gs
z-J@14v;4k-L*7hjj0JfrVy-76x!Jb%r5%gg_5HlEaxW2^{GL1VeZg%EbQchas4(K+
zOwas+alytNPZrCiqtd0^z%vMQRGvo4Wo8g&22rXX+V*v`B`!aHkGxD1`iRVb=&1T3
zHc>TMp>p<9MoaS|k=TkwDd+opdU_gpa9dRck*4?Dg5g6{`=V_0ft?}d`w16!TVYtz
zg%tE&n0?A&m^C?yri2(=$Dz7Tr~6)zS>vW}plDS~gFd^(--Plh2;(Q-R+nM?oOJ*G
zd!}`59Sg*@7teQk@#5vpMPrzLOH3#O^su`*LnIZcMNMwMQ@#KabOl8C5Gj&2=X10U
zW?z8DTJ9k5;9Z3kp`ky%oz_0j)?Qy8g(j*hn{?(5+$CypDe>RceTrat0!WyL@9Bvu
zD=DeXKEnZNRX7Y<3MiwgDYP6P>#Q07!dF`)5^i4XXwpnH5Tm=<F|-flcHd-a?^un>
zTYvp|Ju}=dVL|Y6iT_lll3q@HI+_NTZ(bGZl+@r`zhcH~?e&uwJgg1yy6^wOYuV$4
z+1p~#zHxhBwC*UH{0b9a%3@FU$%~)Mv;9IG=zjJ=%|K3NT34>-M1`WCkh?X`0NX)_
zrZ*lm)aV3wvaN?bdFcvSgO0WSQ}IF9F42-e3=Z;me{=w!Z3Sg;{Ff@Ovcvjn`ek9*
zb95Xm51JTzhk)TZ7E(l}P}1dI-yH>#tLCY)PzJ*x-I=ESJ>^N@1<f2>s3g{ZgTEw*
zw6rFn@iZm9!AoI-N-zo8GquSu>I>y!8gB;9USD4<trCy1@L(YA{bZ}xTQ0UM(fKV7
zS_fk+9)FewCwp+xCjD9`tK|BK?E@9F?zphNo#$wC+PofPdwy#&h?qf5V|T~g6@xt!
zhnyU9!xbrg_vpGkWfO11ThKljOd!$FZ_*f=7++Kzu(6&oRmp9;NSCLs&TuI?SkC2U
zir-qx{EERjnZTdD=P%L);nR61%+y8Ao8w0vQa0UH(`id&E&oTe%tp@TPU8<d+kSUc
zV5DdF8qLif!>^ocbx$YZ^K1(jz6Ps}UvDg!Vie}AYyi4lQkmUI4i{Pk)kgHxKN&QB
z#Jw)_T1eo9RQUlTCG%!#^+AI9e%C7gV#XbZQ5Ca9q-FE_*sO<0e?<36XkWJp-1v$&
z-4T>lXd?Ot?0HUTlfEkRPz=1#0_4+fdq3t`g^?;nn$n-No|~4m98V#QX#-h1q_hts
z1NQveHero@G5l%Q0NHz+!7ugh-8*Qaa_}1$3)g}fWZrs`vQx_{jf_IV%;D((LRMN`
zcTEo^>$Z>-rYZSpZ@XW`*0-gq?~e5@aUHVsvT*wAS(N|6yK&N+H*bJ{k6OeiUws-6
zMIkAZgEG;NsExsY{dB497h-k*j+Fxwca7ys=LV1`=Dgk6!7i6W>obw1`J+uXqj(nW
zY2UINoZx+pcf=SoU~1`&Y@P=~K=WgdsU&8FXKCf)WL|!WoVl&*7Rb}!s9zTAZ=`;<
zY-v;=(&HpRHs8Xw^pkxG7vA{>VO5u=aN83a{?|2z(jg&_#lRVywVTti!>*2pWeY3}
zolZ&3&(;bj^KiLYl}q(W)K=ZCv)MG=oqLwve|e(Qe0}QCZ%!7jAG2ewcsPmC)~$T!
