Fix the bug of non-convergence when use SparseCategoricalCrossentropy

Wanglongzhi2001 · Wanglongzhi2001 · commit 5506f00906c7 · 2023-04-07T17:26:37.000+08:00
diff --git a/src/TensorFlowNET.Keras/Losses/SparseCategoricalCrossentropy.cs b/src/TensorFlowNET.Keras/Losses/SparseCategoricalCrossentropy.cs
@@ -4,17 +4,21 @@ namespace Tensorflow.Keras.Losses
 {
     public class SparseCategoricalCrossentropy : LossFunctionWrapper, ILossFunc
     {
+        private bool _from_logits = false;
         public SparseCategoricalCrossentropy(
             bool from_logits = false,
             string reduction = null,
             string name = null) :
-            base(reduction: reduction, name: name == null ? "sparse_categorical_crossentropy" : name){ }
+            base(reduction: reduction, name: name == null ? "sparse_categorical_crossentropy" : name)
+        {
+            _from_logits = from_logits;
+        }
 
         public override Tensor Apply(Tensor target, Tensor output, bool from_logits = false, int axis = -1)
         {
             target = tf.cast(target, dtype: TF_DataType.TF_INT64);
 
-            if (!from_logits)
+            if (!_from_logits)
             {
                 var epsilon = tf.constant(KerasApi.keras.backend.epsilon(), output.dtype);
                 output = tf.clip_by_value(output, epsilon, 1 - epsilon);

Original file line number	Diff line number	Diff line change
`@@ -4,17 +4,21 @@ namespace Tensorflow.Keras.Losses`
`4`	`4`	`{`
`5`	`5`	`public class SparseCategoricalCrossentropy : LossFunctionWrapper, ILossFunc`
`6`	`6`	`{`
	`7`	`+ private bool _from_logits = false;`
`7`	`8`	`public SparseCategoricalCrossentropy(`
`8`	`9`	`bool from_logits = false,`
`9`	`10`	`string reduction = null,`
`10`	`11`	`string name = null) :`
`11`		`- base(reduction: reduction, name: name == null ? "sparse_categorical_crossentropy" : name){ }`
	`12`	`+ base(reduction: reduction, name: name == null ? "sparse_categorical_crossentropy" : name)`
	`13`	`+ {`
	`14`	`+ _from_logits = from_logits;`
	`15`	`+ }`
`12`	`16`
`13`	`17`	`public override Tensor Apply(Tensor target, Tensor output, bool from_logits = false, int axis = -1)`
`14`	`18`	`{`
`15`	`19`	`target = tf.cast(target, dtype: TF_DataType.TF_INT64);`
`16`	`20`
`17`		`- if (!from_logits)`
	`21`	`+ if (!_from_logits)`
`18`	`22`	`{`
`19`	`23`	`var epsilon = tf.constant(KerasApi.keras.backend.epsilon(), output.dtype);`
`20`	`24`	`output = tf.clip_by_value(output, epsilon, 1 - epsilon);`