ex_05_job_lib.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# A Guided Tour of Ray Core: JobLib"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "[*Distributed scikit-learn*](https://docs.ray.io/en/latest/joblib.html) provides a drop-in replacement to parallelize the [`JobLib`](https://joblib.readthedocs.io/en/latest/) backend for [`scikit-learn`](https://scikit-learn.org/stable/)\n",
    "\n",
    "\n",
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, let's start Ray…"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import logging\n",
    "import ray\n",
    "\n",
    "ray.init(\n",
    "    ignore_reinit_error=True,\n",
    "    logging_level=logging.ERROR,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## JobLib example"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Set up for this example..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from ray.util.joblib import register_ray\n",
    "from sklearn.datasets import load_digits\n",
    "from sklearn.model_selection import RandomizedSearchCV\n",
    "from sklearn.svm import SVC\n",
    "import numpy as np\n",
    "import joblib"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, let's register Ray as the parallelized [*joblib*](https://scikit-learn.org/stable/modules/generated/sklearn.utils.parallel_backend.html) backend for `scikit-learn`, using  Ray actors instead of local processes.\n",
    "This makes it easy to scale existing applications running on a single node to running on a cluster.\n",
    "\n",
    "See: <https://docs.ray.io/en/master/joblib.html>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "\n",
    "register_ray()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Next, load a copy of the UCI machine learning data repository's hand-written *digits* dataset.\n",
    "See: <https://scikit-learn.org/stable/modules/generated/sklearn.datasets.load_digits.html>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "\n",
    "digits = load_digits()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We'll define the hyper-parameter space for training a *support vector machines* model:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "\n",
    "param_space = {\n",
    "    \"C\": np.logspace(-6, 6, 30),\n",
    "    \"gamma\": np.logspace(-8, 8, 30),\n",
    "    \"tol\": np.logspace(-4, -1, 30),\n",
    "    \"class_weight\": [None, \"balanced\"],\n",
    "}\n",
    "\n",
    "model = SVC(kernel=\"rbf\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then use a randomized search to optimize these hyper-parameters. See: <https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.RandomizedSearchCV.html>\n",
    "\n",
    "We'll use 10 cross-validation splits and 50 iterations, which will result in a total of 500 \"fits\". This is enough to illustrate the `joblib` being parallelized, although in practice you'd probably use more iterations."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "\n",
    "clf = RandomizedSearchCV(model, param_space, cv=10, n_iter=50, verbose=True)\n",
    "clf"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Run the cross-validation fits (i.e., the random search for hyper-parameter optimization) using Ray to parallelize the backend processes.\n",
    "\n",
    "NB: **While this runs, check out the performance metrics on the Ray dashboard**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true,
    "tags": []
   },
   "outputs": [],
   "source": [
    "%%time\n",
    "\n",
    "with joblib.parallel_backend(\"ray\"):\n",
    "    search = clf.fit(digits.data, digits.target)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "So far, what is the best set of hyper-parameters found?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "search.best_params_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, shutdown Ray"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "ray.shutdown()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}