z>OTz>Epi0R&FLMwjz^CiImb5ttN;rPJUr!+>ZBV8rLOH7T3B#5@@UswI)qPuy=>HD
zTD59Y_(k`**FxvbAR>?#&_5E!b%#TgQ)9EgOZ8Y)GZek~Ax2T3$6J@iIwdQo1mPQQ
z9nFOEUlt^#EVO;n)Xa5U^#W>Gil*WYi=PesiO7|}-3t5K0i1oO0KOn9Oo9Ry_TTcG
z>=&VrIXC7<42}_*VVDm>;_2@NJ$+eO@h>D`1%BcK`@|NnHQt;!l3DpM#r>rLih5cz
zVItLYkU_+*Eo;w(87gh$3x4eRK32bf)7cTAbGWV>_*6s=73;m(*pf`fBSKvT&Y5rN
zk@6@gMdM*7D2<p8hTeT8H^vvYy5blIuwIR1pQR)oj%%2h_({!FF}hl;bLh?E_98Ls
z!5K|+_hhL>bOHVyx;^K(sji=9J|VP8*nf6gAthFhTbRz5li}LL;-2tkgxHn~ZOyM=
zmpv^9$mk$|tokvg*oww+a?M@I$})xa{cxFIfD!vhb>fAHEFunV`aJD7wSis$3$)t6
z3750XdujCE46AUV-pnX{)f#qL5vs5PpGWV#<}=||wa_6>%H8;UIvusXk>rGb=&f&c
zF^Rk7f3^82BjAiv>9Il&dY4EQIMjLB<ZdzD`hI+T<^yNdEJ|#aopX0{fyUC8Cqcx#
ze2SV(1xU`;RYPsQyU}9P$*Ao}QZwau>B6=o7wTl%QVC1Hw`XVJ{!`XW!PLGewD{fZ
zk|ZHjrn|9`$9SSc5*v|NAFy028F+!SjJuzy<`!Uc|91IzY%YO#|9UtSM3Lsb8qj<U
z5wRP6_S&JV+p;zaSi;VzJLmZgb|IR#QmxK;$YCzK-ci+z`o_+$8YE(E=N29bVXOp&
zC4jN(YaCba>DPJ+58;b0HV{ZbL8lOqx&NEl9}(Wvxus#3?DZSNSZ8$H?;$LY!%=X<
z=yyT+Xsj98(lhGh4URfTV__9zp&6O9_YocXN$rHXD>ZYQ;NBuRw$tMpfx~6!?|~vp
z9D#l2pSRP>eRw$5+q>Bx2xi9LLi^6Yf+a2uruxo*h9-*qHIvE~)LyR1m|A7$3sV<4
zL$0){(Ma_`V!)44NoZnPX-Q2Ip1SrD4xJrD%Ok3cU$_kD?UPBm^Xjg)4z&kHBis3j
zR{G3<7EPE+0VuWuIe5q!bZt>V7c{fqhc@Nrq&d<I6EQgElE~PrRKbd2z~}vkSFKj4
zrh&vp2_J0gwX@>Cv^oD&g<XwUR^LJK(l>_>X4t;8akEj2%e<QlOmyob2FXlb)>ZaP
zA(WlsV@fFg`jVc5eNUwBR|ad9Orbm7j{j^jl~0@jwR}0D1|oi6;H}S=Cfb8<u)wFP
z)dZ2RX$PvE1z>DwI@hdmOn_;wy#xEiqS!GOUX_z`TBdMoVLE2=49E~`YXQFbZ852J
zp9h4kwYcrOqf~Xtr;$6J#Uvi;?+p7-*xS$B0q5lEuC(0qWTHW}XlFiY*R=O@D%r|T
zsq~Q@J0_}se75BF2=ub=%#-xpR4m*ANU{%U8*HnM{yE#TV!z;t`b;?#T}YhS`q_gw
zamtH1b|Bf-*i${#t>LJAv#co|wZ6&s<nvi`Ie?(|6S3{r(sR=Oyfh%5k^B&?YJlB+
zMg?ZENSa7c@)JVu0o=sw_^;d<B-LLtB+<FT%eXyquFSVQ+#W;4ZZ?l@rC6!R6jHEp
zFu)iunZX~x?9(J7vxi^E;(w!BVpv%jC={v%>zbOkS76^<epPJwq`sh?g)EygYj(8W
zyTA3DUapl=a!-Zu_}xw)d|qBNHlS6Rz8y(hTkcL@@4>nvc?iaP%Y<8<EaLSS4sEY_
z@V%BazJvA}18`YwY;Mc-u520e54*IntYnLTa&~wuxnp&=)dry(jrOGhsjriWt?Fr4
