Documentation question regarding N_d and N_a

[mbernico]

In the TabNet docstring it says

"""
- Adjustment of the values of Nd and Na is the most efficient way of obtaining a trade-off
between performance and complexity. Nd = Na is a reasonable choice for most datasets.
"""

However tabnet line 128 is:

"""
if feature_dim <= output_dim:
raise ValueError("To compute features_for_coef, feature_dim must be larger than output dim")
"""

Wondering if this input validation is in error, or if documentation should be adjusted. Happy to PR the fix, I just don't know which way it should go.

[titu1994]

The assertion is required otherwise tensor flow will index 0dim tensors. The doc can be updated in a pr

[mbernico]

Got it, okay, will do.

[Kipkull]

move this line 170 in tabnet.py: features_for_coef = transform_f4[:, self.output_dim:]
into "if" of line 172, would save the 0dim problem