AutoX

This is the documentation of AutoX.

AutoX is a python package. It’s an open source AutoML solution for tabular data.

You can jump right into the package by looking into our Quick Start.

Installation

$ git clone https://github.com/4paradigm/autox.git
$ pip install ./autox

Contents

The following chapters will explain the AutoX package in detail:

Introduction

What is AutoX?

AutoX is an efficient automl tool, mainly aimed at data mining competitions with tabular data. Its features include:

SOTA: AutoX outperforms other solutions in many datasets(see performance improvement under different tasks).

Easy to use: The design of interfaces is similar to sklearn.

Generic & Universal: Supporting tabular data, including binary classification, multi-class classification and regression problems.

Auto: Fully automated pipeline without human-intervention.

Out of the box: Providing flexible modules which can be used alone.

Summary of magics: Organize and publish magics of competitions.

What’s Included in AutoX?

AutoML Competition

autox_competition is AutoML Solutions for Competition.

Demo

data	description	link
Elo	Help understand customer loyalty	autox_elo
Rossmann	Forecast sales using store, promotion, and competitor data	autox_Rossmann
Allstate	How severe is an insurance claim?	autox_Allstate
House Prices	Predict house sales prices	autox_house_price
IEEE	Can you detect fraud from customer transactions?	autox_ieee
springleaf	Forecast sales using store, promotion, and competitor data	autox_springleaf
stumbleupon	Forecast sales using store, promotion, and competitor data	autox_stumbleupon
ventilator	Forecast sales using store, promotion, and competitor data	autox_ventilator
walmart	Use historical markdown data to predict store sales	autox_walmart

Pipeline

API

feature engineer

count features

class autox.autox_competition.feature_engineer.fe_count.FeatureCount[source]

Convert categorical features into the number of occurrences.

fit(df, degree=1, target=None, df_feature_type=None, silence_cols=[], select_all=True, max_num=None)[source]

Parameters:

df – dataframe, train_test.
degree – int, degree equal to 1 or 2.
target – str, target column.
df_feature_type – dict, {col: type of col}.
silence_cols –
select_all –
max_num –

transform(df)[source]

Parameters:: df – dataframe, train_test.
Returns:: dataframe, count features.

cross features

class autox.autox_competition.feature_engineer.fe_cross.FeatureCross(importance_type='split')[source]

synthetic feature formed by multiplying (crossing) two features.

fit(X, y, objective, category_cols, top_k=10, used_cols=[])[source]

Parameters:

X – {array-like, sparse matrix} of shape (n_samples, n_features). Training vector, where n_samples is the number of samples and n_features is the number of features.
y – array-like of shape (n_samples,). Target vector relative to X.
objective – str, objective equal to ‘binary’ or ‘regression’.
category_cols – list, column names of categorical features.
top_k – int, keep the top_k importance cross features, default top_k = 10.
used_cols – list, columns will be used for training model, default top_k = 10.

transform(X)[source]

Parameters:: X – {array-like, sparse matrix} of shape (n_samples, n_features). Training vector, where n_samples is the number of samples and n_features is the number of features.
Returns:: dataframe, cross features.

cumsum features

class autox.autox_competition.feature_engineer.fe_cumsum.FeatureCumsum[source]: cumsum特征描述

denoising autoencoder features

class autox.autox_competition.feature_engineer.fe_denoising_autoencoder.FeatureDenoisingAutoencoder[source]: DenoisingAutoencoder特征描述

diff features

class autox.autox_competition.feature_engineer.fe_diff.FeatureDiff[source]: diff特征描述

dimension reduction features

exp weighted mean features

gbdt features

image to vector features

nlp features

features from other table (one to many relationship)

rank features

rolling statistics features (for time-series data)

shift features

shift features (for time-series data)

statistics features

target encoding features

time features

feature selection

adversarial validation

class autox.autox_competition.feature_selection.adversarial_validation.AdversarialValidation[source]

Bases: object

Remove features with inconsistent distribution between train and test.