zV`IlRzzR>r(FZrymrCV0%!*RvN-^Ei3DC_d_!>1WK5h4(>85fp!+_jx5w)UX_9w=+
zYzH*$ICtYB7=S<&evVK8etn39hoV=IA;leYJ4m}t6xlOZm=PZ<Bn$IFqYNrxbd>LV
zB)oA4Ygt^gCuBy2S?X%V-5{;O(sL-M2&}B4%G?4Qiw~QU%8AW5)0h1Q@(zd+YnB}q
zB3DoXpL~pE;5BU^MygGmtOkQhMV!hoyvd9yJ*(B%@FBz!p;2Nt^>4LU`|-D1M%j&0
z3z6NjfUomLM;5vb6^qht_4mM}3NOn4L8aKU1@Crp9hXulYfMhkm=g`VlL|&4LBtRV
z%!gUZ-RuhDj4<b(`g!u<@izI=FE(+YrO$#g7N-Y-h?{UOuALnmIDX_`lcQapL6fI8
z%rsz&XeJg%?`2Aol65>CST03VvtpQjX|f`-F2CX-7y~uW?^dk*IwPw>j+wdI-Ar)8
zRI;}MLFykV@g%T6Q&2(l_HG1y+wb9j9b~GSZD~4FxXX-<c*hCvgK}U`$+1emOWn==
zGmw~vC!Lj9n~$$T(MoRwZEBa2G4QbMQj5xRYLgYW5Cx?sE*9E~b0?9ayQ7xU4r@jM
z79A5kZ%mWN?6G|NyjW58*_ipzY8-rM)ps2b0%2<9v$DdITa?Vg)C!gJu2gV#tBnH6
z;%l^6E4kir&opZGgw$g;uGHg4g6M^xkb`~H+z+yqPQ;pD?GMW(KEK-3Rpsw4Yv2KI
zPyS%s#w&ANHKnBCbxX_a0)C%_Z3Dublft5`g6otLKglJyN$d^KQswUbqiOtv?|+0j
zKt+<1k{sb$(?#8gI9^IJ!%J3tNL1{dyh>Y=Wm`ie3wue^1%Wa!wg~9`Xd>SXgi+`9
zA*yxDK#8b2XNpa5MoYcHi_?l{ZT90@<YdR&)KBMqp<dw<_u_1eC&T*~39zY5PsNGy
zhdS3Ov%!iW)r#&&Nd>HSrw0;mO8P0ua4&GN>`5dZftAy8A}K_VB&J*1Tb|B#o!2@&
zZSOB?K@ZnU#kJgQ9*eh<h!GMN7DU3biJ?u?r9%=iOf=mB3a6HrB%>3F`3x_sXv2-v
zL;F|i75(?ecn`5QYXqRq5?&enZRXd11(~0-Nx+zZRt#hSTW{hM`xiwGvlD&NOjq&n
z``vD;n3UovnQ6<E*Ec1@C35uB>GNaFtsCKWw%i4DX#L2TvtFYhYi<<iwIkK;#uLBE
z77XlaJ4^@~#AWdMtUnW+F+3a^2whukc-KgwP1IP~)>T@JmgHI=kggrlyc@7pert;v
zQ(tR>0G*T}#$LKlHnn_IWiq4hY@*~=n|wnKdBQn*F{s&i(LtchoSw~OeXN{DG$!eL
zP>HX1(y;J@G1H8~My#to*Ou&Hc~_sxN8V0U_-OH;Bk$)9Y{znPi=F5@kZOMrYLcJt
z+K6&7`d3;-R*3x>#eScBv#@oLwDX?olg9Cm9>P704}*@Tn*vG`buueTr}4SCn)`>9
zjoJ|5?=I%;vXc>Gy%`lm;`gkmYn}t0DZ-n|%TJXm+WKi^!B#_ZhNo=bC20s_m$kv|
za}QL1WA4II7P|CWS0&eHLx2zgwFUz;dwvy2DZuIW3?mPiLnm|`j_o519hR-RTGw`~
z1RuL(>9eGFx=5_ZrPq*=jh6IgBRIju`C$53xE`Xl08}+hd%(gw$wXed*GO1Ki!Nk-
zFyF($CqWN+>3Q(`ii!cbr5T-RAnK&SXQU;*Zix7sbxU4hX6019BsQ;Z%cMu3-!#Ml
zjfs6BVp`op&O2XR-2;n+I~dhd>gFR&Bum%drf@d3-Rh|fu|xlBx{ZC*k`$TKb>3J_
zLbl?#*clB`GwdkAruz?M{qvbnfrLqVAsy;!E}5f+H`BOh`Nc||fCeBuECqD!`<il?
