MonoDense

mono_dense_keras.MonoDense

Bases: Dense

Monotonic counterpart of the regular Dense Layer of tf.keras

This is an implementation of our Monotonic Dense Unit or Constrained Monotone Fully Connected Layer. The below is the figure from the paper for reference.

• the parameter monotonicity_indicator corresponds to t in the figure below, and

• parameters is_convex, is_concave and activation_weights are used to calculate the activation selector s as follows:

  • if is_convex or is_concave is True, then the activation selector s will be (units, 0, 0) and (0, units, 0), respecively.

  • if both is_convex or is_concave is False, then the activation_weights represent ratios between \(\breve{s}\), \(\hat{s}\) and \(\tilde{s}\), respecively. E.g. if activation_weights = (2, 2, 1) and units = 10, then

\[ (\breve{s}, \hat{s}, \tilde{s}) = (4, 4, 2) \]

Source code in mono_dense_keras/_components/mono_dense_layer.py

@export
class MonoDense(Dense):
    """Monotonic counterpart of the regular Dense Layer of tf.keras

    This is an implementation of our Monotonic Dense Unit or Constrained Monotone Fully Connected Layer. The below is the figure from the paper for reference.

    - the parameter `monotonicity_indicator` corresponds to **t** in the figure below, and

    - parameters `is_convex`, `is_concave` and `activation_weights` are used to calculate the activation selector **s** as follows:

        - if `is_convex` or `is_concave` is **True**, then the activation selector **s** will be (`units`, 0, 0) and (0, `units`, 0), respecively.

        - if both  `is_convex` or `is_concave` is **False**, then the `activation_weights` represent ratios between $\\breve{s}$, $\\hat{s}$ and $\\tilde{s}$,
          respecively. E.g. if `activation_weights = (2, 2, 1)` and `units = 10`, then

    $$
    (\\breve{s}, \\hat{s}, \\tilde{s}) = (4, 4, 2)
    $$

    ![mono-dense-layer-diagram.png](../../../images/nbs/images/mono-dense-layer-diagram.png)

    """

    def __init__(
        self,
        units: int,
        *,
        activation: Optional[Union[str, Callable[[TensorLike], TensorLike]]] = None,
        monotonicity_indicator: ArrayLike = 1,
        is_convex: bool = False,
        is_concave: bool = False,
        activation_weights: Tuple[float, float, float] = (7.0, 7.0, 2.0),
        **kwargs: Any,
    ):
        """Constructs a new MonoDense instance.

        Params:
            units: Positive integer, dimensionality of the output space.
            activation: Activation function to use, it is assumed to be convex monotonically
                increasing function such as "relu" or "elu"
            monotonicity_indicator: Vector to indicate which of the inputs are monotonically increasing or
                monotonically decreasing or non-monotonic. Has value 1 for monotonically increasing,
                -1 for monotonically decreasing and 0 for non-monotonic.
            is_convex: convex if set to True
            is_concave: concave if set to True
            activation_weights: relative weights for each type of activation, the default is (1.0, 1.0, 1.0).
                Ignored if is_convex or is_concave is set to True
            **kwargs: passed as kwargs to the constructor of `Dense`

        Raise:
            ValueError:
                - if both **is_concave** and **is_convex** are set to **True**, or
                - if any component of activation_weights is negative or there is not exactly three components
        """
        if is_convex and is_concave:
            raise ValueError(
                "The model cannot be set to be both convex and concave (only linear functions are both)."
            )

        if len(activation_weights) != 3:
            raise ValueError(
                f"There must be exactly three components of activation_weights, but we have this instead: {activation_weights}."
            )

        if (np.array(activation_weights) < 0).any():
            raise ValueError(
                f"Values of activation_weights must be non-negative, but we have this instead: {activation_weights}."
            )

        super(MonoDense, self).__init__(units=units, activation=None, **kwargs)

        self.units = units
        self.org_activation = activation
        self.activation_weights = activation_weights
        self.monotonicity_indicator = monotonicity_indicator
        self.is_convex = is_convex
        self.is_concave = is_concave

        (
            self.convex_activation,
            self.concave_activation,
            self.saturated_activation,
        ) = get_activation_functions(self.org_activation)

    def build(self, input_shape: Tuple, *args: List[Any], **kwargs: Any) -> None:
        """Build

        Args:
            input_shape: input tensor
            args: positional arguments passed to Dense.build()
            kwargs: keyword arguments passed to Dense.build()
        """
        super(MonoDense, self).build(input_shape, *args, **kwargs)
        self.monotonicity_indicator = get_monotonicity_indicator(
            monotonicity_indicator=self.monotonicity_indicator,
            input_shape=input_shape,
            units=self.units,
        )

    def call(self, inputs: TensorLike) -> TensorLike:
        """Call

        Args:
            inputs: input tensor of shape (batch_size, ..., x_length)

        Returns:
            N-D tensor with shape: `(batch_size, ..., units)`.

        """
        # calculate W'*x+y after we replace the kernal according to monotonicity vector
        with replace_kernel_using_monotonicity_indicator(
            self, monotonicity_indicator=self.monotonicity_indicator
        ):
            h = super(MonoDense, self).call(inputs)

        y = apply_activations(
            h,
            units=self.units,
            convex_activation=self.convex_activation,
            concave_activation=self.concave_activation,
            saturated_activation=self.saturated_activation,
            is_convex=self.is_convex,
            is_concave=self.is_concave,
            activation_weights=self.activation_weights,
        )

        return y

Attributes

activation_weights = activation_weights instance-attribute

is_concave = is_concave instance-attribute

is_convex = is_convex instance-attribute

monotonicity_indicator = monotonicity_indicator instance-attribute

org_activation = activation instance-attribute

units = units instance-attribute

Functions

__init__(units: int, *, activation: Optional[Union[str, Callable[[TensorLike], TensorLike]]] = None, monotonicity_indicator: ArrayLike = 1, is_convex: bool = False, is_concave: bool = False, activation_weights: Tuple[float, float, float] = (7.0, 7.0, 2.0), **kwargs: Any)

Constructs a new MonoDense instance.