Example::: elo_AdversarialValidation_AutoX

fit(train, test, id_, target, categorical_features=[], p=0.6)[source]

Parameters:

train – dataframe, the training input samples.
test – dataframe, the testing input samples.
id – list, columns as id.
target – str, target column.
categorical_features – list, columns with categorical type.
p – float, threshold. If the auc is greater than this threshold, the algorithm will continuously remove the most important feature.

transform(df)[source]

Parameters:: df – dataframe, dataframe needs to be transformed.
Returns:: dataframe, transformed dataframe.

GRN feature selection

Each feature weight is output according to the feature column definition.

Example::: GRN_FeatureSelection_AutoX

Metrics

param y_true:: array-like of shape (n_samples,) or (n_samples, n_outputs). Ground truth (correct) target values.
param y_pred:: array-like of shape (n_samples,) or (n_samples, n_outputs). Estimated target values.
param metric:: str, one of [‘mae’, ‘mape’, ‘mse’, ‘rmse’, ‘msle’, ‘rmsle’, ‘smape’], default = ‘mape’.
return:: metric.

Overview of feature engineer API
operation	description
count	count the occurrences of some categorical features within dataset.
cumsum	the calculation of the cumulative sum.
denoising autoencoder	train a denoising autoencoder neural network for feature extraction. reference
dimension reduction	use dimension reduction technology for feature extraction, such as Principal Component Analysis (PCA).
gbdt	Generating Features with Gradient Boosted Decision Trees. reference
rank	Compute numerical data ranks.
rolling	statistics calculation within rolling windows.
shift	lag feature.
diff	first Difference.
statistics	statistics calculation.
time parse feature	parse time feature for time column, such as year, month, day, hour, dayofweek, and so on.
cross feature	synthetic feature formed by multiplying (crossing) two or more features. reference

Overview of feature selection API
operation	description
Adversarial Validation	a feature selection solution for battling overfitting. reference
GRN	a feature selection using Gated Residual Networks (GRN) and Variable Selection Networks (VSN). reference

Overview of Metrics API
operation	description
MAE	mean absolute error
MAPE	mean absolute percentage error
MSE	mean squared error
MSLE	mean squared logarithmic error
RMSLE	root mean squared logarithmic error

AutoML Server

autox_server is AutoML Solutions for Development.

Demo

CASE 1: Customer Loan Risk Prediction

description: Given the user’s basic information, consumption behavior, repayment situation, etc., an accurate overdue prediction model is established to predict whether the user will overdue repayment.

data download link: google cloud

data details: link

autox_server training: bank_train.ipynb

autox_server inference: bank_test.ipynb

Pipeline

AutoML Interpreter

autox_interpreter is AutoML Solutions for Machine Learning interpretation.

AutoX covers following interpretable machine learning methods:

Model-based interpretation

nn model interpretation, see nn_interpret

light model interpretation, see lgb_interpret

lr model interpretation, see lr_interpret

Golbel interpretation

tree-based model, see global_surrogate_tree_demo

Local interpretation

LIME, see lime_demo

SHAP, see shap_demo

Influential interpretation

nn, see influential_interpretation_nn

nn_sgd, see influential_interpretation_nn_sgd

Prototypes and Criticisms

MMD-critic, see MMD_demo

ProtoDash algorithm, see ProtodashExplainer

Quick Start

Install AutoX

As the compiled autox package is hosted on the Python Package Index (PyPI) you can easily install it with pip

pip install automl-x -i https://www.pypi.org/simple/

or

$ git clone https://github.com/4paradigm/autox.git
$ pip install ./autox

Before boring yourself by reading the docs in detail, you can dive right into AutoX with the following examples:

Binary classification example(Transaction Prediction)

We are provided with an anonymized dataset containing numeric feature variables, the binary target column, and a string ID_code column. The task is to predict the value of target column in the test set.

print(train.head())

print(test.head())

print(train.shape(), test.shape())

We build the automl pipeline with AutoX as following:

from autox import AutoX
path = f'../input/santander-customer-transaction-prediction'
autox = AutoX(target = 'target', train_name = 'train.csv', test_name = 'test.csv', id = ['ID_code'], path = path)
sub = autox.get_submit()
sub.to_csv("./autox_Santander.csv", index = False)

We get a pandas.DataFrame sub which has the same number of rows as test.

print(sub.shape(), test.shape())

print(sub.head())

You can execute this example with this link: santander-autox.