z?%@f-bmaxjIIG%+IjN#CaCpezHQ&w_t^Mln7J&G+mr4P;P4Z8zf)ALn5Ot=4qXjDd
z%0uZl6X*HIu#&p8a*VMc>;2LF+JB6Hdt<nlihcN^a&Cw5g+QGt?yUb2QT^)Msq-IY
z*h$_?DaA9O6%HhV@pvJHx3b)u?=KPqU&_2S*M65em-C=>{honoaYAaq`U;d`;mI^9
zTL<`jAcpDIDDEuY9rp5<=-CA2nJr9git3F7e|>sHeO1=WitGw?o3*Za8_xCl#;9=R
z;+)8cI6qAfei+ckt=--t|Gr1yZ-Um%@RE;!b(D)Vx$3)}v?+~5;9@Kbyr@0ee^(4G
zKFd%LdDy}|8b0JW=R1)Z5=aEOB7%377Wb-I*)&@ZUTFu>>#Pd_k(ta>-@kpP7<O3y
z>>S><GmagzJvau!A08ggXi1{B<Y}v@h{kx6_rL<7y3Kx_kN^(f;PyTlFth1?IxG&x
zjnG7%-$XeOj)E>=o<>T_|Lk(hkn(5^;?(0oGi2jQ<X(OK(yw*nUwZEPAFXQ$39qk2
zMoFvSkL9xdB^v)vfbE_$ReL9=7$>XP(4rwDVrs9i6I05s9PNUKlUz_kjJ|?CGlyK;
z%rP*+yyL*ivRiK5A#2~}GlXq1%xh84d3!#}@_2e|IcCuNdm`KNQH=nUaky%^K4U87
zN!xig-LV+R%dZxQM{sN#WH)85LCl^%)*KI%TwdXYJ33Rgyb-mi<ChP_z>Ooljm@(f
zED+y^{Tjy4?ryD6#W{_qa$-#_kH&!IMhBOiu`ygjG6e_M&WX{_O4nW-(<;EL9!+E6
zNs}KlrpA+9(?U)@3;Bg@|GeRYFe>4o-`Mx9m2)(l*)`BbFgP<xWI3~!gZoO{Q)09h
zQG5OXWrUAGVT&{{16=WUfTxb{*i_iD8P=SdzO#yICefj(C#(gU%xk6xDV4U<)6+s!
zEW|~zIsnMbP{wGKwnU7#@C+D>f)3Ss?%mBOKq%QZ`oAMn-!$&NQ;NT@YfWDA_vv3D
zjLrpi<5p?w#*;qys`b^YYxCJ#eWnNibuB79MEbLD%`M3xUlhi0(<5?ml}o{_s6jU{
zbPz-~$H~jy3vhHiE5oI{dF1;Gm7FTGrP<iNKqjZ!J_%}QsDGG~i;y(c#imM&$%?Ay
zo4rI#so7Orc7K_K7ZVHL3u5BVKSg)v4p18|RWe_iC_WfETe|fd8Fp9pY5%TohxdDf
z(*_awYd{(Uw~^TMD>xK&8TChBuNc`&>wAsCti5$F1fwnP+WX9!70SQgKJ3Hs+{Mv&
zQPSqa&bn9lmPbDN+g=F`vxzbZ|K<CI>z)*DYQ|JY<~0z|{@!b65vN{j9n{$mB%3oQ
zcU6C9KA~IsI2+kBpIq_}26tm<KV|TEf?3_&N@GF?53n73@*AHIGy!8g@v2#`ik;bJ
z^NVACbD)I^jz|Zc`v08iiHni032Hiw;%sVd>qZ)E2$l@nu=<Glt@_51@}iB0!I4mw
zB@|0j;okifHvszGM_e#ImwH36Af9($-b0NLqRcOdS5+`~oDcNZM7Q-|`)d2|wTt)n
z_KL)KbAmx7yl<BarYVWa0^NTzwSci$K+8mAVF_%6F&=A4R+CZpT`mAB@o{-IoygGC
zm4`@u%u;czv=5j-U0vHx2>9XkSx)Do1bOX~njo9(Qqa~~u~#Q2T=(;%JKmBwHqg+U
zSA8qNJrv_4nzD>qee$if?~aK;Q-0exHr-6(aI<eGMcgsXXtF)>i*Sxcr6_B9*$<Xk
z5oVxF4N0*OdDkj$+gSJlWuByB%S>A3nHlG3aZvm|<rnpno8n2>&)l)?^Kq|RSXl6t
z+-h*nmCU34{YEFJbgNp#KPs7qJDAo$#wC6t(XW%#pF^&HsQCY9*J8j!S3-p5a9$}m
P6y&y!!FBvq>%jj4uImUS

literal 0
HcmV?d00001

diff --git a/docs/search.json b/docs/search.json
index 1ef9fa2..c33968b 100644
--- a/docs/search.json
+++ b/docs/search.json
@@ -6,6 +6,20 @@
     "section": "",
     "text": "%cd ..\n\nc:\\Users\\s2558406\\Documents\\Repos\\cmr-experiments\n\n\nc:\\Users\\s2558406\\Documents\\Repos\\cmr-experiments\\venv\\Lib\\site-packages\\IPython\\core\\magics\\osm.py:417: UserWarning: This is now an optional IPython functionality, setting dhist requires you to install the `pickleshare` library.\n  self.shell.db['dhist'] = compress_dhist(dhist)[-100:]\n\n\n\nfrom utils.edipo.data.transforms import CineNetDataTransform\nfrom utils.edipo.data.mri_data import RawDataSample\nfrom utils.edipo.models.crnn import CRNN\nfrom utils import *\nfrom pathlib import Path\nfrom torch.utils.data import DataLoader, random_split\nimport torch\nimport numpy as np\n\n\n# ONLY RUN when new dataset_cache run on Linux\nimport pathlib, pickle\nwith open(\"dataset_cache.pkl\", \"rb\") as f:\n    with set_posix_windows():\n        dataset_cache = pickle.load(f)\n    \n    new_cache = {\n        pathlib.WindowsPath(r\"M:\\data\\CMRxRecon\\SingleCoil\\Cine\\TrainingSet\"):\n    [\n        RawDataSample(r.fname.replace(\"/home/s2558406/RDS\", \"M:\").replace(\"/\", \"\\\\\"), r.slice_ind, r.metadata) for r in dataset_cache[\n            pathlib.WindowsPath('/home/s2558406/RDS/data/CMRxRecon/SingleCoil/Cine/TrainingSet')\n        ]\n    ]}\n\nwith open(\"dataset_cache_windows.pkl\", \"wb\") as f:\n    pickle.dump(new_cache, f)\n    \n\n\ndataset = DeepinvSliceDataset(\n    Path(r\"M:\\data\\CMRxRecon\"), \n    transform=CineNetDataTransform(time_window=12, apply_mask=True, normalize=False), \n    set_name=\"TrainingSet\",\n    acc_folders=[\"FullSample\"],\n    