Parameters:

Name	Type	Description	Default
units	int	Positive integer, dimensionality of the output space.	required
activation	Optional[Union[str, Callable[[TensorLike], TensorLike]]]	Activation function to use, it is assumed to be convex monotonically increasing function such as "relu" or "elu"	None
monotonicity_indicator	ArrayLike	Vector to indicate which of the inputs are monotonically increasing or monotonically decreasing or non-monotonic. Has value 1 for monotonically increasing, -1 for monotonically decreasing and 0 for non-monotonic.	1
is_convex	bool	convex if set to True	False
is_concave	bool	concave if set to True	False
activation_weights	Tuple[float, float, float]	relative weights for each type of activation, the default is (1.0, 1.0, 1.0). Ignored if is_convex or is_concave is set to True	(7.0, 7.0, 2.0)
**kwargs	Any	passed as kwargs to the constructor of Dense	{}

Raise

ValueError: - if both is_concave and is_convex are set to True, or - if any component of activation_weights is negative or there is not exactly three components

Source code in mono_dense_keras/_components/mono_dense_layer.py

def __init__(
    self,
    units: int,
    *,
    activation: Optional[Union[str, Callable[[TensorLike], TensorLike]]] = None,
    monotonicity_indicator: ArrayLike = 1,
    is_convex: bool = False,
    is_concave: bool = False,
    activation_weights: Tuple[float, float, float] = (7.0, 7.0, 2.0),
    **kwargs: Any,
):
    """Constructs a new MonoDense instance.

    Params:
        units: Positive integer, dimensionality of the output space.
        activation: Activation function to use, it is assumed to be convex monotonically
            increasing function such as "relu" or "elu"
        monotonicity_indicator: Vector to indicate which of the inputs are monotonically increasing or
            monotonically decreasing or non-monotonic. Has value 1 for monotonically increasing,
            -1 for monotonically decreasing and 0 for non-monotonic.
        is_convex: convex if set to True
        is_concave: concave if set to True
        activation_weights: relative weights for each type of activation, the default is (1.0, 1.0, 1.0).
            Ignored if is_convex or is_concave is set to True
        **kwargs: passed as kwargs to the constructor of `Dense`

    Raise:
        ValueError:
            - if both **is_concave** and **is_convex** are set to **True**, or
            - if any component of activation_weights is negative or there is not exactly three components
    """
    if is_convex and is_concave:
        raise ValueError(
            "The model cannot be set to be both convex and concave (only linear functions are both)."
        )

    if len(activation_weights) != 3:
        raise ValueError(
            f"There must be exactly three components of activation_weights, but we have this instead: {activation_weights}."
        )

    if (np.array(activation_weights) < 0).any():
        raise ValueError(
            f"Values of activation_weights must be non-negative, but we have this instead: {activation_weights}."
        )

    super(MonoDense, self).__init__(units=units, activation=None, **kwargs)

    self.units = units
    self.org_activation = activation
    self.activation_weights = activation_weights
    self.monotonicity_indicator = monotonicity_indicator
    self.is_convex = is_convex
    self.is_concave = is_concave

    (
        self.convex_activation,
        self.concave_activation,
        self.saturated_activation,
    ) = get_activation_functions(self.org_activation)

add_loss(losses, **kwargs)

Add loss tensor(s), potentially dependent on layer inputs.

Some losses (for instance, activity regularization losses) may be dependent on the inputs passed when calling a layer. Hence, when reusing the same layer on different inputs a and b, some entries in layer.losses may be dependent on a and some on b. This method automatically keeps track of dependencies.

This method can be used inside a subclassed layer or model's call function, in which case losses should be a Tensor or list of Tensors.

Example:

class MyLayer(tf.keras.layers.Layer):
  def call(self, inputs):
    self.add_loss(tf.abs(tf.reduce_mean(inputs)))
    return inputs

The same code works in distributed training: the input to add_loss() is treated like a regularization loss and averaged across replicas by the training loop (both built-in Model.fit() and compliant custom training loops).

The add_loss method can also be called directly on a Functional Model during construction. In this case, any loss Tensors passed to this Model must be symbolic and be able to be traced back to the model's Inputs. These losses become part of the model's topology and are tracked in get_config.

Example:

inputs = tf.keras.Input(shape=(10,))
x = tf.keras.layers.Dense(10)(inputs)
outputs = tf.keras.layers.Dense(1)(x)
model = tf.keras.Model(inputs, outputs)
# Activity regularization.
model.add_loss(tf.abs(tf.reduce_mean(x)))

If this is not the case for your loss (if, for example, your loss references a Variable of one of the model's layers), you can wrap your loss in a zero-argument lambda. These losses are not tracked as part of the model's topology since they can't be serialized.

Example:

inputs = tf.keras.Input(shape=(10,))
d = tf.keras.layers.Dense(10)
x = d(inputs)
outputs = tf.keras.layers.Dense(1)(x)
model = tf.keras.Model(inputs, outputs)
# Weight regularization.
model.add_loss(lambda: tf.reduce_mean(d.kernel))

Parameters:

Name	Type	Description	Default
losses		Loss tensor, or list/tuple of tensors. Rather than tensors, losses may also be zero-argument callables which create a loss tensor.	required
**kwargs		Used for backwards compatibility only.	{}

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

def add_loss(self, losses, **kwargs):
    """Add loss tensor(s), potentially dependent on layer inputs.

    Some losses (for instance, activity regularization losses) may be
    dependent on the inputs passed when calling a layer. Hence, when reusing
    the same layer on different inputs `a` and `b`, some entries in
    `layer.losses` may be dependent on `a` and some on `b`. This method
    automatically keeps track of dependencies.

    This method can be used inside a subclassed layer or model's `call`
    function, in which case `losses` should be a Tensor or list of Tensors.

    Example:

    ```python
    class MyLayer(tf.keras.layers.Layer):
      def call(self, inputs):
        self.add_loss(tf.abs(tf.reduce_mean(inputs)))
        return inputs
    ```

    The same code works in distributed training: the input to `add_loss()`
    is treated like a regularization loss and averaged across replicas
    by the training loop (both built-in `Model.fit()` and compliant custom
    training loops).

    The `add_loss` method can also be called directly on a Functional Model
    during construction. In this case, any loss Tensors passed to this Model
    must be symbolic and be able to be traced back to the model's `Input`s.
    These losses become part of the model's topology and are tracked in
    `get_config`.

    Example:

    ```python
    inputs = tf.keras.Input(shape=(10,))
    x = tf.keras.layers.Dense(10)(inputs)
    outputs = tf.keras.layers.Dense(1)(x)
    model = tf.keras.Model(inputs, outputs)
    # Activity regularization.
    model.add_loss(tf.abs(tf.reduce_mean(x)))
    ```

    If this is not the case for your loss (if, for example, your loss
    references a `Variable` of one of the model's layers), you can wrap your
    loss in a zero-argument lambda. These losses are not tracked as part of
    the model's topology since they can't be serialized.

