Tabular Transformer fails to fit

[jcorreia11]

Sometimes when fitting the Tabulat Transformer model it fails and raises:

Node: 'model_1/embedding_1/embedding_lookup' indices[24,5] = 222 is not in [0, 208) [[{{node model_1/embedding_1/embedding_lookup}}]] [Op:__inference_train_function_9384]

tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,

Not sure what is causing that behavior. Maybe something related to the input size after feature selection but not sure. It happens regularly when the featurizer is Mol2Vec but also happens with others.

This should be later addressed.