Document base Loss class

Author: fchollet

scripts/api_master.py

@@ -1190,7 +1190,7 @@
                         "keras.ops.inv",
                         "keras.ops.logdet",
                         "keras.ops.lstsq",
-                        "keras.ops.lu_factor",                        
+                        "keras.ops.lu_factor",
                         "keras.ops.norm",
                         "keras.ops.qr",
                         "keras.ops.solve",
@@ -1493,6 +1493,9 @@
             "path": "losses/",
             "title": "Losses",
             "toc": True,
+            "generate": [
+                "keras.losses.Loss",
+            ],
             "children": [
                 {
                     "path": "probabilistic_losses",
@@ -1526,6 +1529,7 @@
                         "keras.losses.Huber",
                         "keras.losses.LogCosh",
                         "keras.losses.Tversky",
+                        "keras.losses.Dice",
                         "keras.losses.mean_squared_error",
                         "keras.losses.mean_absolute_error",
                         "keras.losses.mean_absolute_percentage_error",
@@ -1534,6 +1538,7 @@
                         "keras.losses.huber",
                         "keras.losses.log_cosh",
                         "keras.losses.tversky",
+                        "keras.losses.dice",
                     ],
                 },
                 {

scripts/cv_api_master.py

@@ -532,7 +532,7 @@
             "generate": [
                 "keras_cv.models.RetinaNet",
                 "keras_cv.models.RetinaNet.from_preset",
-                "keras_cv.models.retinanet.PredictionHead"
+                "keras_cv.models.retinanet.PredictionHead",
             ],
         },
         {

scripts/master.py

@@ -50,7 +50,7 @@
             "path": "keras_cv/",
             "title": "KerasCV: Computer Vision Workflows",
         },
-                {
+        {
             "path": "keras_nlp/",
             "title": "KerasNLP: Natural Language Workflows",
         },

templates/api/losses/index.md

@@ -15,6 +15,11 @@ and they perform reduction by default when used in a standalone way (see details
 
 {{toc}}
 
+---
+
+## Base Loss API
+
+{{autogenerated}}
 
 ---
 
@@ -74,8 +79,9 @@ A loss is a callable with arguments `loss_fn(y_true, y_pred, sample_weight=None)
 By default, loss functions return one scalar loss value per input sample, e.g.
 
 ```
->>> keras.losses.mean_squared_error(tf.ones((2, 2,)), tf.zeros((2, 2)))
-<tf.Tensor: shape=(2,), dtype=float32, numpy=array([1., 1.], dtype=float32)>
+>>> from keras import ops
+>>> keras.losses.mean_squared_error(ops.ones((2, 2,)), ops.zeros((2, 2)))
+<Array: shape=(2,), dtype=float32, numpy=array([1., 1.], dtype=float32)>
 ```
 
 However, loss class instances feature a `reduction` constructor argument,
@@ -89,18 +95,18 @@ which defaults to `"sum_over_batch_size"` (i.e. average). Allowable values are
 
 ```
 >>> loss_fn = keras.losses.MeanSquaredError(reduction='sum_over_batch_size')
->>> loss_fn(tf.ones((2, 2,)), tf.zeros((2, 2)))
-<tf.Tensor: shape=(), dtype=float32, numpy=1.0>
+>>> loss_fn(ops.ones((2, 2,)), ops.zeros((2, 2)))
+<Array: shape=(), dtype=float32, numpy=1.0>
 ```
 ```
 >>> loss_fn = keras.losses.MeanSquaredError(reduction='sum')
->>> loss_fn(tf.ones((2, 2,)), tf.zeros((2, 2)))
-<tf.Tensor: shape=(), dtype=float32, numpy=2.0>
+>>> loss_fn(ops.ones((2, 2,)), ops.zeros((2, 2)))
+<Array: shape=(), dtype=float32, numpy=2.0>
 ```
 ```
 >>> loss_fn = keras.losses.MeanSquaredError(reduction='none')
->>> loss_fn(tf.ones((2, 2,)), tf.zeros((2, 2)))
-<tf.Tensor: shape=(2,), dtype=float32, numpy=array([1., 1.], dtype=float32)>
+>>> loss_fn(ops.ones((2, 2,)), ops.zeros((2, 2)))
+<Array: shape=(2,), dtype=float32, numpy=array([1., 1.], dtype=float32)>
 ```
 
 Note that this is an important difference between loss functions like `keras.losses.mean_squared_error`
@@ -109,13 +115,13 @@ does not perform reduction, but by default the class instance does.
 
 ```
 >>> loss_fn = keras.losses.mean_squared_error
->>> loss_fn(tf.ones((2, 2,)), tf.zeros((2, 2)))
-<tf.Tensor: shape=(2,), dtype=float32, numpy=array([1., 1.], dtype=float32)>
+>>> loss_fn(ops.ones((2, 2,)), ops.zeros((2, 2)))
+<Array: shape=(2,), dtype=float32, numpy=array([1., 1.], dtype=float32)>
 ```
 ```
 >>> loss_fn = keras.losses.MeanSquaredError()
->>> loss_fn(tf.ones((2, 2,)), tf.zeros((2, 2)))
-<tf.Tensor: shape=(), dtype=float32, numpy=1.0>
+>>> loss_fn(ops.ones((2, 2,)), ops.zeros((2, 2)))
+<Array: shape=(), dtype=float32, numpy=1.0>
 ```
 
 When using `fit()`, this difference is irrelevant since reduction is handled by the framework.

templates/api/optimizers/index.md

@@ -50,12 +50,11 @@ Check out [the learning rate schedule API documentation](/api/optimizers/learnin
 
 ---
 
-## Core Optimizer API
+## Base Optimizer API
 
 These methods and attributes are common to all Keras optimizers.
 
 {{autogenerated}}
 
 
 
-