Tab Transformer (wip)

Implementation of Tab Transformer, attention network for tabular data, in Pytorch. This simple architecture came within a hair's breadth of GBDT's performance.

Install

$ pip install tab-transformer-pytorch

Usage

import torch
from tab_transformer_pytorch import TabTransformer

cont_mean_std = torch.randn(10, 2)

model = TabTransformer(
    categories = (10, 5, 6, 5, 8),      # tuple containing the number of unique values within each category
    num_continuous = 10,                # number of continuous values
    dim = 32,                           # dimension, paper set at 32
    dim_out = 1,                        # binary prediction, but could be anything
    depth = 6,                          # depth, paper recommended 6
    heads = 8,                          # heads, paper recommends 8
    continuous_mean_std = cont_mean_std # (optional) - normalize the continuous values before layer norm
)

x_categ = torch.randint(0, 5, (1, 5))     # category values, from 0 - max number of categories, in the order as passed into the constructor above
x_cont = torch.randn(1, 10)               # assume continuous values are already normalized individually

pred = model(x_categ, x_cont)

Citations

@misc{huang2020tabtransformer,
    title={TabTransformer: Tabular Data Modeling Using Contextual Embeddings}, 
    author={Xin Huang and Ashish Khetan and Milan Cvitkovic and Zohar Karnin},
    year={2020},
    eprint={2012.06678},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}

GitHub