This repository contains a small standalone implementation of RealMLP-TD-S, a neural network for tabular datasets, consisting of
- preprocessing code (91 lines of code) in
preprocessing.py, which includes- one-hot encoding with custom missing/unknown value encoding and encoding binary categories to 1/-1
- robust scaling and smooth clipping
- the MLP implementation (211 lines of code) in
mlp.py.
The implementation is standalone in the sense that it only uses
numpy, pandas, sklearn, and torch.
This code is not available on pip since it can just be copied.
RealMLP-TD-S is also available with more functionality in PyTabKit.
The file check_mlp.py
checks that this implementation matches the one in PyTabKit
and requires to install PyTabKit to run.