Merge pull request #1484 from rstudio/retether-3.7.0

Retether Keras 3.7.0

Author: t-kalinowski

.tether/man/Loss.txt

@@ -5,10 +5,17 @@ class Loss(keras.src.saving.keras_saveable.KerasSaveable)
  |
  |  Loss base class.
  |
+ |  This is the class to subclass in order to create new custom losses.
+ |
  |  Args:
  |      reduction: Type of reduction to apply to the loss. In almost all cases
- |          this should be `"sum_over_batch_size"`.
- |          Supported options are `"sum"`, `"sum_over_batch_size"` or `None`.
+ |          this should be `"sum_over_batch_size"`. Supported options are
+ |          `"sum"`, `"sum_over_batch_size"`, `"mean"`,
+ |          `"mean_with_sample_weight"` or `None`. `"sum"` sums the loss,
+ |          `"sum_over_batch_size"` and `"mean"` sum the loss and divide by the
+ |          sample size, and `"mean_with_sample_weight"` sums the loss and
+ |          divides by the sum of the sample weights. `"none"` and `None`
+ |          perform no aggregation. Defaults to `"sum_over_batch_size"`.
  |      name: Optional name for the loss instance.
  |      dtype: The dtype of the loss's computations. Defaults to `None`, which
  |          means using `keras.backend.floatx()`. `keras.backend.floatx()` is a

.tether/man/activation_celu.txt

@@ -0,0 +1,19 @@
+__signature__
+keras.activations.celu(x, alpha=1.0)
+__doc__
+Continuously Differentiable Exponential Linear Unit.
+
+The CeLU activation function is defined as:
+
+`celu(x) = alpha * (exp(x / alpha) - 1) for x < 0`,`celu(x) = x for x >= 0`.
+
+where `alpha` is a scaling parameter that controls the activation's shape.
+
+Args:
+    x: Input tensor.
+    alpha: The α value for the CeLU formulation. Defaults to `1.0`.
+
+Reference:
+
+- [Barron, J. T., 2017](https://arxiv.org/abs/1704.07483)
+

.tether/man/activation_glu.txt

@@ -0,0 +1,19 @@
+__signature__
+keras.activations.glu(x, axis=-1)
+__doc__
+Gated Linear Unit (GLU) activation function.
+
+The GLU activation function is defined as:
+
+`glu(x) = a * sigmoid(b)`,
+
+where `x` is split into two equal parts `a` and `b` along the given axis.
+
+Args:
+    x: Input tensor.
+    axis: The axis along which to split the input tensor. Defaults to `-1`.
+
+Reference:
+
+- [Dauphin et al., 2017](https://arxiv.org/abs/1612.08083)
+

.tether/man/activation_hard_shrink.txt

@@ -0,0 +1,14 @@
+__signature__
+keras.activations.hard_shrink(x, threshold=0.5)
+__doc__
+Hard Shrink activation function.
+
+It is defined as:
+
+`hard_shrink(x) = x` if `|x| > threshold`,
+`hard_shrink(x) = 0` otherwise.
+
+Args:
+    x: Input tensor.
+    threshold: Threshold value. Defaults to 0.5.
+

.tether/man/activation_hard_tanh.txt

@@ -0,0 +1,13 @@
+__signature__
+keras.activations.hard_tanh(x)
+__doc__
+HardTanh activation function.
+
+It is defined as:
+`hard_tanh(x) = -1 for x < -1`,
+`hard_tanh(x) = x for -1 <= x <= 1`,
+`hard_tanh(x) = 1 for x > 1`.
+
+Args:
+    x: Input tensor.
+

.tether/man/activation_log_sigmoid.txt

@@ -0,0 +1,10 @@
+__signature__
+keras.activations.log_sigmoid(x)
+__doc__
+Logarithm of the sigmoid activation function.
+
+It is defined as `f(x) = log(1 / (1 + exp(-x)))`.
+
+Args:
+    x: Input tensor.
+

.tether/man/activation_soft_shrink.txt

@@ -0,0 +1,15 @@
+__signature__
+keras.activations.soft_shrink(x, threshold=0.5)
+__doc__
+Soft Shrink activation function.
+
+It is defined as:
+
+`soft_shrink(x) = x - threshold` if `x > threshold`,
+`soft_shrink(x) = x + threshold` if `x < -threshold`,
+`soft_shrink(x) = 0` otherwise.
+
+Args:
+    x: Input tensor.
+    threshold: Threshold value. Defaults to 0.5.
+

.tether/man/activation_squareplus.txt

@@ -0,0 +1,19 @@
+__signature__
+keras.activations.squareplus(x, b=4)
+__doc__
+Squareplus activation function.
+
+The Squareplus activation function is defined as:
+
+`f(x) = (x + sqrt(x^2 + b)) / 2`
+
+Where `b` is a smoothness parameter.
+
+Args:
+    x: Input tensor.
+    b: Smoothness parameter. Defaults to 4.
+
+Reference:
+
+- [Ramachandran et al., 2021](https://arxiv.org/abs/2112.11687)
+

.tether/man/activation_tanh_shrink.txt

@@ -0,0 +1,12 @@
+__signature__
+keras.activations.tanh_shrink(x)
+__doc__
+Tanh shrink activation function.
+
+It is defined as:
+
+`f(x) = x - tanh(x)`.
+
+Args:
+    x: Input tensor.
+

.tether/man/callback_backup_and_restore.txt

@@ -1,7 +1,7 @@
 Help on class BackupAndRestore in module keras.src.callbacks.backup_and_restore:
 
 class BackupAndRestore(keras.src.callbacks.callback.Callback)
- |  BackupAndRestore(backup_dir, save_freq='epoch', delete_checkpoint=True)
+ |  BackupAndRestore(backup_dir, save_freq='epoch', double_checkpoint=False, delete_checkpoint=True)
  |
  |  Callback to back up and restore the training state.
  |
@@ -59,6 +59,12 @@ class BackupAndRestore(keras.src.callbacks.callback.Callback)
  |        When set to an integer, the callback saves the checkpoint every
  |        `save_freq` batches. Set `save_freq=False` only if using
  |        preemption checkpointing (i.e. with `save_before_preemption=True`).
+ |      double_checkpoint: Boolean. If enabled, `BackupAndRestore` callback
+ |        will save 2 last training states (current and previous). After
+ |        interruption if current state can't be loaded due to IO error
+ |        (e.g. file corrupted) it will try to restore previous one. Such
+ |        behaviour will consume twice more space on disk, but increase fault
+ |        tolerance. Defaults to `False`.
  |      delete_checkpoint: Boolean. This `BackupAndRestore`
  |        callback works by saving a checkpoint to back up the training state.
  |        If `delete_checkpoint=True`, the checkpoint will be deleted after
@@ -76,6 +82,7 @@ class BackupAndRestore(keras.src.callbacks.callback.Callback)
  |    self,
  |    backup_dir,
  |    save_freq='epoch',
+ |    double_checkpoint=False,
  |    delete_checkpoint=True
  |  )
  |      Initialize self.  See help(type(self)) for accurate signature.
@@ -116,7 +123,13 @@ class BackupAndRestore(keras.src.callbacks.callback.Callback)
  |          logs: Dict. Aggregated metric results up until this batch.
  |
  |  on_train_begin(self, logs=None)
- |      Get training state from temporary file and restore it.
+ |      Called at the beginning of training.
+ |
+ |      Subclasses should override for any actions to run.
+ |
+ |      Args:
+ |          logs: Dict. Currently no data is passed to this argument for this
+ |            method but that may change in the future.
  |
  |  on_train_end(self, logs=None)
  |      Called at the end of training.