model_variable

tfsnippet.utils.model_variable(name, shape=None, dtype=None, initializer=None, regularizer=None, constraint=None, trainable=True, collections=None, **kwargs)

Get or create a model variable.

When the variable is created, it will be added to both GLOBAL_VARIABLES and MODEL_VARIABLES collection.

Parameters:

• name – Name of the variable.
• shape – Shape of the variable.
• dtype – Data type of the variable.
• initializer – Initializer of the variable.
• regularizer – Regularizer of the variable.
• constraint – Constraint of the variable.
• trainable (bool) – Whether or not the variable is trainable?
• collections – In addition to GLOBAL_VARIABLES and MODEL_VARIABLES, also add the variable to these collections.
• **kwargs – Other named arguments passed to tf.get_variable().

Returns:

The variable.

Return type:

tf.Variable