Regression example(House Prices)

With 79 explanatory variables describing (almost) every aspect of residential homes in Ames, Iowa, we need predict the SalePrice of each home.

print(train.head())

print(test.head())

print(train.shape(), test.shape())

We build the automl pipeline with AutoX as following:

from autox import AutoX
path = '../input/house-prices-advanced-regression-techniques'
autox = AutoX(target = 'SalePrice', train_name = 'train.csv', test_name = 'test.csv', id = ['Id'], path = path)
sub = autox.get_submit()
sub.to_csv("submission.csv", index = False)

We get a pandas.DataFrame sub which has the same number of rows as test.

print(sub.shape(), test.shape())

print(sub.head())

You can execute this example with this link: house_price-autox.

Community

Welcome to join our community.

Achievement

performance improvement under different tasks

data_type	compare with AutoGluon	compare with H2O
binary classification	+20.44%	+2.98%
regression	+37.54%	+39.66%
time-series	+28.40%	+32.46%

results comparison

data_type	single-or-multi	data_name	metric	AutoX	AutoGluon	H2o
binary classification	single-table	Springleaf	auc	0.78865	0.61141	0.78186
binary classification	single-table	stumbleupon	auc	0.87177	0.81025	0.79039
binary classification	single-table	santander	auc	0.89196	0.64643	0.88775
binary classification	multi-table	IEEE	accuracy	0.920809	0.724925	0.907818
regression	single-table	ventilator	mae	0.755	8.434	4.221
regression	single-table	Allstate Claims Severity	mae	1137.07885	1173.35917	1163.12014
regression	single-table	zhidemai	mse	1.0034	1.9466	1.1927
regression	single-table	Tabular Playground Series - Aug 2021	rmse	7.87731	10.3944	7.8895
regression	single-table	House Prices	rmse	0.13043	0.13104	0.13161
regression	single-table	Restaurant Revenue	rmse	2133204.32146	31913829.59876	28958013.69639
regression	multi-table	Elo Merchant Category Recommendation	rmse	3.72228	3.80801	22.88899
regression-ts	single-table	Demand Forecasting	smape	13.79241	25.39182	18.89678
regression-ts	multi-table	Walmart Recruiting	wmae	4660.99174	5024.16179	5128.31622
regression-ts	multi-table	Rossmann Store Sales	RMSPE	0.13850	0.20453	0.35757

competition

First place solution for Alibaba Cloud Infrastructure Supply Chain Competition

Enterprise support

值得买

慕尚

How to contribute

We want AutoX to become a leading international AutoML solution. To achieve this goal, we need your help!

All contributions, bug reports, bug fixes, documentation improvements, enhancements and ideas are welcome.

If you want to help, just create a pull request on our github page. To the new user, working with Git can sometimes be confusing and frustrating. If you are not familiar with Git you can also contact us by email.

We are looking forward to hear from you! =)

FAQ

How can I use autox with windows?

We recommend to use Anaconda. After installing, open the Anaconda Prompt, create an environment and set up AutoX (Please be aware that we’re using multiprocessing, which can be problematic.):
conda create -n ENV_NAME python=VERSION
activate ENV_NAME
pip install autox

AutoX

Installation

Contents

Introduction

What is AutoX?

What’s Included in AutoX?

AutoML Competition

Demo

Pipeline

API

feature engineer

count features

cross features

cumsum features

denoising autoencoder features

diff features

dimension reduction features

exp weighted mean features

gbdt features

image to vector features

nlp features

features from other table (one to many relationship)

rank features

rolling statistics features (for time-series data)

shift features

shift features (for time-series data)

statistics features

target encoding features

time features

feature selection

adversarial validation

GRN feature selection

Metrics

AutoML Server

Demo

CASE 1: Customer Loan Risk Prediction

Pipeline

AutoML Interpreter

Quick Start

Install AutoX

Binary classification example(Transaction Prediction)

Regression example(House Prices)

Community

Achievement

performance improvement under different tasks

results comparison

competition

Enterprise support

How to contribute

FAQ

Authors

Core Development Team

Contributions

Indices and tables

Acknowledgements