GANExecutor

Inheritance Diagram

---

class ashpy.losses.gan.GANExecutor(fn=None)

Bases: ashpy.losses.executor.Executor, abc.ABC

Executor for GANs.

Implements the basic functions needed by the GAN losses.

Methods

__init__([fn])	Initialize the GANExecutor.
call(context, **kwargs)	Execute the function, using the information provided by the context.
get_discriminator_inputs(context, …)	Return the discriminator inputs.

Attributes

fn	Return the Keras loss function to execute.
global_batch_size	Global batch size comprises the batch size for each cpu.
weight	Return the loss weight.

__init__(fn=None)

Initialize the GANExecutor.

Parameters

fn (tf.keras.losses.Loss) – A Keras Loss to execute.

Returns

None

abstract call(context, **kwargs)

Execute the function, using the information provided by the context.

Parameters

context (ashpy.contexts.BaseContext) – The function execution Context.

Returns

tf.Tensor – Output Tensor.

static get_discriminator_inputs(context, fake_or_real, condition, training)

Return the discriminator inputs. If needed it uses the encoder.

The current implementation uses the number of inputs to determine whether the discriminator is conditioned or not.

Parameters

• context (ashpy.contexts.gan.GANContext) – Context for GAN models.

• fake_or_real (tf.Tensor) – Discriminator input tensor, it can be fake (generated) or real.

• condition (tf.Tensor) – Discriminator condition (it can also be generator noise).

• training (bool) – whether is training phase or not

Return type

Union[Tensor, List[Tensor]]

Returns

The discriminator inputs.