tests passing again.

src/mpol/losses.py

@@ -198,7 +198,8 @@ def neg_log_likelihood_avg(
     Returns
     -------
     :class:`torch.Tensor` of :class:`torch.double`
-        the :math:`\chi^2_\mathrm{r}`, summed over all dimensions of input array.
+        the average of the negative log likelihood, summed over all dimensions of
+        input array.
     """
     N = len(torch.ravel(data_vis))  # number of complex visibilities
     ll = log_likelihood(model_vis, data_vis, weight)
@@ -297,8 +298,7 @@ def reduced_chi_squared_gridded(
     Returns
     -------
     :class:`torch.Tensor` of :class:`torch.double`
-        the normalized negative log likelihood likelihood loss, summed over all input
-        values
+        the :math:`\chi^2_\mathrm{r}` value summed over all input dimensions
     """
     model_vis = griddedDataset(modelVisibilityCube)
 

src/mpol/training.py

@@ -2,7 +2,7 @@
 import numpy as np
 import torch
 
-from mpol.losses import TSV, TV_image, entropy, nll_gridded, sparsity
+from mpol.losses import TSV, TV_image, entropy, reduced_chi_squared_gridded, sparsity
 from mpol.plot import train_diagnostics_fig
 from mpol.utils import torch2npy
 
@@ -205,7 +205,7 @@ def loss_eval(self, vis, dataset, sky_cube=None):
             Value of loss function
         """
         # negative log-likelihood loss function
-        loss = nll_gridded(vis, dataset)
+        loss = reduced_chi_squared_gridded(vis, dataset)
 
         # regularizers
         if sky_cube is not None:

test/losses_test.py

@@ -2,7 +2,7 @@
 import pytest
 import torch
 
-from mpol import fourier, images, losses, utils
+from mpol import fourier, images, losses
 
 
 # create a fixture that returns nchan and an image
@@ -102,7 +102,7 @@ def test_nll_hermitian_pairs(loose_visibilities, mock_visibility_data):
     data = torch.tensor(data_re + 1.0j * data_im)
     weight = torch.tensor(weight)
 
-    log_like = losses.nll(loose_visibilities, data, weight)
+    log_like = losses.reduced_chi_squared(loose_visibilities, data, weight)
     print("loose nll", log_like)
 
     # calculate it with Hermitian pairs
@@ -116,7 +116,7 @@ def test_nll_hermitian_pairs(loose_visibilities, mock_visibility_data):
     data = torch.cat([data, torch.conj(data)], axis=1)
     weight = torch.cat([weight, weight], axis=1)
 
-    log_like = losses.nll(loose_visibilities, data, weight)
+    log_like = losses.reduced_chi_squared(loose_visibilities, data, weight)
     print("loose nll w/ Hermitian", log_like)
 
 
@@ -133,11 +133,11 @@ def test_nll_evaluation(
     data = torch.tensor(data_re + 1.0j * data_im)
     weight = torch.tensor(weight)
 
-    log_like = losses.nll(loose_visibilities, data, weight)
+    log_like = losses.reduced_chi_squared(loose_visibilities, data, weight)
     print("loose nll", log_like)
 
     # calculate the gridded log likelihood
-    log_like_gridded = losses.nll_gridded(gridded_visibilities, dataset)
+    log_like_gridded = losses.reduced_chi_squared_gridded(gridded_visibilities, dataset)
     print("gridded nll", log_like_gridded)
 
 
@@ -156,7 +156,7 @@ def test_nll_1D_zero():
     data_im = model_im
     data_vis = torch.complex(data_re, data_im)
 
-    loss = losses.nll(model_vis, data_vis, weights)
+    loss = losses.reduced_chi_squared(model_vis, data_vis, weights)
     assert loss.item() == 0.0
 
 
@@ -175,7 +175,7 @@ def test_nll_1D_random():
     data_im = torch.randn_like(weights)
     data_vis = torch.complex(data_re, data_im)
 
-    losses.nll(model_vis, data_vis, weights)
+    losses.reduced_chi_squared(model_vis, data_vis, weights)
 
 
 def test_nll_2D_zero():
@@ -195,7 +195,7 @@ def test_nll_2D_zero():
     data_im = model_im
     data_vis = torch.complex(data_re, data_im)
 
-    loss = losses.nll(model_vis, data_vis, weights)
+    loss = losses.reduced_chi_squared(model_vis, data_vis, weights)
     assert loss.item() == 0.0
 
 
@@ -215,7 +215,7 @@ def test_nll_2D_random():
     data_im = torch.randn_like(weights)
     data_vis = torch.complex(data_re, data_im)
 
-    losses.nll(model_vis, data_vis, weights)
+    losses.reduced_chi_squared(model_vis, data_vis, weights)
 
 
 def test_entropy_raise_error_negative():

test/train_test_test.py

@@ -122,7 +122,7 @@ def test_standalone_init_train(coords, dataset):
     rml.zero_grad()
 
     # calculate a loss
-    loss = losses.nll_gridded(vis, dataset)
+    loss = losses.reduced_chi_squared_gridded(vis, dataset)
 
     # calculate gradients of parameters
     loss.backward()
@@ -147,7 +147,7 @@ def test_standalone_train_loop(coords, dataset_cont, tmp_path):
         vis = rml()
 
         # calculate a loss
-        loss = losses.nll_gridded(vis, dataset_cont)
+        loss = losses.reduced_chi_squared_gridded(vis, dataset_cont)
 
         # calculate gradients of parameters
         loss.backward()
@@ -194,7 +194,7 @@ def test_tensorboard(coords, dataset_cont):
         vis = rml()
 
         # calculate a loss
-        loss = losses.nll_gridded(vis, dataset_cont)
+        loss = losses.reduced_chi_squared_gridded(vis, dataset_cont)
 
         writer.add_scalar("loss", loss.item(), i)