Revisiting Tabular Deep Learning

This repository contains:

• the official implementation of the paper "Revisiting Deep Learning Models for Tabular Data" (link)
• rtdl (Revisiting Tabular Deep Learning):
  • It is a PyTorch-based package that provides a user-friendly API for the main models (FT-Transformer, ResNet, MLP) used in the paper
  • It can be used by practitioners looking for Deep Learning models for tabular data
  • It can serve as a source of baselines for researchers (excluding FT-Transformer, see the warning below)
  • You can follow releases by hitting "Watch" / "Custom" / "Releases" in the right upper corner of the GitHub interface
  • See the website for more details
  • See the discussion

Warning: if you are a researcher (not a practitioner) and plan to use the FT-Transformer model as a baseline in your paper, please, use the original implementation from ft_transformer.py. We will remove this limitation soon (i.e. rtdl will become the recommended way to use FT-Transformer in papers).

The rest of this document is dedicated to the implementation of the paper.

---

Note that the paper reports results based on thousands of experiments, so there can be rough edges in the implementation. Feel free to open issues and ask questions in discussions.

• 1. The main results
• 2. Overview
• 3. Setup the environment
  • 3.1. PyTorch environment
  • 3.2. TensorFlow environment
  • 3.3. Data
• 4. Tutorial (how to reproduce results)
  • 4.1. Check the environment
  • 4.2. Tuning
  • 4.3. Evaluation
  • 4.4. Ensembling
  • 4.5. "Visualize" results
  • 4.6. What about other models and datasets?
• 5. How to work with the repository
  • 5.1. How to run scripts
  • 5.2. stats.json and other results
  • 5.3. Conclusion
• 6. How to cite

1. The main results

The main table, where all models are sorted by their final (test) performance on all datasets, can be found in the last cell of this notebook.

2. Overview

The code is organized as follows:

• bin:
  • training code for all the models
  • ensemble.py performs ensembling
  • tune.py tunes models
  • report.ipynb summarizes all the results
  • create_tables.ipynb builds Table 1 and Table 2 from the paper
  • code for the section "When FT-Transformer is better than ResNet?" of the paper:
    • analysis_gbdt_vs_nn.py runs the experiments
    • create_synthetic_data_plots.py builds plots
• lib contains common tools used by programs in bin
• output contains configuration files (inputs for programs in bin) and results (metrics, tuned configurations, etc.)
• the remaining files and directories are mostly related to the rtdl package

The results are represented with numerous JSON files that are scatterd all over the output directory. The notebook bin/report.ipynb summarizes them into a report (a much more detailed version of Table 1 and Table 2 from the paper). Use this notebook to:

• build a detailed report where all the models are compared on all datasets
• get insights about test, validation and train scores

3. Setup the environment

3.1. PyTorch environment

Install conda

export REPO_DIR=<ABSOLUTE path to the desired repository directory>
git clone <repository url> $REPO_DIR
cd $REPO_DIR

conda create -n rtdl python=3.8.8
conda activate rtdl

conda install pytorch==1.7.1 torchvision==0.8.2 cudatoolkit=10.1.243 numpy=1.19.2 -c pytorch -y
conda install cudnn=7.6.5 -c anaconda -y
pip install -r requirements.txt
conda install -c conda-forge nodejs -y
jupyter labextension install @jupyter-widgets/jupyterlab-manager

# if the following commands do not succeed, update conda
conda env config vars set PYTHONPATH=${PYTHONPATH}:${REPO_DIR}
conda env config vars set PROJECT_DIR=${REPO_DIR}
conda env config vars set LD_LIBRARY_PATH=${CONDA_PREFIX}/lib:${LD_LIBRARY_PATH}
conda env config vars set CUDA_HOME=${CONDA_PREFIX}
conda env config vars set CUDA_ROOT=${CONDA_PREFIX}

conda deactivate
conda activate rtdl

3.2. TensorFlow environment

This environment is needed only for experimenting with TabNet. For all other cases use the PyTorch environment.

The instructions are the same as for the PyTorch environment (including installation of PyTorch!), but:

• python=3.7.10
• cudatoolkit=10.0
• right before pip install -r requirements.txt do the following:
  • pip install tensorflow-gpu==1.14
  • comment out tensorboard in requirements.txt

3.3. Data

LICENSE: by downloading our dataset you accept licenses of all its components. We do not impose any new restrictions in addition to those licenses. You can find the list of sources in the section "References" of our paper.

1. Download the data: wget https://www.dropbox.com/s/o53umyg6mn3zhxy/rtdl_data.tar.gz?dl=1
2. Move the archive to the root of the repository: mv rtdl_data.tar.gz $PROJECT_DIR
3. Go to the root of the repository: cd $PROJECT_DIR
4. Unpack the archive: tar -xvf rtdl_data.tar.gz

4. Tutorial (how to reproduce results)

This section only provides specific commands with few comments. After completing the tutorial, we recommend checking the next section for better understanding of how to work with the repository. It will also help to better understand the tutorial.

