Merge pull request #28 from imdeepmind/test-scripts-for-loss-functions

Test scripts for loss functions

neuralpy/loss_functions/cross_entropy.py

@@ -41,7 +41,7 @@ def __init__(self, weight=None, ignore_index=-100, reduction='mean'):
         if reduction not in ["none", "mean", "sum"]:
             raise ValueError("Invalid reduction")
 
-        if not isinstance(ignore_index, int):
+        if isinstance(ignore_index, bool) or not isinstance(ignore_index, int):
             raise ValueError("Invalid ignore index")
 
         self.__weight = weight

neuralpy/loss_functions/mse.py

@@ -4,7 +4,6 @@
 
 # pylint: disable=too-few-public-methods
 
-
 class MSELoss:
     """
         Applies a Mean Squared Error loss function to the model.

tests/neuralpy/loss_functions/test_bce_loss.py

@@ -5,31 +5,24 @@
 import numpy as np
 import torch
 
-# Possible values that are invalid
-weights=["asd", 12, -.3]
-reductions=["asdas", "", 12, 6.3]
-pos_weights=["asd", 12, -.3]
-
 @pytest.mark.parametrize(
 	"weight, reduction, pos_weight", 
-	[(weight, reduction, pos_weight) for weight in weights
-								     for reduction in reductions
-								     for pos_weight in pos_weights]
+	[("invalid", 12, -.3),
+	(12, 12, -.3),
+	([1.0, 1.0, 1.0], "sump", -.3),
+	([1.0, 1.0, 1.0], 3, -.3),
+	([1.0, 1.0, 1.0], "sum", -.3),
+	([1.0, 1.0, 1.0], "sum", "invalid")]
 )
 def test_bce_should_throw_value_error(weight, reduction, pos_weight):
     with pytest.raises(ValueError) as ex:
         x = BCELoss(weight=weight, reduction=reduction, pos_weight=pos_weight)
 
-# Possible values that are valid
-weights=[[1.0, 1.0, 1.0], [2.0, 1.0, 2.0], np.ones([3])]
-reductions=["mean"]
-pos_weights=[[1.0, 1.0, 1.0], [2.0, 1.0, 2.0], np.ones([3])]
-
 @pytest.mark.parametrize(
 	"weight, reduction, pos_weight", 
-	[(weight, reduction, pos_weight) for weight in weights
-								     for reduction in reductions
-								     for pos_weight in pos_weights]
+	[([1.0, 1.0, 1.0], "mean", np.ones([3])),
+	(np.ones([3]), "mean", [2.0, 1.0, 2.0]),
+	]
 )
 def test_bce_get_layer_method(weight, reduction, pos_weight):
 	x = BCELoss(weight=weight, reduction=reduction, pos_weight=pos_weight)

tests/neuralpy/loss_functions/test_cross_entropy_loss.py

@@ -5,31 +5,30 @@
 import numpy as np
 import torch
 
-# Possible values that are invalid
-weights=["asd", 12, -.3]
-reductions=["asdas", "", 12, 6.3]
-ignore_indexes=["asd", False, "", 2.36]
-
 @pytest.mark.parametrize(
 	"weight, reduction, ignore_index", 
-	[(weight, reduction, ignore_index) for weight in weights
-								     for reduction in reductions
-								     for ignore_index in ignore_indexes]
+	[
+		("invalid", "invalid", "invalid"),
+		(12, "invalid", "invalid"),
+		(np.ones([3]), "invalid", "invalid"),
+		(np.ones([3]), 12, "invalid"),
+		(np.ones([3]), "mean", "mean"),
+		(np.ones([3]), "sum", "sum"),
+		(np.ones([3]), "none", "asd"),
+		(np.ones([3]), "none", False)
+	]
 )
 def test_cce_should_throw_value_error(weight, reduction, ignore_index):
     with pytest.raises(ValueError) as ex:
         x = CrossEntropyLoss(weight=weight, reduction=reduction, ignore_index=ignore_index)
 
-# Possible values that are valid
-weights=[[1.0, 1.0, 1.0], [2.0, 1.0, 2.0], np.ones([3])]
-reductions=["mean"]
-ignore_indexes=[1, -100, 5]
-
 @pytest.mark.parametrize(
 	"weight, reduction, ignore_index", 
-	[(weight, reduction, ignore_index) for weight in weights
-								     for reduction in reductions
-								     for ignore_index in ignore_indexes]
+	[
+		([2.0, 1.0, 2.0], "mean", -100),
+		(np.ones([3]), "sum", 1),
+		(np.ones([3]), "none", 1),
+	]
 )
 def test_cce_get_layer_method(weight, reduction, ignore_index):
 	x = CrossEntropyLoss(weight=weight, reduction=reduction, ignore_index=ignore_index)

tests/neuralpy/loss_functions/test_mse.py

@@ -5,23 +5,17 @@
 import numpy as np
 import torch
 
-# Possible values that are invalid
-reductions=["asdas", "", 12, 6.3]
-
 @pytest.mark.parametrize(
 	"reduction", 
-	[(reduction) for reduction in reductions]
+	[("invalid", "", 12, 6.3)]
 )
 def test_cce_should_throw_value_error(reduction):
     with pytest.raises(ValueError) as ex:
         x = MSELoss(reduction=reduction)
 
-# Possible values that are valid
-reductions=["mean"]
-
 @pytest.mark.parametrize(
 	"reduction", 
-	[(reduction) for reduction in reductions]
+	[("none"), ("mean"), ("sum")]
 )
 def test_mse_get_layer_method(reduction):
 	x = MSELoss(reduction=reduction)