generated Cov is not that close about gan-for-tabular-data HOT 2 CLOSED

I didnt qet what you would like to achieve. Why is cov so important?

First of all you may with GAN params:

gen_x, gen_y = GANGenerator(bot_filter_quantile=0.001, top_filter_quantile=0.999, is_post_process=True,
           adversarial_model_params={
               "metrics": "AUC", "max_depth": 2, "max_bin": 100, 
               "learning_rate": 0.02, "random_state": 42, "n_estimators": 500,
           }, pregeneration_frac=2, only_generated_data=False,
           gan_params = {"batch_size": 500, "patience": 50, "epochs" : 500,},
           gen_x_times=1.2).generate_data_pipe(train_df, target, test_df)

Secondly:
Your generated data with np.random.multivariate_normal doest have same cov as you passed:
what you wanted
[[1. 0.66666667 0.33333333]
[0.66666667 1. 0.66666667]
[0.33333333 0.66666667 1. ]]
you actually generated
[[0.91722526 0.62890772 0.3325362 ]
[0.62890772 1.04560479 0.74625155]
[0.3325362 0.74625155 1.09697827]]
generated from gan - looks prery similar to me
[[ 1.22316084 0.06813614 -0.02172127]
[ 0.06813614 1.29201178 0.22556574]
[-0.02172127 0.22556574 1.25738251]]

from gan-for-tabular-data.

Sorry for the late answer, if you didnt get required answers, please reopen question

from gan-for-tabular-data.