add tests for losses (CCE currently failing)

fancompute · Jan 9, 2019 · a2f2a2c · a2f2a2c
1 parent 117b0bf
commit a2f2a2c
Show file tree

Hide file tree

Showing 3 changed files with 51 additions and 1 deletion.
diff --git a/neuroptica/losses.py b/neuroptica/losses.py
@@ -43,7  43,7 @@ class CategoricalCrossEntropy(Loss):
     @staticmethod
     def L(X: np.ndarray, T: np.ndarray) -> np.ndarray:
         X_softmax = np.exp(X) / np.sum(np.exp(X), axis=0)
-        X_clip = np.clip(X_softmax, 1e-7, 1 - 1e-7)
         X_clip = np.clip(X_softmax, 1e-9, 1 - 1e-9)
         return -np.sum(T * np.log(X_clip), axis=0)
 
     @staticmethod
@@ -54,3  54,4 @@ def dL(X: np.ndarray, T: np.ndarray) -> np.ndarray:
             return -T / X_clip
         else:
             return np.conj(X - T)
             # return X - T
diff --git a/tests/test_losses.py b/tests/test_losses.py
@@ -0,0  1,47 @@
 import unittest
 
 from neuroptica.layers import Activation, ClementsLayer
 from neuroptica.losses import CategoricalCrossEntropy, MeanSquaredError
 from neuroptica.models import Sequential
 from neuroptica.nonlinearities import *
 from neuroptica.optimizers import Optimizer
 from tests.base import NeuropticaTest
 from tests.test_models import TestModels
 
 
 class TestLosses(NeuropticaTest):
     '''Tests for model losses'''
 
     def test_loss_gradients(self):
         for N in [9, 10]:
 
             losses = [MeanSquaredError, CategoricalCrossEntropy]
 
             for loss in losses:
 
                 print("Testing loss {}".format(loss))
 
                 batch_size = 6
                 n_samples = batch_size * 4
 
                 X_all = (2 * np.random.rand(N * n_samples) - 1).reshape((N, n_samples))
                 Y_all = np.abs(X_all)
 
                 # Make a single-layer model
                 model = Sequential([ClementsLayer(N),
                                     Activation(Abs(N))
                                     ])
 
                 for X, Y in Optimizer.make_batches(X_all, Y_all, batch_size):
                     # Propagate the data forward
                     Y_hat = model.forward_pass(X)
                     d_loss = loss.dL(Y_hat, Y)
 
                     # Compute the backpropagated signals for the model
                     gradients = model.backward_pass(d_loss)
 
                     TestModels.verify_model_gradients(model, X, Y, loss.L, gradients, epsilon=1e-6)
 
 
 if __name__ == "__main__":
     unittest.main()
diff --git a/tests/test_nonlinearities.py b/tests/test_nonlinearities.py
@@ -41,6  41,8 @@ def test_OpticalMesh_adjoint_optimize(self):
 
             # nonlinearities that may be applied to complex outpus
             nonlinearities_complex = [Abs(N, mode="full"),
                                       Abs(N, mode="condensed"),
                                       Abs(N, mode="polar"),
                                       SoftMax(N),
                                       AbsSquared(N),
                                       ElectroOpticActivation(N, **eo_settings),