    Example:

    ```python
    inputs = tf.keras.Input(shape=(10,))
    d = tf.keras.layers.Dense(10)
    x = d(inputs)
    outputs = tf.keras.layers.Dense(1)(x)
    model = tf.keras.Model(inputs, outputs)
    # Weight regularization.
    model.add_loss(lambda: tf.reduce_mean(d.kernel))
    ```

    Args:
      losses: Loss tensor, or list/tuple of tensors. Rather than tensors,
        losses may also be zero-argument callables which create a loss
        tensor.
      **kwargs: Used for backwards compatibility only.
    """
    kwargs.pop("inputs", None)
    if kwargs:
        raise TypeError(f"Unknown keyword arguments: {kwargs.keys()}")

    def _tag_callable(loss):
        """Tags callable loss tensor as `_unconditional_loss`."""
        if callable(loss):
            # We run the loss without autocasting, as regularizers are often
            # numerically unstable in float16.
            with autocast_variable.enable_auto_cast_variables(None):
                loss = loss()
        if loss is None:
            # Will be filtered out when computing the .losses property
            return None
        if not tf.is_tensor(loss):
            loss = tf.convert_to_tensor(loss, dtype=backend.floatx())
        loss._unconditional_loss = True
        return loss

    losses = tf.nest.flatten(losses)

    callable_losses = []
    eager_losses = []
    symbolic_losses = []
    for loss in losses:
        if callable(loss):
            callable_losses.append(functools.partial(_tag_callable, loss))
            continue
        if loss is None:
            continue
        if not tf.is_tensor(loss) and not isinstance(
            loss, keras_tensor.KerasTensor
        ):
            loss = tf.convert_to_tensor(loss, dtype=backend.floatx())
        # TF Functions should take the eager path.
        if (
            tf_utils.is_symbolic_tensor(loss)
            or isinstance(loss, keras_tensor.KerasTensor)
        ) and not base_layer_utils.is_in_tf_function():
            symbolic_losses.append(loss)
        elif tf.is_tensor(loss):
            eager_losses.append(loss)

    self._callable_losses.extend(callable_losses)

    in_call_context = base_layer_utils.call_context().in_call
    if eager_losses and not in_call_context:
        raise ValueError(
            "Expected a symbolic Tensors or a callable for the loss value. "
            "Please wrap your loss computation in a zero argument `lambda`."
        )

    self._eager_losses.extend(eager_losses)

    for symbolic_loss in symbolic_losses:
        if getattr(self, "_is_graph_network", False):
            self._graph_network_add_loss(symbolic_loss)
        else:
            # Possible a loss was added in a Layer's `build`.
            self._losses.append(symbolic_loss)

add_metric(value, name = None, **kwargs)

Adds metric tensor to the layer.

This method can be used inside the call() method of a subclassed layer or model.

class MyMetricLayer(tf.keras.layers.Layer):
  def __init__(self):
    super(MyMetricLayer, self).__init__(name='my_metric_layer')
    self.mean = tf.keras.metrics.Mean(name='metric_1')

  def call(self, inputs):
    self.add_metric(self.mean(inputs))
    self.add_metric(tf.reduce_sum(inputs), name='metric_2')
    return inputs

This method can also be called directly on a Functional Model during construction. In this case, any tensor passed to this Model must be symbolic and be able to be traced back to the model's Inputs. These metrics become part of the model's topology and are tracked when you save the model via save().

inputs = tf.keras.Input(shape=(10,))
x = tf.keras.layers.Dense(10)(inputs)
outputs = tf.keras.layers.Dense(1)(x)
model = tf.keras.Model(inputs, outputs)
model.add_metric(math_ops.reduce_sum(x), name='metric_1')

Note: Calling add_metric() with the result of a metric object on a Functional Model, as shown in the example below, is not supported. This is because we cannot trace the metric result tensor back to the model's inputs.

inputs = tf.keras.Input(shape=(10,))
x = tf.keras.layers.Dense(10)(inputs)
outputs = tf.keras.layers.Dense(1)(x)
model = tf.keras.Model(inputs, outputs)
model.add_metric(tf.keras.metrics.Mean()(x), name='metric_1')

Parameters:

Name	Type	Description	Default
value		Metric tensor.	required
name		String metric name.	None
**kwargs		Additional keyword arguments for backward compatibility. Accepted values: aggregation - When the value tensor provided is not the result of calling a keras.Metric instance, it will be aggregated by default using a keras.Metric.Mean.	{}

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

def add_metric(self, value, name=None, **kwargs):
    """Adds metric tensor to the layer.

    This method can be used inside the `call()` method of a subclassed layer
    or model.

    ```python
    class MyMetricLayer(tf.keras.layers.Layer):
      def __init__(self):
        super(MyMetricLayer, self).__init__(name='my_metric_layer')
        self.mean = tf.keras.metrics.Mean(name='metric_1')

      def call(self, inputs):
        self.add_metric(self.mean(inputs))
        self.add_metric(tf.reduce_sum(inputs), name='metric_2')
        return inputs
    ```

    This method can also be called directly on a Functional Model during
    construction. In this case, any tensor passed to this Model must
    be symbolic and be able to be traced back to the model's `Input`s. These
    metrics become part of the model's topology and are tracked when you
    save the model via `save()`.

    ```python
    inputs = tf.keras.Input(shape=(10,))
    x = tf.keras.layers.Dense(10)(inputs)
    outputs = tf.keras.layers.Dense(1)(x)
    model = tf.keras.Model(inputs, outputs)
    model.add_metric(math_ops.reduce_sum(x), name='metric_1')
    ```

    Note: Calling `add_metric()` with the result of a metric object on a
    Functional Model, as shown in the example below, is not supported. This
    is because we cannot trace the metric result tensor back to the model's
    inputs.

    ```python
    inputs = tf.keras.Input(shape=(10,))
    x = tf.keras.layers.Dense(10)(inputs)
    outputs = tf.keras.layers.Dense(1)(x)
    model = tf.keras.Model(inputs, outputs)
    model.add_metric(tf.keras.metrics.Mean()(x), name='metric_1')
    ```

    Args:
      value: Metric tensor.
      name: String metric name.
      **kwargs: Additional keyword arguments for backward compatibility.
        Accepted values:
        `aggregation` - When the `value` tensor provided is not the result
        of calling a `keras.Metric` instance, it will be aggregated by
        default using a `keras.Metric.Mean`.
    """
    kwargs_keys = list(kwargs.keys())
    if len(kwargs_keys) > 1 or (
        len(kwargs_keys) == 1 and kwargs_keys[0] != "aggregation"
    ):
        raise TypeError(
            f"Unknown keyword arguments: {kwargs.keys()}. "
            "Expected `aggregation`."
        )

    from_metric_obj = hasattr(value, "_metric_obj")
    is_symbolic = isinstance(value, keras_tensor.KerasTensor)
    in_call_context = base_layer_utils.call_context().in_call

    if name is None and not from_metric_obj:
        # Eg. `self.add_metric(math_ops.reduce_sum(x))` In eager mode, we
        # use metric name to lookup a metric. Without a name, a new Mean
        # metric wrapper will be created on every model/layer call. So, we
        # raise an error when no name is provided. We will do the same for
        # symbolic mode for consistency although a name will be generated if
        # no name is provided.