mask_folder=\"TimeVaryingGaussianMask16\",#\"AccFactor08\",\n    dataset_cache_file=\"dataset_cache_windows.pkl\"\n    )\n\nUsing dataset cache file\n\n\n\ntrain_dataset, test_dataset = random_split(dataset, (0.8, 0.2))\n\n\ntrain_dataloader = DataLoader(dataset=train_dataset, batch_size=1, shuffle=False)\ntest_dataloader  = DataLoader(dataset= test_dataset, batch_size=1, shuffle=False)\n\n\nx, y, mask = next(iter(train_dataloader))\n\n\nphysics = DynamicMRI((1, 2, 12, 512, 256))\ny2 = physics(x, mask=mask)\n\n\nx_hat = physics.A_adjoint(y2, mask=mask)\n\n\nmodel = ArtifactRemovalCRNN(\n    CRNN(num_cascades=2)\n).to(\"cpu\")\n\n\nphysics.set_mask(mask)\nx_recon = model(y, physics)\n\n\nplot_gif([x, y, y2, x_hat, mask], titles=[\"x\", \"y\", \"y2\", \"x_hat\", \"mask\"], display=True)\n\n\n\n\n\n\n\n\n  \n  \n    \n    \n      \n          \n      \n        \n      \n          \n      \n          \n      \n          \n      \n          \n      \n          \n      \n          \n      \n          \n    \n    \n      \n      Once\n      \n      Loop\n      \n      Reflect\n    \n  \n\n\n\n\n\n\n\n\n\n\nimport deepinv as dinv\n\n\nclass Undersampling(dinv.physics.DecomposablePhysics):\n    def __init__(self, mask, device=\"cpu\", **kwargs):\n        super().__init__(**kwargs)\n        self.mask = torch.nn.Parameter(mask.to(device), requires_grad=False)\n\n\nloss = YamanSplittingLoss(split_ratio=0.9)\n\nmask2 = loss.subsample_mask(loss.rng, physics.mask.data.detach().clone())\n\ninp = Undersampling(mask2, device=y.device)\n\ninp2 = Undersampling(mask - mask2, device=y.device)\n\nphysics1 = DynamicMRI(physics.img_size)\nphysics1.set_mask(mask2)\nphysics2 = DynamicMRI(physics.img_size)\nphysics2.set_mask(mask - mask2)\n\n# divide measurements\ny1 = inp.A(y)\ny2 = inp2.A(y)\n\n\nplot_videos([\n    mask, mask2, mask-mask2, \n    y, y1, y2, \n    physics.A_adjoint(y, mask=mask), physics1.A_adjoint(y1, mask=mask2), physics2.A_adjoint(y2, mask=mask-mask2)\n], display=True)\n\n\n\n\n\n\n\n\n  \n  \n    \n    \n      \n          \n      \n        \n      \n          \n      \n          \n      \n          \n      \n          \n      \n          \n      \n          \n      \n          \n    \n    \n      \n      Once\n      \n      Loop\n      \n      Reflect"
   },
+  {
+    "objectID": "demo_splitting_loss.html",
+    "href": "demo_splitting_loss.html",
+    "title": "Self-supervised learning with measurement splitting",
+    "section": "",
+    "text": "We demonstrate self-supervised learning with measurement splitting, to train a denoiser network on the MNIST dataset.\nMeasurement splitting constructs a ground-truth free loss \\(\\frac{m}{m_2}\\| y_2 - A_2 \\inversef{y_1}{A_1}\\|^2\\) by splitting the measurement and the forward operator using a randomly generated mask.\nSee :class:deepinv.loss.SplittingLoss for full details.\nimport deepinv as dinv\nfrom torch.utils.data import DataLoader\nimport torch\nfrom torchvision import transforms, datasets\nfrom deepinv.models.utils import get_weights_url\n\nc:\\Users\\s2558406\\Documents\\Repos\\deepinv\\venv\\Lib\\site-packages\\tqdm\\auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n  from .autonotebook import tqdm as notebook_tqdm\ntorch.manual_seed(0)\ndevice = dinv.utils.get_freer_gpu() if torch.cuda.is_available() else \"cpu\""
+  },
+  {
+    "objectID": "demo_splitting_loss.html#train-and-test-network",
+    "href": "demo_splitting_loss.html#train-and-test-network",
+    "title": "Self-supervised learning with measurement splitting",
+    "section": "Train and test network",
