Usage

Acquire Data

The first step is to acquire data from OpenML.

import openml as oml
import datacleanbot.dataclean as dc
import numpy as np

data = oml.datasets.get_dataset(id) # id: openml dataset id
X, y, categorical_indicator, features = data.get_data(target=data.default_target_attribute, dataset_format='array')
Xy = np.concatenate((X,y.reshape((y.shape[0],1))), axis=1)

Show Impotant Features

datacleanbot computes the most important features of the given dataset using random forest and present the 15 most useful features to the user.

dc.show_important_features(X, y, data.name, features)

Unify Column Names

Inconsistent capitalization of column names can be detected and reported to the user. Users can decide whether to unify them or not. The capitalization can be unified to either upper case or lower case.

dc.unify_name_consistency(features)

Show Statistical Inforamtion

datacleanbot can present the statistical information to help users gain a better understanding of the data distribution.

dc.show_statistical_info(Xy)

Discover Data Types

datacleanbot can discover feature data types. Basic data types discovered are ‘datetime’, ‘float’, ‘integer’, ‘bool’ and ‘string’. datacleanbot also can discover statistical data types (real, positive real, categorical and count) using Bayesian Model abda.

dc.discover_types(Xy)

Clean Duplicated Rows

datacleanbot detects the duplicated records and reports them to users.

dc.clean_duplicated_rows(Xy)

Handle Missing Values

datacleanbot identifies characters ‘n/a’, ‘na’, ‘–’ and ‘?’ as missing values. Users can add extra characters to be considered as missing. After the missing values being detected, datacleanbot will present the missing values in effective visualizations to help users identify the missing mechanism. Afterwards, datacleanbot recommends the appropriate approach to clean missing values according the missing mechanism.

features, Xy = dc.handle_missing(features, Xy)

Outlier Detection

A meta-learner is trained beforehand to recommend the outlier detection algorithm according to the meta features og the given dataset. Users can apply the recommended algorithm or any other available algorithm to detect outliers. After the detection, outliers will be present to users in effective visualizations and users can choose to drop them or not.

Xy = dc.handle_outlier(features, Xy)