        # We will not raise this error in the foll use case for the sake of
        # consistency as name in provided in the metric constructor.
        # mean = metrics.Mean(name='my_metric')
        # model.add_metric(mean(outputs))
        raise ValueError(
            "Please provide a name for your metric like "
            "`self.add_metric(tf.reduce_sum(inputs), "
            "name='mean_activation')`"
        )
    elif from_metric_obj:
        name = value._metric_obj.name

    if not in_call_context and not is_symbolic:
        raise ValueError(
            "Expected a symbolic Tensor for the metric value, received: "
            + str(value)
        )

    # If a metric was added in a Layer's `call` or `build`.
    if in_call_context or not getattr(self, "_is_graph_network", False):
        # TF Function path should take the eager path.

        # If the given metric is available in `metrics` list we just update
        # state on it, otherwise we create a new metric instance and
        # add it to the `metrics` list.
        metric_obj = getattr(value, "_metric_obj", None)
        # Tensors that come from a Metric object already updated the Metric
        # state.
        should_update_state = not metric_obj
        name = metric_obj.name if metric_obj else name

        with self._metrics_lock:
            match = self._get_existing_metric(name)
            if match:
                metric_obj = match
            elif metric_obj:
                self._metrics.append(metric_obj)
            else:
                # Build the metric object with the value's dtype if it
                # defines one
                metric_obj = metrics_mod.Mean(
                    name=name, dtype=getattr(value, "dtype", None)
                )
                self._metrics.append(metric_obj)

        if should_update_state:
            metric_obj(value)
    else:
        if from_metric_obj:
            raise ValueError(
                "Using the result of calling a `Metric` object "
                "when calling `add_metric` on a Functional "
                "Model is not supported. Please pass the "
                "Tensor to monitor directly."
            )

        # Insert layers into the Keras Graph Network.
        aggregation = None if from_metric_obj else "mean"
        self._graph_network_add_metric(value, aggregation, name)

add_update(updates)

Add update op(s), potentially dependent on layer inputs.

Weight updates (for instance, the updates of the moving mean and variance in a BatchNormalization layer) may be dependent on the inputs passed when calling a layer. Hence, when reusing the same layer on different inputs a and b, some entries in layer.updates may be dependent on a and some on b. This method automatically keeps track of dependencies.

This call is ignored when eager execution is enabled (in that case, variable updates are run on the fly and thus do not need to be tracked for later execution).

Parameters:

Name	Type	Description	Default
updates		Update op, or list/tuple of update ops, or zero-arg callable that returns an update op. A zero-arg callable should be passed in order to disable running the updates by setting trainable=False on this Layer, when executing in Eager mode.	required

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_doc_inheritable
def add_update(self, updates):
    """Add update op(s), potentially dependent on layer inputs.

    Weight updates (for instance, the updates of the moving mean and
    variance in a BatchNormalization layer) may be dependent on the inputs
    passed when calling a layer. Hence, when reusing the same layer on
    different inputs `a` and `b`, some entries in `layer.updates` may be
    dependent on `a` and some on `b`. This method automatically keeps track
    of dependencies.

    This call is ignored when eager execution is enabled (in that case,
    variable updates are run on the fly and thus do not need to be tracked
    for later execution).

    Args:
      updates: Update op, or list/tuple of update ops, or zero-arg callable
        that returns an update op. A zero-arg callable should be passed in
        order to disable running the updates by setting `trainable=False`
        on this Layer, when executing in Eager mode.
    """
    call_context = base_layer_utils.call_context()
    # No need to run updates during Functional API construction.
    if call_context.in_keras_graph:
        return

    # Callable updates are disabled by setting `trainable=False`.
    if not call_context.frozen:
        for update in tf.nest.flatten(updates):
            if callable(update):
                update()

add_variable(*args, **kwargs)

Deprecated, do NOT use! Alias for add_weight.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_doc_inheritable
def add_variable(self, *args, **kwargs):
    """Deprecated, do NOT use! Alias for `add_weight`."""
    warnings.warn(
        "`layer.add_variable` is deprecated and "
        "will be removed in a future version. "
        "Please use the `layer.add_weight()` method instead.",
        stacklevel=2,
    )
    return self.add_weight(*args, **kwargs)

add_weight(name = None, shape = None, dtype = None, initializer = None, regularizer = None, trainable = None, constraint = None, use_resource = None, synchronization = tf.VariableSynchronization.AUTO, aggregation = tf.VariableAggregation.NONE, **kwargs)

Adds a new variable to the layer.

Parameters:

Name	Type	Description	Default
name		Variable name.	None
shape		Variable shape. Defaults to scalar if unspecified.	None
dtype		The type of the variable. Defaults to self.dtype.	None
initializer		Initializer instance (callable).	None
regularizer		Regularizer instance (callable).	None
trainable		Boolean, whether the variable should be part of the layer's "trainable_variables" (e.g. variables, biases) or "non_trainable_variables" (e.g. BatchNorm mean and variance). Note that trainable cannot be True if synchronization is set to ON_READ.	None
constraint		Constraint instance (callable).	None
use_resource		Whether to use a ResourceVariable or not. See this guide for more information.	None
synchronization		Indicates when a distributed a variable will be aggregated. Accepted values are constants defined in the class tf.VariableSynchronization. By default the synchronization is set to AUTO and the current DistributionStrategy chooses when to synchronize. If synchronization is set to ON_READ, trainable must not be set to True.	tf.VariableSynchronization.AUTO
aggregation		Indicates how a distributed variable will be aggregated. Accepted values are constants defined in the class tf.VariableAggregation.	tf.VariableAggregation.NONE
**kwargs		Additional keyword arguments. Accepted values are getter, collections, experimental_autocast and caching_device.	{}

Returns:

Type	Description
	The variable created.

Raises:

Type	Description
ValueError	When giving unsupported dtype and no initializer or when trainable has been set to True with synchronization set as ON_READ.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.for_subclass_implementers
def add_weight(
    self,
    name=None,
    shape=None,
    dtype=None,
    initializer=None,
    regularizer=None,
    trainable=None,
    constraint=None,
    use_resource=None,
    synchronization=tf.VariableSynchronization.AUTO,
    aggregation=tf.VariableAggregation.NONE,
    **kwargs,
):
    """Adds a new variable to the layer.

