@guthl Yes, the NN’s are struggling in this competition. It took me a week to build a Tensorflow model from scratch and the results were mediocre - Gini values between 0.265 and 0.270.
@mindtrinket and I now have a working implementation based on the ColumnarModelData but a lot of optimization is required before the closing date. The anonymized features don’t help either.
I’m working on as many competitive Kaggle competitions as I can until January. I’ve found it to be the best form of practice and learning. Let me know if you want to team up on the Corporacion competition.