new_metric_class: Define new keras types

Description

These functions can be used to make custom objects that fit in the family of existing keras types. For example, new_layer_class() will return a class constructor, an object that behaves like other layer functions such as layer_dense(). new_callback_class() will return an object that behaves similarly to other callback functions, like callback_reduce_lr_on_plateau(), and so on. All arguments with a default NULL value are optional methods that can be provided.

Usage

new_metric_class(classname, ..., initialize, update_state, result)
new_loss_class(classname, ..., call = NULL)
new_callback_class(
  classname,
  ...,
  on_epoch_begin = NULL,
  on_epoch_end = NULL,
  on_train_begin = NULL,
  on_train_end = NULL,
  on_batch_begin = NULL,
  on_batch_end = NULL,
  on_predict_batch_begin = NULL,
  on_predict_batch_end = NULL,
  on_predict_begin = NULL,
  on_predict_end = NULL,
  on_test_batch_begin = NULL,
  on_test_batch_end = NULL,
  on_test_begin = NULL,
  on_test_end = NULL,
  on_train_batch_begin = NULL,
  on_train_batch_end = NULL
)
new_model_class(
  classname,
  ...,
  initialize = NULL,
  call = NULL,
  train_step = NULL,
  predict_step = NULL,
  test_step = NULL,
  compute_loss = NULL,
  compute_metrics = NULL
)
new_layer_class(
  classname,
  ...,
  initialize = NULL,
  build = NULL,
  call = NULL,
  get_config = NULL
)
mark_active(x)

Value

A new class generator object that inherits from the appropriate Keras base class.

Arguments

classname: The classname as a string. Convention is for the classname to be a CamelCase version of the constructor.
...: Additional fields and methods for the new type.
initialize, build, call, get_config, on_epoch_begin, on_epoch_end, on_train_begin, on_train_end, on_batch_begin, on_batch_end, on_predict_batch_begin, on_predict_batch_end, on_predict_begin, on_predict_end, on_test_batch_begin, on_test_batch_end, on_test_begin, on_test_end, on_train_batch_begin, on_train_batch_end, update_state, result, train_step, predict_step, test_step, compute_loss, compute_metrics: Optional methods that can be overridden.
x: A function that should be converted to an active property of the class type.

Details

mark_active() is a decorator that can be used to indicate functions that should become active properties of the class instances.