    Args:
      name: Variable name.
      shape: Variable shape. Defaults to scalar if unspecified.
      dtype: The type of the variable. Defaults to `self.dtype`.
      initializer: Initializer instance (callable).
      regularizer: Regularizer instance (callable).
      trainable: Boolean, whether the variable should be part of the layer's
        "trainable_variables" (e.g. variables, biases)
        or "non_trainable_variables" (e.g. BatchNorm mean and variance).
        Note that `trainable` cannot be `True` if `synchronization`
        is set to `ON_READ`.
      constraint: Constraint instance (callable).
      use_resource: Whether to use a `ResourceVariable` or not.
        See [this guide](
        https://www.tensorflow.org/guide/migrate/tf1_vs_tf2#resourcevariables_instead_of_referencevariables)
         for more information.
      synchronization: Indicates when a distributed a variable will be
        aggregated. Accepted values are constants defined in the class
        `tf.VariableSynchronization`. By default the synchronization is set
        to `AUTO` and the current `DistributionStrategy` chooses when to
        synchronize. If `synchronization` is set to `ON_READ`, `trainable`
        must not be set to `True`.
      aggregation: Indicates how a distributed variable will be aggregated.
        Accepted values are constants defined in the class
        `tf.VariableAggregation`.
      **kwargs: Additional keyword arguments. Accepted values are `getter`,
        `collections`, `experimental_autocast` and `caching_device`.

    Returns:
      The variable created.

    Raises:
      ValueError: When giving unsupported dtype and no initializer or when
        trainable has been set to True with synchronization set as
        `ON_READ`.
    """
    if shape is None:
        shape = ()
    kwargs.pop("partitioner", None)  # Ignored.
    # Validate optional keyword arguments.
    for kwarg in kwargs:
        if kwarg not in [
            "collections",
            "experimental_autocast",
            "caching_device",
            "getter",
            "layout",
        ]:
            raise TypeError("Unknown keyword argument:", kwarg)
    collections_arg = kwargs.pop("collections", None)
    # 'experimental_autocast' can be set to False by the caller to indicate
    # an AutoCastVariable should never be created.
    autocast = kwargs.pop("experimental_autocast", True)
    # See the docstring for tf.Variable about the details for
    # caching_device.
    caching_device = kwargs.pop("caching_device", None)

    layout = kwargs.pop("layout", None)
    # Specially handling of auto layout fetch, based on the variable name
    # and attribute name. For built-in keras layers, usually the variable
    # name, eg 'kernel', will match with a 'kernel_layout' attribute name on
    # the instance. We will try to do this auto fetch if layout is not
    # explicitly specified. This is mainly a quick workaround for not
    # applying too many interface change to built-in layers, until DTensor
    # is a public API.  Also see dtensor.utils.allow_initializer_layout for
    # more details.
    # TODO(scottzhu): Remove this once dtensor is public to end user.
    if not layout and name:
        layout = getattr(self, name + "_layout", None)

    if dtype is None:
        dtype = self.dtype or backend.floatx()
    dtype = tf.as_dtype(dtype)
    if self._dtype_policy.variable_dtype is None:
        # The policy is "_infer", so we infer the policy from the variable
        # dtype.
        self._set_dtype_policy(policy.Policy(dtype.base_dtype.name))
    initializer = initializers.get(initializer)
    regularizer = regularizers.get(regularizer)
    constraint = constraints.get(constraint)

    if synchronization == tf.VariableSynchronization.ON_READ:
        if trainable:
            raise ValueError(
                "Synchronization value can be set to "
                "VariableSynchronization.ON_READ only for non-trainable "
                "variables. You have specified trainable=True and "
                "synchronization=VariableSynchronization.ON_READ."
            )
        else:
            # Set trainable to be false when variable is to be synced on
            # read.
            trainable = False
    elif trainable is None:
        trainable = True

    # Initialize variable when no initializer provided
    if initializer is None:
        # If dtype is DT_FLOAT, provide a uniform unit scaling initializer
        if dtype.is_floating:
            initializer = initializers.get("glorot_uniform")
        # If dtype is DT_INT/DT_UINT, provide a default value `zero`
        # If dtype is DT_BOOL, provide a default value `FALSE`
        elif dtype.is_integer or dtype.is_unsigned or dtype.is_bool:
            initializer = initializers.get("zeros")
        # NOTES:Do we need to support for handling DT_STRING and DT_COMPLEX
        # here?
        elif "getter" not in kwargs:
            # When `getter` is specified, it's possibly fine for
            # `initializer` to be None since it's up to the custom `getter`
            # to raise error in case it indeed needs `initializer`.
            raise ValueError(
                f"An initializer for variable {name} of type "
                f"{dtype.base_dtype} is required for layer "
                f"{self.name}. Received: {initializer}."
            )

    getter = kwargs.pop("getter", base_layer_utils.make_variable)
    if (
        autocast
        and self._dtype_policy.compute_dtype
        != self._dtype_policy.variable_dtype
        and dtype.is_floating
    ):
        old_getter = getter

        # Wrap variable constructor to return an AutoCastVariable.
        def getter(*args, **kwargs):
            variable = old_getter(*args, **kwargs)
            return autocast_variable.create_autocast_variable(variable)

        # Also the caching_device does not work with the mixed precision
        # API, disable it if it is specified.
        # TODO(b/142020079): Re-enable it once the bug is fixed.
        if caching_device is not None:
            tf_logging.warning(
                "`caching_device` does not work with mixed precision API. "
                "Ignoring user specified `caching_device`."
            )
            caching_device = None
    if layout:
        getter = functools.partial(getter, layout=layout)

    variable = self._add_variable_with_custom_getter(
        name=name,
        shape=shape,
        # TODO(allenl): a `make_variable` equivalent should be added as a
        # `Trackable` method.
        getter=getter,
        # Manage errors in Layer rather than Trackable.
        overwrite=True,
        initializer=initializer,
        dtype=dtype,
        constraint=constraint,
        trainable=trainable,
        use_resource=use_resource,
        collections=collections_arg,
        synchronization=synchronization,
        aggregation=aggregation,
        caching_device=caching_device,
    )
    if regularizer is not None:
        # TODO(fchollet): in the future, this should be handled at the
        # level of variable creation, and weight regularization losses
        # should be variable attributes.
        name_in_scope = variable.name[: variable.name.find(":")]
        self._handle_weight_regularization(
            name_in_scope, variable, regularizer
        )
    if base_layer_utils.is_split_variable(variable):
        for v in variable:
            backend.track_variable(v)
            if trainable:
                self._trainable_weights.append(v)
            else:
                self._non_trainable_weights.append(v)
    else:
        backend.track_variable(variable)
        if trainable:
            self._trainable_weights.append(variable)
        else:
            self._non_trainable_weights.append(variable)
    return variable

build(input_shape: Tuple, *args: List[Any], **kwargs: Any) -> None

Build

Parameters:

