Semi-Supervised Learning for Tabular Data

This repository contains all data, code, and results we gathered, created, and obtained during the works on our machine-learning project.

The contents of this repository are organized into the following directories:

• data

  • in this directory, you can find the prepared training, validation, and test sets
  • we do not provide the entire original dataset as it can be easily downloaded from UCI ML Repository

• code

  • all code for data preparation and experiments is located here
  • there is a separate file for each tested type of model, including all hyper-parameter configurations
  • you should be able to replicate our experiments with the same results simply by running the respective scripts

• results

  • this directory contains the results from all the conducted experiments, both in a raw text form and in the the form of confusion matrices
  • the results from training on the dataset data/covtype-train.csv are stored in the subdirectory original, while the subdirectory balanced holds the results from training on data/covtype-train-balanced.csv
  • for each set of experiments with one model, there are two pdf files with the above-mentioned confusion matrices, normalized either by the predicted classes (with the suffix prec.pdf) or by the true classes (with the suffix rec.pdf)

Requirements

The code in this directory was run under the following setup:

• python 3.10.12
• pandas 2.1.4
• matplotlib 3.6.0
• numpy 1.26.3
• scipy 1.11.4
• scikit-learn 1.3.2