Layer weight constraints

[source]

MaxNorm class

tf_keras.constraints.MaxNorm(max_value=2, axis=0)

MaxNorm weight constraint.

Constrains the weights incident to each hidden unit to have a norm less than or equal to a desired value.

Also available via the shortcut function tf.keras.constraints.max_norm.

Arguments

• max_value: the maximum norm value for the incoming weights.
• axis: integer, axis along which to calculate weight norms. For instance, in a Dense layer the weight matrix has shape (input_dim, output_dim), set axis to 0 to constrain each weight vector of length (input_dim,). In a Conv2D layer with data_format="channels_last", the weight tensor has shape (rows, cols, input_depth, output_depth), set axis to [0, 1, 2] to constrain the weights of each filter tensor of size (rows, cols, input_depth).

[source]

MinMaxNorm class

tf_keras.constraints.MinMaxNorm(min_value=0.0, max_value=1.0, rate=1.0, axis=0)

MinMaxNorm weight constraint.

Constrains the weights incident to each hidden unit to have the norm between a lower bound and an upper bound.

Also available via the shortcut function tf.keras.constraints.min_max_norm.

Arguments

• min_value: the minimum norm for the incoming weights.
• max_value: the maximum norm for the incoming weights.
• rate: rate for enforcing the constraint: weights will be rescaled to yield (1 - rate) * norm + rate * norm.clip(min_value, max_value). Effectively, this means that rate=1.0 stands for strict enforcement of the constraint, while rate<1.0 means that weights will be rescaled at each step to slowly move towards a value inside the desired interval.
• axis: integer, axis along which to calculate weight norms. For instance, in a Dense layer the weight matrix has shape (input_dim, output_dim), set axis to 0 to constrain each weight vector of length (input_dim,). In a Conv2D layer with data_format="channels_last", the weight tensor has shape (rows, cols, input_depth, output_depth), set axis to [0, 1, 2] to constrain the weights of each filter tensor of size (rows, cols, input_depth).

[source]

NonNeg class

tf_keras.constraints.NonNeg()

Constrains the weights to be non-negative.

Also available via the shortcut function tf.keras.constraints.non_neg.

[source]

UnitNorm class

tf_keras.constraints.UnitNorm(axis=0)

Constrains the weights incident to each hidden unit to have unit norm.

Also available via the shortcut function tf.keras.constraints.unit_norm.

Arguments

• axis: integer, axis along which to calculate weight norms. For instance, in a Dense layer the weight matrix has shape (input_dim, output_dim), set axis to 0 to constrain each weight vector of length (input_dim,). In a Conv2D layer with data_format="channels_last", the weight tensor has shape (rows, cols, input_depth, output_depth), set axis to [0, 1, 2] to constrain the weights of each filter tensor of size (rows, cols, input_depth).

[source]

RadialConstraint class

tf_keras.constraints.RadialConstraint()

Constrains Conv2D kernel weights to be the same for each radius.

Also available via the shortcut function tf.keras.constraints.radial_constraint.

For example, the desired output for the following 4-by-4 kernel:

    kernel = [[v_00, v_01, v_02, v_03],
              [v_10, v_11, v_12, v_13],
              [v_20, v_21, v_22, v_23],
              [v_30, v_31, v_32, v_33]]

is this::

    kernel = [[v_11, v_11, v_11, v_11],
              [v_11, v_33, v_33, v_11],
              [v_11, v_33, v_33, v_11],
              [v_11, v_11, v_11, v_11]]

This constraint can be applied to any Conv2D layer version, including Conv2DTranspose and SeparableConv2D, and with either "channels_last" or "channels_first" data format. The method assumes the weight tensor is of shape (rows, cols, input_depth, output_depth).