Name	Type	Description	Default
input_shape	Tuple	input tensor	required
args	List[Any]	positional arguments passed to Dense.build()	()
kwargs	Any	keyword arguments passed to Dense.build()	{}

Source code in mono_dense_keras/_components/mono_dense_layer.py

def build(self, input_shape: Tuple, *args: List[Any], **kwargs: Any) -> None:
    """Build

    Args:
        input_shape: input tensor
        args: positional arguments passed to Dense.build()
        kwargs: keyword arguments passed to Dense.build()
    """
    super(MonoDense, self).build(input_shape, *args, **kwargs)
    self.monotonicity_indicator = get_monotonicity_indicator(
        monotonicity_indicator=self.monotonicity_indicator,
        input_shape=input_shape,
        units=self.units,
    )

call(inputs: TensorLike) -> TensorLike

Call

Parameters:

Name	Type	Description	Default
inputs	TensorLike	input tensor of shape (batch_size, ..., x_length)	required

Returns:

Type	Description
TensorLike	N-D tensor with shape: (batch_size, ..., units).

Source code in mono_dense_keras/_components/mono_dense_layer.py

def call(self, inputs: TensorLike) -> TensorLike:
    """Call

    Args:
        inputs: input tensor of shape (batch_size, ..., x_length)

    Returns:
        N-D tensor with shape: `(batch_size, ..., units)`.

    """
    # calculate W'*x+y after we replace the kernal according to monotonicity vector
    with replace_kernel_using_monotonicity_indicator(
        self, monotonicity_indicator=self.monotonicity_indicator
    ):
        h = super(MonoDense, self).call(inputs)

    y = apply_activations(
        h,
        units=self.units,
        convex_activation=self.convex_activation,
        concave_activation=self.concave_activation,
        saturated_activation=self.saturated_activation,
        is_convex=self.is_convex,
        is_concave=self.is_concave,
        activation_weights=self.activation_weights,
    )

    return y

compute_mask(inputs, mask = None)

Computes an output mask tensor.

Parameters:

Name	Type	Description	Default
inputs		Tensor or list of tensors.	required
mask		Tensor or list of tensors.	None

Returns:

Type	Description
	None or a tensor (or list of tensors, one per output tensor of the layer).

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@generic_utils.default
def compute_mask(self, inputs, mask=None):
    """Computes an output mask tensor.

    Args:
        inputs: Tensor or list of tensors.
        mask: Tensor or list of tensors.

    Returns:
        None or a tensor (or list of tensors,
            one per output tensor of the layer).
    """
    if not self._supports_masking:
        if any(m is not None for m in tf.nest.flatten(mask)):
            raise TypeError(
                "Layer " + self.name + " does not support masking, "
                "but was passed an input_mask: " + str(mask)
            )
        # masking not explicitly supported: return None as mask.
        return None
    # if masking is explicitly supported, by default
    # carry over the input mask
    return mask

compute_output_signature(input_signature)

Compute the output tensor signature of the layer based on the inputs.

Unlike a TensorShape object, a TensorSpec object contains both shape and dtype information for a tensor. This method allows layers to provide output dtype information if it is different from the input dtype. For any layer that doesn't implement this function, the framework will fall back to use compute_output_shape, and will assume that the output dtype matches the input dtype.

Parameters:

Name	Type	Description	Default
input_signature		Single TensorSpec or nested structure of TensorSpec objects, describing a candidate input for the layer.	required

Returns:

Type	Description
	Single TensorSpec or nested structure of TensorSpec objects, describing how the layer would transform the provided input.

Raises:

Type	Description
TypeError	If input_signature contains a non-TensorSpec object.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.for_subclass_implementers
def compute_output_signature(self, input_signature):
    """Compute the output tensor signature of the layer based on the inputs.

    Unlike a TensorShape object, a TensorSpec object contains both shape
    and dtype information for a tensor. This method allows layers to provide
    output dtype information if it is different from the input dtype.
    For any layer that doesn't implement this function,
    the framework will fall back to use `compute_output_shape`, and will
    assume that the output dtype matches the input dtype.

    Args:
      input_signature: Single TensorSpec or nested structure of TensorSpec
        objects, describing a candidate input for the layer.

    Returns:
      Single TensorSpec or nested structure of TensorSpec objects,
        describing how the layer would transform the provided input.

    Raises:
      TypeError: If input_signature contains a non-TensorSpec object.
    """

    def check_type_return_shape(s):
        if not isinstance(s, tf.TensorSpec):
            raise TypeError(
                "Only TensorSpec signature types are supported. "
                f"Received: {s}."
            )
        return s.shape

    input_shape = tf.nest.map_structure(
        check_type_return_shape, input_signature
    )
    output_shape = self.compute_output_shape(input_shape)
    dtype = self._compute_dtype
    if dtype is None:
        input_dtypes = [s.dtype for s in tf.nest.flatten(input_signature)]
        # Default behavior when self.dtype is None, is to use the first
        # input's dtype.
        dtype = input_dtypes[0]
    return tf.nest.map_structure(
        lambda s: tf.TensorSpec(dtype=dtype, shape=s), output_shape
    )

count_params()

Count the total number of scalars composing the weights.

Returns:

Type	Description
	An integer count.

Raises:

Type	Description
ValueError	if the layer isn't yet built (in which case its weights aren't yet defined).

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

def count_params(self):
    """Count the total number of scalars composing the weights.

    Returns:
        An integer count.

    Raises:
        ValueError: if the layer isn't yet built
          (in which case its weights aren't yet defined).
    """
    if not self.built:
        if getattr(self, "_is_graph_network", False):
            with tf_utils.maybe_init_scope(self):
                self._maybe_build(self.inputs)
        else:
            raise ValueError(
                "You tried to call `count_params` "
                f"on layer {self.name}"
                ", but the layer isn't built. "
                "You can build it manually via: "
                f"`{self.name}.build(batch_input_shape)`."
            )
    return layer_utils.count_params(self.weights)

finalize_state()

Finalizes the layers state after updating layer weights.

This function can be subclassed in a layer and will be called after updating a layer weights. It can be overridden to finalize any additional layer state after a weight update.

This function will be called after weights of a layer have been restored from a loaded model.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_generate_docs
def finalize_state(self):
    """Finalizes the layers state after updating layer weights.

    This function can be subclassed in a layer and will be called after
    updating a layer weights. It can be overridden to finalize any
    additional layer state after a weight update.

    This function will be called after weights of a layer have been restored
    from a loaded model.
    """
    pass

get_input_at(node_index)

Retrieves the input tensor(s) of a layer at a given node.

