How to configure the dim sizes and learning rates?

[vinay-k12]

I'm trying to customize the Gen/Des dim sizes, learning rates and other params in the models. I'm using the same API as SDV.

import pandas as pd
from gretel_synthetics.actgan import ACTGAN
model = ACTGAN(
verbose=True,
binary_encoder_cutoff=10,
auto_transform_datetimes=True,
epochs=100,
epoch_callback=epoch_tracker.add,
generator_dim=(1024, 1024),
discriminator_dim=(1024, 1024),
generator_lr=0.0000001, generator_decay=1e-05, discriminator_lr=0.000001, discriminator_decay=1e-06,
)

This worked. Does this mean I can use the same API dictionary as SDV? And can I load models like in SDV (conditions, seeing etc)?

[johntmyers]

Yes most of the functionality of CTGAN is supported with ACTGAN, you can see the supported parameter list is just about the same, plus the ones we added:

https://synthetics.docs.gretel.ai/en/stable/models/actgan.html

[vinay-k12]

That's great. But I notice that there is not enough info about 'generate' like 'Conditional' or 'Seeding with columns' etc. Are there any references for the same?

[johntmyers]

It should support the same methods as CTGAN. So methods like sample_remaining_columns are all supported.

[vinay-k12]

I tried doing conditional sampling (Column B === 'Value'), but it didn't work.

[johntmyers]

Can you be more specific on what did not work? Did you receive errors that you can share?