Generated data from GANGenerator() gives worse result in training for regression task

[diyago]

Generated data from GANGenerator() gives worse result in training for regression task

[Diyago]

Could you please send data stats you pass on to model this way:

df.describe()

[jomariya23156]

Here you are. The "metal_linker", "organic_linker1" and "organic_linker2" are categorical. The last column "CO2_working_capacity" is the target. These are the stats after dropping some NaNs and invalid values and before doing one-hot encoding and feature scaling. I have tried both versions with feature scaling and without feature scaling. Both produced bad newly generated data and decrease the score.

[Diyago]

1. Dummy training I just tried - works, but I suggest you change lightgbm params training:
2. "Accuracy may be bad since you didn't set num_leaves and 2^max_depth > num_leaves" - try setting max_depth = 2 or 3
3. What makes you feel that regression task quality much lower?

[Diyago]

4. Added params for early stopping - gan_params = {"batch_size": 500, "patience": 25, "epochs" : 500,}

Try it out by installing new library version:
!pip install tabgan==1.1.1

[Diyago]

Reopen if problems still persists and provide details for point "GANGenerator in training and it gives worse results for the regression"

Sry for late answer