Parameters:

Name	Type	Description	Default
node_index		Integer, index of the node from which to retrieve the attribute. E.g. node_index=0 will correspond to the first input node of the layer.	required

Returns:

Type	Description
	A tensor (or list of tensors if the layer has multiple inputs).

Raises:

Type	Description
RuntimeError	If called in Eager mode.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_doc_inheritable
def get_input_at(self, node_index):
    """Retrieves the input tensor(s) of a layer at a given node.

    Args:
        node_index: Integer, index of the node
            from which to retrieve the attribute.
            E.g. `node_index=0` will correspond to the
            first input node of the layer.

    Returns:
        A tensor (or list of tensors if the layer has multiple inputs).

    Raises:
      RuntimeError: If called in Eager mode.
    """
    return self._get_node_attribute_at_index(
        node_index, "input_tensors", "input"
    )

get_input_mask_at(node_index)

Retrieves the input mask tensor(s) of a layer at a given node.

Parameters:

Name	Type	Description	Default
node_index		Integer, index of the node from which to retrieve the attribute. E.g. node_index=0 will correspond to the first time the layer was called.	required

Returns:

Type	Description
	A mask tensor
	(or list of tensors if the layer has multiple inputs).

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_doc_inheritable
def get_input_mask_at(self, node_index):
    """Retrieves the input mask tensor(s) of a layer at a given node.

    Args:
        node_index: Integer, index of the node
            from which to retrieve the attribute.
            E.g. `node_index=0` will correspond to the
            first time the layer was called.

    Returns:
        A mask tensor
        (or list of tensors if the layer has multiple inputs).
    """
    inputs = self.get_input_at(node_index)
    if isinstance(inputs, list):
        return [getattr(x, "_keras_mask", None) for x in inputs]
    else:
        return getattr(inputs, "_keras_mask", None)

get_input_shape_at(node_index)

Retrieves the input shape(s) of a layer at a given node.

Parameters:

Name	Type	Description	Default
node_index		Integer, index of the node from which to retrieve the attribute. E.g. node_index=0 will correspond to the first time the layer was called.	required

Returns:

Type	Description
	A shape tuple
	(or list of shape tuples if the layer has multiple inputs).

Raises:

Type	Description
RuntimeError	If called in Eager mode.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_doc_inheritable
def get_input_shape_at(self, node_index):
    """Retrieves the input shape(s) of a layer at a given node.

    Args:
        node_index: Integer, index of the node
            from which to retrieve the attribute.
            E.g. `node_index=0` will correspond to the
            first time the layer was called.

    Returns:
        A shape tuple
        (or list of shape tuples if the layer has multiple inputs).

    Raises:
      RuntimeError: If called in Eager mode.
    """
    return self._get_node_attribute_at_index(
        node_index, "input_shapes", "input shape"
    )

get_output_at(node_index)

Retrieves the output tensor(s) of a layer at a given node.

Parameters:

Name	Type	Description	Default
node_index		Integer, index of the node from which to retrieve the attribute. E.g. node_index=0 will correspond to the first output node of the layer.	required

Returns:

Type	Description
	A tensor (or list of tensors if the layer has multiple outputs).

Raises:

Type	Description
RuntimeError	If called in Eager mode.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_doc_inheritable
def get_output_at(self, node_index):
    """Retrieves the output tensor(s) of a layer at a given node.

    Args:
        node_index: Integer, index of the node
            from which to retrieve the attribute.
            E.g. `node_index=0` will correspond to the
            first output node of the layer.

    Returns:
        A tensor (or list of tensors if the layer has multiple outputs).

    Raises:
      RuntimeError: If called in Eager mode.
    """
    return self._get_node_attribute_at_index(
        node_index, "output_tensors", "output"
    )

get_output_mask_at(node_index)

Retrieves the output mask tensor(s) of a layer at a given node.

Parameters:

Name	Type	Description	Default
node_index		Integer, index of the node from which to retrieve the attribute. E.g. node_index=0 will correspond to the first time the layer was called.	required

Returns:

Type	Description
	A mask tensor
	(or list of tensors if the layer has multiple outputs).

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_doc_inheritable
def get_output_mask_at(self, node_index):
    """Retrieves the output mask tensor(s) of a layer at a given node.

    Args:
        node_index: Integer, index of the node
            from which to retrieve the attribute.
            E.g. `node_index=0` will correspond to the
            first time the layer was called.

    Returns:
        A mask tensor
        (or list of tensors if the layer has multiple outputs).
    """
    output = self.get_output_at(node_index)
    if isinstance(output, list):
        return [getattr(x, "_keras_mask", None) for x in output]
    else:
        return getattr(output, "_keras_mask", None)

get_output_shape_at(node_index)

Retrieves the output shape(s) of a layer at a given node.

Parameters:

Name	Type	Description	Default
node_index		Integer, index of the node from which to retrieve the attribute. E.g. node_index=0 will correspond to the first time the layer was called.	required

Returns:

Type	Description
	A shape tuple
	(or list of shape tuples if the layer has multiple outputs).

Raises:

Type	Description
RuntimeError	If called in Eager mode.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

@doc_controls.do_not_doc_inheritable
def get_output_shape_at(self, node_index):
    """Retrieves the output shape(s) of a layer at a given node.

    Args:
        node_index: Integer, index of the node
            from which to retrieve the attribute.
            E.g. `node_index=0` will correspond to the
            first time the layer was called.

    Returns:
        A shape tuple
        (or list of shape tuples if the layer has multiple outputs).

    Raises:
      RuntimeError: If called in Eager mode.
    """
    return self._get_node_attribute_at_index(
        node_index, "output_shapes", "output shape"
    )

get_weights()

Returns the current weights of the layer, as NumPy arrays.

The weights of a layer represent the state of the layer. This function returns both trainable and non-trainable weight values associated with this layer as a list of NumPy arrays, which can in turn be used to load state into similarly parameterized layers.

For example, a Dense layer returns a list of two values: the kernel matrix and the bias vector. These can be used to set the weights of another Dense layer:

layer_a = tf.keras.layers.Dense(1, ... kernel_initializer=tf.constant_initializer(1.)) a_out = layer_a(tf.convert_to_tensor([[1., 2., 3.]])) layer_a.get_weights() [array([[1.], [1.], [1.]], dtype=float32), array([0.], dtype=float32)] layer_b = tf.keras.layers.Dense(1, ... kernel_initializer=tf.constant_initializer(2.)) b_out = layer_b(tf.convert_to_tensor([[10., 20., 30.]])) layer_b.get_weights() [array([[2.], [2.], [2.]], dtype=float32), array([0.], dtype=float32)] layer_b.set_weights(layer_a.get_weights()) layer_b.get_weights() [array([[1.], [1.], [1.]], dtype=float32), array([0.], dtype=float32)]

Returns:

Type	Description
	Weights values as a list of NumPy arrays.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

def get_weights(self):
    """Returns the current weights of the layer, as NumPy arrays.

