/tabulartorch

beyond decision tree

Primary LanguageJupyter Notebook

tabulartorch

Evaluating SOTA library/implementation on tabular data such as:

  • widedeep
  • tabzilla

Evaluation is based on:

  • confusion matrix
  • pytorch with GPU support
  • memory usage (must fit in 24GB vmem GPU)

Installation

  • Install latest package from RAPIS.ai with pytorch support.

How to Run

Widedeep

  • Install widedeep libarary 
    pip install pytorch-widedeep

Tabzilla

  • Clone the repo
  • Install dependencies 
    pip instal openml
  • Download sample dataset from openml. 
    cd tabzilla
python tabzilla_data_preprocessing.py --dataset_name openml__california__361089
  • Run test experiment with GPU config. 
    python ./tabzilla_experiment.py --experiment_config ./tabzilla_experiment_config_gpu.yml --model_name SAINT --dataset_dir ./datasets/openml__california__361089/

Enviroment

  • Ubuntu 20.04 WSL
  • Intel i7 12700k
  • 32 GB RAM
  • Nvidia RTX3090 24GB