+    "text": "Train and test network\n\ntrainer = dinv.Trainer(\n    model=model,\n    physics=physics,\n    epochs=1,\n    losses=loss,\n    optimizer=optimizer,\n    device=device,\n    train_dataloader=train_dataloader,\n    plot_images=False,\n    save_path=None,\n    verbose=True,\n    show_progress_bar=False,\n    wandb_vis=False,\n)\n\nmodel = trainer.train()\n\nThe model has 444737 trainable parameters\nTrain epoch 0: TotalLoss=0.017, PSNR=11.59\n\n\nTest and visualise the model outputs using a small test set. We set the output to average over 50 iterations of random mask realisations. The trained model improves on the no-learning reconstruction by ~3dB.\n\ntrainer.plot_images = True\nmodel.MC_samples = 50\ntrainer.test(test_dataloader)\n\n\n\n\nTest PSNR: No learning rec.: 19.356+-1.523 | Model: 23.150+-1.996. \n\n\n(23.150262069702148,\n 1.9961603806415074,\n 19.355849266052246,\n 1.5233685706910973)\n\n\nSince this is a denoising example, above, we have set eval_split_output to True (see :class:deepinv.loss.SplittingLoss for details). Alternatively, we get worse results when we set eval_split_output to False:\n\nmodel.eval_split_output = False\ntrainer.test(test_dataloader)\n\n\n\n\nTest PSNR: No learning rec.: 19.356+-1.523 | Model: 14.441+-1.520. \n\n\n(14.44115686416626, 1.5199918435073472, 19.355849266052246, 1.5233685706910973)\n\n\nFurthermore, we can disable measurement splitting at evaluation altogether by setting eval_split_input to False (this is done in SSDU):\n\nmodel.eval_split_input = False\ntrainer.test(test_dataloader)\n\n\n\n\nTest PSNR: No learning rec.: 19.356+-1.523 | Model: 9.650+-1.670. \n\n\n(9.650477170944214, 1.6700495788637173, 19.355849266052246, 1.5233685706910973)"
+  },
   {
     "objectID": "demo_ei.html",
     "href": "demo_ei.html",
@@ -14,17 +28,17 @@
     "text": "Train a neural network to solve an image inpainting inverse problem, using perspective-EI and the deepinv library.\n\nimport deepinv as dinv\n\n\n\ntorch imports\nimport torch\nfrom torch.utils.data import DataLoader, random_split\nfrom torchvision.datasets import ImageFolder\nfrom torchvision.transforms import Compose, ToTensor, CenterCrop, Resize\nfrom torchvision.datasets.utils import download_and_extract_archive\n\n\nDefine inpainting experiment to reconstruct images from images masked with a random mask:\n\nphysics = dinv.physics.Inpainting((3, 256, 256), mask=0.6, device=\"cpu\")\n\nLoad Urban100 dataset of natural urban scenes:\n\n\nDownload dataset from HuggingFace\ndownload_and_extract_archive(\n    \"https://huggingface.co/datasets/eugenesiow/Urban100/resolve/main/data/Urban100_HR.tar.gz?download=true\",\n    \"Urban100\",\n    filename=\"Urban100_HR.tar.gz\",\n    md5=\"65d9d84a34b72c6f7ca1e26a12df1e4c\"\n)\n\n\n\ntrain_dataset, test_dataset = random_split(\n    ImageFolder(\"Urban100\", transform=Compose([ToTensor(), Resize(256), CenterCrop(256)])),\n    (0.8, 0.2)\n    )\n    \ntrain_dataloader, test_dataloader = DataLoader(train_dataset, shuffle=True), DataLoader(test_dataset)\n\nAs these scenes are imaged with a camera free to move and rotate in the world, we can impose perspective invariance on the unknown image set \\(x\\in X\\). Define measurement consistency and EI losses:\n\ntransform = dinv.transform.Homography(theta_max=10)\n\nlosses = [\n    dinv.loss.MCLoss(), \n    dinv.loss.EILoss(transform)\n]\n\nFor training, use a small UNet for the model with Adam optimizer:\n\nmodel = dinv.models.UNet(\n    in_channels=3, \n    out_channels=3,\n    scales=2,\n    circular_padding=True,\n    batch_norm=False\n)\n\noptimizer = torch.optim.Adam(model.parameters(), lr=1e-3, weight_decay=1e-8)\n\nTrain the model using deepinv’s Trainer:\n\nmodel = dinv.Trainer(\n    model=model,\n    physics=physics,\n    online_measurements=True,\n    train_dataloader=train_dataloader,\n    eval_dataloader=test_dataloader,\n    epochs=1,\n    losses=losses,\n    optimizer=optimizer,\n    verbose=True,\n    show_progress_bar=False,\n    save_path=None,\n    device=\"cpu\"\n).train()\n\nThe model has 444867 trainable parameters\nEval epoch 0: PSNR=10.078\nTrain epoch 0: MCLoss=0.002, EILoss=0.021, TotalLoss=0.023, PSNR=15.948\n\n\nShow results of a pretrained model trained using a larger UNet for 40 epochs:\n\n\nLoad pretrained model from HuggingFace\nmodel = dinv.models.UNet(\n    in_channels=3, \n    out_channels=3,\n    scales=3,\n    circular_padding=True,\n    batch_norm=False\n)\n\nckpt = torch.hub.load_state_dict_from_url(\n    dinv.models.utils.get_weights_url(\"ei\", \"Urban100_inpainting_homography_model.pth\"),\n    map_location=\"cpu\",\n)\n\nmodel.load_state_dict(ckpt[\"state_dict\"])\n\n\n&lt;All keys matched successfully&gt;\n\n\n\nx, _ = next(iter(train_dataloader))\ny = physics(x)\nx_hat = model(y)\n\ndinv.utils.plot([x, y, x_hat], [\"x\", \"y\", \"reconstruction\"])"
   },
   {
-    "objectID": "demo_splitting_loss.html",
-    "href": "demo_splitting_loss.html",
+    "objectID": "demo_splitting_loss_tomography.html",
+    "href": "demo_splitting_loss_tomography.html",
     "title": "Self-supervised learning with measurement splitting",
     "section": "",
-    "text": "We demonstrate self-supervised learning with measurement splitting, to train a denoiser network on the MNIST dataset.\nMeasurement splitting constructs a ground-truth free loss \\(\\frac{m}{m_2}\\| y_2 - A_2 \\inversef{y_1}{A_1}\\|^2\\) by splitting the measurement and the forward operator using a randomly generated mask.\nSee :class:deepinv.loss.SplittingLoss for full details.\nimport deepinv as dinv\nfrom torch.utils.data import DataLoader\nimport torch\nfrom torchvision import transforms, datasets\nfrom deepinv.models.utils import get_weights_url\n\nc:\\Users\\s2558406\\Documents\\Repos\\deepinv\\venv\\Lib\\site-packages\\tqdm\\auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n  from .autonotebook import tqdm as notebook_tqdm\ntorch.manual_seed(0)\ndevice = dinv.utils.get_freer_gpu() if torch.cuda.is_available() else \"cpu\""
+    "text": "We demonstrate self-supervised learning with measurement splitting, to train a denoiser network on the MNIST dataset. The physics here is noisy computed tomography, as is the case in Noise2Inverse. Note this example can also be easily applied to undersampled multicoil MRI as is the case in SSDU.\nMeasurement splitting constructs a ground-truth free loss \\(\\frac{m}{m_2}\\| y_2 - A_2 \\inversef{y_1}{A_1}\\|^2\\) by splitting the measurement and the forward operator using a randomly generated mask.\nSee :class:deepinv.loss.SplittingLoss for full details.\nimport deepinv as dinv\nfrom torch.utils.data import DataLoader\nimport torch\nfrom torchvision import transforms, datasets\nfrom deepinv.models.utils import get_weights_url\n\nc:\\Users\\s2558406\\Documents\\Repos\\deepinv\\venv\\Lib\\site-packages\\tqdm\\auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n  from .autonotebook import tqdm as notebook_tqdm\ntorch.manual_seed(0)\ndevice = dinv.utils.get_freer_gpu() if torch.cuda.is_available() else \"cpu\""
   },
   {
-    "objectID": "demo_splitting_loss.html#train-and-test-network",
-    "href": "demo_splitting_loss.html#train-and-test-network",
+    "objectID": "demo_splitting_loss_tomography.html#train-and-test-network",
+    "href": "demo_splitting_loss_tomography.html#train-and-test-network",
     "title": "Self-supervised learning with measurement splitting",