    The weights of a layer represent the state of the layer. This function
    returns both trainable and non-trainable weight values associated with
    this layer as a list of NumPy arrays, which can in turn be used to load
    state into similarly parameterized layers.

    For example, a `Dense` layer returns a list of two values: the kernel
    matrix and the bias vector. These can be used to set the weights of
    another `Dense` layer:

    >>> layer_a = tf.keras.layers.Dense(1,
    ...   kernel_initializer=tf.constant_initializer(1.))
    >>> a_out = layer_a(tf.convert_to_tensor([[1., 2., 3.]]))
    >>> layer_a.get_weights()
    [array([[1.],
           [1.],
           [1.]], dtype=float32), array([0.], dtype=float32)]
    >>> layer_b = tf.keras.layers.Dense(1,
    ...   kernel_initializer=tf.constant_initializer(2.))
    >>> b_out = layer_b(tf.convert_to_tensor([[10., 20., 30.]]))
    >>> layer_b.get_weights()
    [array([[2.],
           [2.],
           [2.]], dtype=float32), array([0.], dtype=float32)]
    >>> layer_b.set_weights(layer_a.get_weights())
    >>> layer_b.get_weights()
    [array([[1.],
           [1.],
           [1.]], dtype=float32), array([0.], dtype=float32)]

    Returns:
        Weights values as a list of NumPy arrays.
    """
    weights = self.weights
    output_weights = []
    for weight in weights:
        if isinstance(weight, base_layer_utils.TrackableWeightHandler):
            output_weights.extend(weight.get_tensors())
        else:
            output_weights.append(weight)
    return backend.batch_get_value(output_weights)

set_weights(weights)

Sets the weights of the layer, from NumPy arrays.

The weights of a layer represent the state of the layer. This function sets the weight values from numpy arrays. The weight values should be passed in the order they are created by the layer. Note that the layer's weights must be instantiated before calling this function, by calling the layer.

For example, a Dense layer returns a list of two values: the kernel matrix and the bias vector. These can be used to set the weights of another Dense layer:

layer_a = tf.keras.layers.Dense(1, ... kernel_initializer=tf.constant_initializer(1.)) a_out = layer_a(tf.convert_to_tensor([[1., 2., 3.]])) layer_a.get_weights() [array([[1.], [1.], [1.]], dtype=float32), array([0.], dtype=float32)] layer_b = tf.keras.layers.Dense(1, ... kernel_initializer=tf.constant_initializer(2.)) b_out = layer_b(tf.convert_to_tensor([[10., 20., 30.]])) layer_b.get_weights() [array([[2.], [2.], [2.]], dtype=float32), array([0.], dtype=float32)] layer_b.set_weights(layer_a.get_weights()) layer_b.get_weights() [array([[1.], [1.], [1.]], dtype=float32), array([0.], dtype=float32)]

Parameters:

Name	Type	Description	Default
weights		a list of NumPy arrays. The number of arrays and their shape must match number of the dimensions of the weights of the layer (i.e. it should match the output of get_weights).	required

Raises:

Type	Description
ValueError	If the provided weights list does not match the layer's specifications.

Source code in /opt/hostedtoolcache/Python/3.9.16/x64/lib/python3.9/site-packages/keras/engine/base_layer.py

def set_weights(self, weights):
    """Sets the weights of the layer, from NumPy arrays.

    The weights of a layer represent the state of the layer. This function
    sets the weight values from numpy arrays. The weight values should be
    passed in the order they are created by the layer. Note that the layer's
    weights must be instantiated before calling this function, by calling
    the layer.

    For example, a `Dense` layer returns a list of two values: the kernel
    matrix and the bias vector. These can be used to set the weights of
    another `Dense` layer:

    >>> layer_a = tf.keras.layers.Dense(1,
    ...   kernel_initializer=tf.constant_initializer(1.))
    >>> a_out = layer_a(tf.convert_to_tensor([[1., 2., 3.]]))
    >>> layer_a.get_weights()
    [array([[1.],
           [1.],
           [1.]], dtype=float32), array([0.], dtype=float32)]
    >>> layer_b = tf.keras.layers.Dense(1,
    ...   kernel_initializer=tf.constant_initializer(2.))
    >>> b_out = layer_b(tf.convert_to_tensor([[10., 20., 30.]]))
    >>> layer_b.get_weights()
    [array([[2.],
           [2.],
           [2.]], dtype=float32), array([0.], dtype=float32)]
    >>> layer_b.set_weights(layer_a.get_weights())
    >>> layer_b.get_weights()
    [array([[1.],
           [1.],
           [1.]], dtype=float32), array([0.], dtype=float32)]

    Args:
      weights: a list of NumPy arrays. The number
        of arrays and their shape must match
        number of the dimensions of the weights
        of the layer (i.e. it should match the
        output of `get_weights`).

    Raises:
      ValueError: If the provided weights list does not match the
        layer's specifications.
    """
    params = self.weights

    expected_num_weights = 0
    for param in params:
        if isinstance(param, base_layer_utils.TrackableWeightHandler):
            expected_num_weights += param.num_tensors
        else:
            expected_num_weights += 1

    if expected_num_weights != len(weights):
        raise ValueError(
            'You called `set_weights(weights)` on layer "%s" '
            "with a weight list of length %s, but the layer was "
            "expecting %s weights. Provided weights: %s..."
            % (
                self.name,
                len(weights),
                expected_num_weights,
                str(weights)[:50],
            )
        )

    weight_index = 0
    weight_value_tuples = []
    for param in params:
        if isinstance(param, base_layer_utils.TrackableWeightHandler):
            num_tensors = param.num_tensors
            tensors = weights[weight_index : weight_index + num_tensors]
            param.set_weights(tensors)
            weight_index += num_tensors
        else:
            weight = weights[weight_index]
            weight_shape = weight.shape if hasattr(weight, "shape") else ()
            ref_shape = param.shape
            if not ref_shape.is_compatible_with(weight_shape):
                raise ValueError(
                    f"Layer {self.name} weight shape {ref_shape} "
                    "is not compatible with provided weight "
                    f"shape {weight_shape}."
                )
            weight_value_tuples.append((param, weight))
            weight_index += 1

    backend.batch_set_value(weight_value_tuples)

    # Perform any layer defined finalization of the layer state.
    for layer in self._flatten_layers():
        layer.finalize_state()