How can we create more synthetic data?

[limhasic]

i tried

child_samples = child_model.sample(
    n_samples = len(child_df),
    input_unique_ids=parent_samples[join_on],
    input_df=parent_samples.drop(join_on, axis=1),
    gen_batch=64,
     )

but, n_samples = len(child_df), wasn't work

[limhasic]

Is this the only alternative?

child_samples = [] for n_child, df in parent_samples.sort_values("n_child").groupby("n_child"): \ print(n_child) _child_samples = child_rtf.sample(input_unique_ids=df[join_on], input_df=df.drop(join_on, axis=1),gen_batch=64) child_samples.append(_child_samples)

[avsolatorio]

Is this the only alternative?

child_samples = [] for n_child, df in parent_samples.sort_values("n_child").groupby("n_child"): \ print(n_child) _child_samples = child_rtf.sample(input_unique_ids=df[join_on], input_df=df.drop(join_on, axis=1),gen_batch=64) child_samples.append(_child_samples)

@limhasic , yes, that looks like a reasonable workaround, but it really depends on the application.

In theory, the number of children given a row from the parent table should be learned by the model. But if your use case does not require constraints over the number of children conditional on the parent, your alternative could work.