ZAPS: Pythonic Exploratory Data Analysis Framework

ZAPS is a lightweight, low-code Python wrapper designed to simplify and accelerate the exploratory data analysis (EDA) process. Built on top of industry-standard libraries, it provides an intuitive and efficient framework for data inspection, visualization, and preparation.

With ZAPS, you can quickly and easily perform a wide range of EDA tasks, without the need for complex code or extensive programming expertise; allowing you to focus on insights and decision-making rather than tedious data manipulation, all for unlocking deeper understanding and actionable insights from your data.

Table of Contents

• Main Features
• Installation
• Dependencies
• Quickstart
• Tutorials
• License
• API Documentation
• Contributing to ZAPS

Main Features

• Combined capabilities of multiple modules in a concise single interface
• Highlighting potential underlying problems and summarizing input data
• Rich tabular reports and highly customizable interactive visualizations
• Statistical modeling with commonly used algorithms and metrics
• Flexible alternation between numeric and categorical data
• User friendly error handling and informative workflow
• Easy integration with Scikit-learn Pipeline

Installation

ZAPS is tested and supported on 64-bit systems with:

• Python 3.9 or latter
• Debian 12
• Windows 8.1 or later

You can install ZAPS with Python's pip package manager:

# install zaps
pip install zaps

Dependencies

• Numpy
• Pandas
• Seaborn
• Matplotlib
• Plotly
• Statsmodels
• Scipy
• Scikit-learn
• Distfit

Quickstart

Analysing and highlighting data problems

import pandas as pd
from zaps.eda import UniStat, Dist, Olrs, NumAna, CatAna

# loading sample dataset
df = pd.read_csv('...')

# univariate stats - highlighting skewed numeric features and rare categories and  
u_s = UniStat(df)

num_cols, cat_cols, dup_df = u_s.peek()

# visualizing data problems
u_s.stats_plot()

# skewed features goodness of fit: Normal distribution
u_s.skew_plot()

Visualizing distributions

# plotting numeric distributions with no user input
dsts = Dist(df = df, cols = num_cols, silent = True)

# Histograms iterative plotting 
dsts.hs()

# analysing best fitting distribution
dsts.best_fit(
            cols = num_cols,
            distr = ['norm', 'expon', 'lognorm', 'uniform']
          )

# visualize best fitting distribution
dsts.best_vis()

Identifying and handling outliers

# outliers: identifying and capping using different methods
lrs = Olrs(num_cols,
		   mapping = {'column_name': ('iqr', 1.5)},
		   method = 'q',
           hide_p_bar = True
           )

trans_df = lrs.fit_transform(df)

Numeric features analysis

# numeric analysis - fitting regression models and displaying results
n_a = NumAna(df, num_cols, 'numeric_target_column_name', 
			 hide_p_bar = True).fit_models()

# target and feature correlation - with filtered display
corr_df, feat_corr = n_a.corr(plot = True, thresh = .5)

# visualizing trend lines and overlaying outliers on a selected subset
n_a.vis_fit(olrs_mapping = lrs.z_olrs_)

# assess fit results visually - OLS
n_a.vis_ols_fit()

# check fit results
n_a.z_fit_results_['column_name'].summary()

# interactive multivariate analysis
n_a.vis_multi(col = 'column_name1', color = 'column_name2', symbol = 'column_name3', 
              trendline = 'ols', olrs_idx = lrs.z_olrs_idx_)

Categorical features analysis

# categorical analysis - cats vs num
c_a = CatAna(df, cat_cols, 'numeric_target_column_name', hide_p_bar = True)

# ANOVA with assumptions and mutual info scores
anova = c_a.ana_owva()

# Post-hoc displaying groups that could be merged
post_hoc = c_a.ana_post(multi_tst_corrc = 'bonf')

# categorical analysis - cats vs cat
c_a = CatAna(df, cat_cols, 'categorical_target_column_name', hide_p_bar = True)

# chi2 test of independence
chi2 = c_a.ana_chi2()

Preprocessing Pipeline

# sklearn pipline integration
from sklearn.preprocessing import PolynomialFeatures
from sklearn.pipeline import Pipeline

# setup
feats = ['column_name1', 'column_name2']

lrs = Olrs(cols = feats, hide_p_bar = True)
poly = PolynomialFeatures(interaction_only = True, include_bias = False).set_output(transform = "pandas")

# pipline
pl = Pipeline([
    ('pf', poly),
    ('olrs', lrs),
    ])
    
pl.fit_transform(df[feats])

Tutorials

ZAPS in Colab

License

MIT

API Documentation

The official documentation is hosted at Read The Docs.

Contributing to ZAPS

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

TOC