In this tutorial, we will reproduce the results for MLP on the California Housing dataset. We will cover:

• tuning
• evaluation
• ensembling
• comparing models with each other

Note that the chances to get exactly the same results are rather low, however, they should not differ much from ours. Before running anything, go to the root of the repository and explicitly set CUDA_VISIBLE_DEVICES (if you plan to use GPU):

cd $PROJECT_DIR
export CUDA_VISIBLE_DEVICES=0

4.1. Check the environment

Before we start, let's check that the environment is configured successfully. The following commands should train one MLP on the California Housing dataset:

mkdir draft
cp output/california_housing/mlp/tuned/0.toml draft/check_environment.toml
python bin/mlp.py draft/check_environment.toml

The result should be in the directory draft/check_environment. For now, the content of the result is not important.

4.2. Tuning

Our config for tuning MLP on the California Housing dataset is located at output/california_housing/mlp/tuning/0.toml. In order to reproduce the tuning, copy our config and run your tuning:

# you can choose any other name instead of "reproduced.toml"; it is better to keep this
# name while completing the tutorial
cp output/california_housing/mlp/tuning/0.toml output/california_housing/mlp/tuning/reproduced.toml
# let's reduce the number of tuning iterations to make tuning fast (and ineffective)
python -c "
from pathlib import Path
p = Path('output/california_housing/mlp/tuning/reproduced.toml')
p.write_text(p.read_text().replace('n_trials = 100', 'n_trials = 5'))
"
python bin/tune.py output/california_housing/mlp/tuning/reproduced.toml

The result of your tuning will be located at output/california_housing/mlp/tuning/reproduced, you can compare it with ours: output/california_housing/mlp/tuning/0. The file best.toml contains the best configuration that we will evaluate in the next section.

4.3. Evaluation

Now we have to evaluate the tuned configuration with 15 different random seeds.

# create a directory for evaluation
mkdir -p output/california_housing/mlp/tuned_reproduced

# clone the best config from the tuning stage with 15 different random seeds
python -c "
for seed in range(15):
    open(f'output/california_housing/mlp/tuned_reproduced/{seed}.toml', 'w').write(
        open('output/california_housing/mlp/tuning/reproduced/best.toml').read().replace('seed = 0', f'seed = {seed}')
    )
"

# train MLP with all 15 configs
for seed in {0..14}
do
    python bin/mlp.py output/california_housing/mlp/tuned_reproduced/${seed}.toml
done

Our directory with evaluation results is located right next to yours, namely, at output/california_housing/mlp/tuned.

4.4. Ensembling

# just run this single command
python bin/ensemble.py mlp output/california_housing/mlp/tuned_reproduced

Your results will be located at output/california_housing/mlp/tuned_reproduced_ensemble, you can compare it with ours: output/california_housing/mlp/tuned_ensemble.

4.5. "Visualize" results

Use bin/report.ipynb:

• find the cell with the list of results; the list includes many lines of this kind: ('algorithm/experiment', N_SEEDS, ENSEMBLES_3_5, 'PrettyAlgorithmName', datasets)
• uncomment the line relevant to the tutorial; it should look like this: ('mlp/tuned_reproduced', N_SEEDS, ENSEMBLES_3_5, 'MLP | reproduced', [CALIFORNIA]),
• run the notebook from scratch

4.6. What about other models and datasets?

Similar steps can be performed for all models and datasets. The tuning process is slightly different in the case of grid search: you have to run all desired configurations and manually choose the best one based on the validation performance. For example, see output/epsilon/ft_transformer.

5. How to work with the repository

5.1. How to run scripts

You should run Python scripts from the root of the repository. Most programs expect a configuration file as the only argument. The output will be a directory with the same name as the config, but without the extention. Configs are written in TOML. The lists of possible arguments for the programs are not provided and should be inferred from scripts (usually, the config is represented with the args variable in scripts). If you want to use CUDA, you must explicitly set the CUDA_VISIBLE_DEVICES environment variable. For example:

# The result will be at "path/to/my_experiment"
CUDA_VISIBLE_DEVICES=0 python bin/mlp.py path/to/my_experiment.toml

# The following example will run WITHOUT CUDA
python bin/mlp.py path/to/my_experiment.toml

If you are going to use CUDA all the time, you can save the environment variable in the Conda environment:

conda env config vars set CUDA_VISIBLE_DEVICES="0"

The -f (--force) option will remove the existing results and run the script from scratch:

python bin/whatever.py path/to/config.toml -f  # rewrites path/to/config

bin/tune.py supports continuation:

python bin/tune.py path/to/config.toml --continue

5.2. stats.json and other results

For all scripts, stats.json is the most important part of output. The content varies from program to program. It can contain:

• metrics
• config that was passed to the program
• hardware info
• execution time
• and other information

Predictions for train, validation and test sets are usually also saved.

5.3. Conclusion

Now, you know everything you need to reproduce all the results and extend this repository for your needs. The tutorial also should be more clear now. Feel free to open issues and ask questions.

6. How to cite

@article{gorishniy2021revisiting,
    title={Revisiting Deep Learning Models for Tabular Data},
    author={Yury Gorishniy and Ivan Rubachev and Valentin Khrulkov and Artem Babenko},
    journal={arXiv},
    volume={2106.11959},
    year={2021},
}