     "section": "Train and test network",
-    "text": "Train and test network\n\ntrainer = dinv.Trainer(\n    model=model,\n    physics=physics,\n    epochs=1,\n    losses=loss,\n    optimizer=optimizer,\n    device=device,\n    train_dataloader=train_dataloader,\n    plot_images=False,\n    save_path=None,\n    verbose=True,\n    show_progress_bar=False,\n    wandb_vis=False,\n)\n\nmodel = trainer.train()\n\nThe model has 444737 trainable parameters\nTrain epoch 0: TotalLoss=0.017, PSNR=11.59\n\n\nTest and visualise the model outputs using a small test set. We set the output to average over 50 iterations of random mask realisations. The trained model improves on the no-learning reconstruction by ~3dB.\n\ntrainer.plot_images = True\nmodel.MC_samples = 50\ntrainer.test(test_dataloader)\n\n\n\n\nTest PSNR: No learning rec.: 19.356+-1.523 | Model: 23.150+-1.996. \n\n\n(23.150262069702148,\n 1.9961603806415074,\n 19.355849266052246,\n 1.5233685706910973)\n\n\nSince this is a denoising example, above, we have set eval_split_output to True (see :class:deepinv.loss.SplittingLoss for details). Alternatively, we get worse results when we set eval_split_output to False:\n\nmodel.eval_split_output = False\ntrainer.test(test_dataloader)\n\n\n\n\nTest PSNR: No learning rec.: 19.356+-1.523 | Model: 14.441+-1.520. \n\n\n(14.44115686416626, 1.5199918435073472, 19.355849266052246, 1.5233685706910973)\n\n\nFurthermore, we can disable measurement splitting at evaluation altogether by setting eval_split_input to False (this is done in SSDU):\n\nmodel.eval_split_input = False\ntrainer.test(test_dataloader)\n\n\n\n\nTest PSNR: No learning rec.: 19.356+-1.523 | Model: 9.650+-1.670. \n\n\n(9.650477170944214, 1.6700495788637173, 19.355849266052246, 1.5233685706910973)"
+    "text": "Train and test network\n\ntrainer = dinv.Trainer(\n    model=model,\n    physics=physics,\n    epochs=3,\n    losses=loss,\n    optimizer=optimizer,\n    device=device,\n    train_dataloader=train_dataloader,\n    plot_images=False,\n    save_path=None,\n    verbose=True,\n    show_progress_bar=False,\n    wandb_vis=False,\n)\n\nmodel = trainer.train()\n\nThe model has 444737 trainable parameters\nTrain epoch 0: TotalLoss=0.032, PSNR=29.155\nTrain epoch 1: TotalLoss=0.035, PSNR=28.895\nTrain epoch 2: TotalLoss=0.035, PSNR=28.837\n\n\nTest and visualise the model outputs using a small test set. We set the output to average over 5 iterations of random mask realisations. The trained model improves on the no-learning reconstruction by ~7dB.\n\ntrainer.plot_images = True\ntrainer.test(test_dataloader, pinv=True)\n\n\n\n\nTest PSNR: No learning rec.: 24.549+-1.052 | Model: 31.911+-2.220. \n\n\n(31.911066627502443, 2.219884675922773, 24.548791694641114, 1.0523162766060832)\n\n\nDemonstrate the effect of not averaging over multiple realisations of the splitting mask at evaluation time, by setting eval_n_samples=1. We have a worse performance:\n\nmodel.eval_n_samples = 1\ntrainer.test(test_dataloader, pinv=True)\n\n\n\n\nTest PSNR: No learning rec.: 24.549+-1.052 | Model: 30.434+-2.487. \n\n\n(30.434445762634276, 2.486670644991658, 24.548791694641114, 1.0523162766060832)\n\n\nFurthermore, we can disable measurement splitting at evaluation altogether by setting eval_split_input to False (this is done in SSDU). This generally is worse than MC averaging:\n\nmodel.eval_split_input = False\ntrainer.test(test_dataloader, pinv=True)\n\n\n\n\nTest PSNR: No learning rec.: 24.549+-1.052 | Model: 31.003+-2.107. \n\n\n(31.002875900268556, 2.106733038650352, 24.548791694641114, 1.0523162766060832)"
   }
 ]
\ No newline at end of file