Model Generation Configuration

In generation, we need to create the generate action corresponding to the selected trained Rockfish model with its configuration.

Basic Configuration Setup

Model	Generate Configuration	Generate Action
RF-Time-GAN	`generate_config = ra.GenerateTimeGAN.Config(doppelganger=ra.GenerateTimeGAN.DGConfig())`	`generate = ra.GenerateTimeGAN(generate_config)`
RF-Time-Transformer	`generate_config = ra.GenerateTimeTransformer.Config()`	`generate = ra.GenerateTimeTransformer(generate_config)`
RF-Tab-GAN	`generate_config = ra.GenerateTabGAN.Config(tabular_gan=ra.GenerateTabGAN.GenerateConfig())`	`generate = ra.GenerateTabGAN(generate_config)`
RF-Tab-Transformer	`generate_config = ra.GenerateTabTransformer.Config()`	`generate = ra.GenerateTabTransformer(generate_config)`

The generation configuration uses default parameter values if not specified.

Available Parameters

Model	Parameters	Description	Default Value	Guidelines
RF-Time-GAN	`given_metadata`	A dict to specify metadata values for conditional generation. If the dict is None, generate all possible metadata values.	`None`	Users can provide `given_metadata` in a dict - keys to indicate field names and values for the values prior to generation
	`clip_per_sample`	When True, it ensures only generating in-range values	`True`	It is effective when `activate_normalization_per_sample = True` in training configuration. When `activate_normalization_per_sample = False`, it by default only generates within range values.
	`sessions`	Number of generated sessions	`None`	If None, it will generate the `min(1000, number of source sessions)`. To generate more, users need to indicate the number in `ra.SessionTarget`. It is only used if users want to generate smaller number than the source number.
RF-Time-Transformer	`sessions`	Number of sessions to generate.	`None`	If None, it will generate the `min(1000, number of source sessions)`. To generate more, users need to indicate the number in `ra.SessionTarget`. It is only used if users want to generate smaller number than the source number.
RF-Tab-Transformer	`records`	Number of records to generate.	`None`	If None, it will generate the `min(1000, number of source records)`. To generate more, users need to indicate the number in `ra.SessionTarget`. It is only used if users want to generate smaller number than the source number.
RF-Tab-GAN	`clip_in_range`	When True, enable generated continuous values to be within the range	`False`	It clips the out-of-range values by the boundary of training data so that it may affect the numerical columns' distribution due to the spikes around the boundary.
	`records`	Number of records to generate.	`None`	If None, it will generate the `min(23,000, number of source records)`. To generate more, users need to indicate the number in `ra.SessionTarget`. It is only used if users want to generate smaller number than the source number.

Usage Examples

You can customize the generation behavior by passing parameters to the configuration objects. Here are some examples:

Note: The sessions and records parameters should only be specified when you want to generate fewer items than the default limit. For generating more than the default, use ra.SessionTarget instead.

For comprehensive information on how to apply the generate action with different use cases, please refer to the Generate Module.

RF-Time-GAN

generate_config = ra.GenerateTimeGAN.Config(
    doppelganger=ra.GenerateTimeGAN.DGConfig(
        given_metadata={'user_type': 'premium'},  # Conditional generation
        clip_per_sample=True,                     # Ensure in-range values
        sessions=500                              # Generate 500 sessions (only when < min(1000, source sessions))
    )
)
generate = ra.GenerateTimeGAN(generate_config)

RF-Time-Transformer

generate_config = ra.GenerateTimeTransformer.Config(
    sessions=200  # Generate 200 sessions (only when < min(1000, source sessions))
)
generate = ra.GenerateTimeTransformer(generate_config)

RF-Tab-GAN

generate_config = ra.GenerateTabGAN.Config(
    tabular_gan=ra.GenerateTabGAN.GenerateConfig(
        clip_in_range=True,  # Clip values to training range
        records=1000         # Generate 1000 records (only when < min(23000, source records))
    )
)
generate = ra.GenerateTabGAN(generate_config)

RF-Tab-Transformer

generate_config = ra.GenerateTabTransformer.Config(
    records=500  # Generate 500 records (only when < min(1000, source records))
)
generate = ra.GenerateTabTransformer(generate_config)