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)
- Install latest package from RAPIS.ai with pytorch support.
- Install widedeep libarary
pip install pytorch-widedeep
- 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/
- Ubuntu 20.04 WSL
- Intel i7 12700k
- 32 GB RAM
- Nvidia RTX3090 24GB