deeptables 0.1.13

Deep-learning Toolkit for Tabular datasets

DeepTables

DeepTables: Deep-learning Toolkit for Tabular data

DeepTables(DT) is a easy-to-use toolkit that enables deep learning to unleash great power on tabular data.

Overview

MLP (also known as Fully-connected neural networks) have been shown inefficient in learning distribution representation. The "add" operations of the perceptron layer have been proven poor performance to exploring multiplicative feature interactions. In most cases, manual feature engineering is necessary and this work requires extensive domain knowledge and very cumbersome. How learning feature interactions efficiently in neural networks becomes the most important problem.

Various models have been proposed to CTR prediction and continue to outperform existing state-of-the-art approaches to the late years. Well-known examples include FM, DeepFM, Wide&Deep, DCN, PNN, etc. These models can also provide good performance on tabular data under reasonable utilization.

DT aims to utilize the latest research findings to provide users with an end-to-end toolkit on tabular data.

DT has been designed with these key goals in mind:

• Easy to use, non-experts can also use.
• Provide good performance out of the box.
• Flexible architecture and easy expansion by user.

Tutorials

Please refer to the official docs at https://deeptables.readthedocs.io/en/latest/.

• Quick Start
• Examples
• ModelConfig
• Models
• Layers

Installation

pip install deeptables

GPU Setup (Optional)

pip install deeptables[gpu]

Verify the install:

python -c "from deeptables.utils.quicktest import test; test()”

Optional dependencies

Following libraries are not hard dependencies and are not automatically installed when you install DeepTables. To use all functionalities of DT, these optional dependencies must be installed.

pip install shap

Example：

A simple binary classification example

import numpy as np
from deeptables.models import deeptable, deepnets
from deeptables.datasets import dsutils
from sklearn.model_selection import train_test_split

#loading data
df = dsutils.load_bank()
df_train, df_test = train_test_split(df, test_size=0.2, random_state=42)

y = df_train.pop('y')
y_test = df_test.pop('y')

#training
config = deeptable.ModelConfig(nets=deepnets.DeepFM)
dt = deeptable.DeepTable(config=config)
model, history = dt.fit(df_train, y, epochs=10)

#evaluation
result = dt.evaluate(df_test,y_test, batch_size=512, verbose=0)
print(result)

#scoring
preds = dt.predict(df_test)

A solution using DeepTables to win the 1st place in Kaggle Categorical Feature Encoding Challenge II

Click here

DataCanvas

DeepTables is an open source project created by DataCanvas.