funpymodeling 0.2.5

A package designed for data scientists and teachers, to speed up their ML projects, focused on exploratory data analysis, data preparation, and model performance.

funpymodeling

A Python package designed for data scientists and teachers to speed up Exploratory Data Analysis (EDA), Data Preparation, and Model Performance evaluation.

Companion package for the book Data Science Live Book.

Install

pip install funpymodeling

Quick-start

from funpymodeling import *
import pandas as pd

All functions are well documented. Access the documentation via help(function_name).

Some functions have comments in the source code so users can open the black-box and learn how they were developed, or fine-tune/improve any of them.

---

Exploratory Data Analysis

status — Dataset health check

For each variable: quantity and percentage of zeros, NAs, and infinite values. Percentages are rounded to 2 decimal places. Also reports data type and number of unique values.

status(heart_disease)

profiling_num — Numerical variable profiling

Returns mean, std_dev, variation_coef, percentiles (p_01 to p_99), skewness, kurtosis, IQR, range_98 and range_80.

profiling_num(heart_disease)

plot_num — Histograms for numerical variables

Plots histograms for all numerical variables in a single figure.

plot_num(heart_disease)
plot_num(heart_disease, bins=20)

freq_tbl — Frequency table for categorical variables

Returns frequency, percentage (as proportion 0 to 1), and cumulative percentage for each category. Handles category dtype, NaN values, and high-cardinality variables.

freq_tbl(heart_disease, input='chest_pain')

# Multiple variables
freq_tbl(heart_disease, input=['chest_pain', 'gender'])

# All categorical variables at once
freq_tbl(heart_disease)

freq_plot — Bar chart for categorical variables

Horizontal bar charts showing frequency and percentage for each category.

freq_plot(heart_disease, input='chest_pain')

---

Target & Variable Analysis

var_rank_info — Variable importance via Information Theory

Computes entropy (en), mutual information (mi), information gain (ig), and gain ratio (gr) for each variable against the target.

var_rank_info(heart_disease, target='has_heart_disease')

cross_plot — Input vs. target distribution

Shows how the input variable relates to the target variable, with likelihood rates for each bin/bucket. Supports plot_type='both' (default), 'percentual', or 'quantity'.

cross_plot(heart_disease, input='chest_pain', target='has_heart_disease')

# Multiple variables at once
cross_plot(heart_disease, input=['age', 'chest_pain'], target='has_heart_disease')

plotar — Boxplot and density histograms by target

Useful to visually assess whether a variable discriminates between target classes.

# Boxplot
plotar(heart_disease, input='age',
       target='has_heart_disease', plot_type='boxplot')

# Density histogram
plotar(heart_disease, input='age',
       target='has_heart_disease', plot_type='histdens')

Notes:

• input must be numeric and target must be categorical.
• target can be multi-class (not only binary).

categ_analysis — Categorical variable vs. binary target

Profiling of a categorical input against a binary target: representativeness (perc_rows) and accuracy (perc_target) of each category.

categ_analysis(data, input='country', target='has_flu')

corr_pair — Pairwise correlation

Calculate pairwise correlation (R and R²) among all numeric features.

corr_pair(heart_disease)

---

Data Preparation

equal_freq — Equal-frequency binning

Splits a numeric variable into n_bins segments of approximately equal size.

equal_freq(heart_disease['age'], n_bins=5)

discretize_get_bins / discretize_df — Two-step discretization

First get the bins, then apply them. Useful for applying the same bins to train and test sets.

d_bins = discretize_get_bins(heart_disease, input=['age', 'oldpeak'], n_bins=5)
heart_disease_disc = discretize_df(heart_disease, d_bins)

convert_df_to_categoric — Convert all columns to categorical

Numeric variables are binned using equal-frequency; other variables are cast to string.

convert_df_to_categoric(heart_disease, n_bins=5)

range01 — Min-max normalization

Scales a numeric vector to the [0, 1] range.

range01(heart_disease['oldpeak'])

auto_grouping — Reduce cardinality via clustering

Groups categories of a high-cardinality variable using KMeans on profiling metrics (perc_rows, perc_target).

res = auto_grouping(data, input='country', target='has_flu', n_groups=3)
res['df_equivalence']   # mapping from original to new categories
res['recateg_results']  # profiling of the new groups

---

Outlier Treatment

tukey_outlier / hampel_outlier — Outlier thresholds

tukey_outlier(heart_disease['age'])
# {'lower': ..., 'upper': ...}

hampel_outlier(heart_disease['age'])
# {'lower': ..., 'upper': ...}

prep_outliers — Handle outliers in a DataFrame

Two modes:

• type='stop' — clips values at the threshold (recommended for predictive modeling)
• type='set_na' — converts outliers to NaN (recommended for statistical analysis)

Three methods: 'tukey', 'hampel', 'bottom_top'.

# Clip outliers using Tukey method
prep_outliers(heart_disease, input=['age', 'oldpeak'], type='stop', method='tukey')

# Set outliers to NA using Hampel method
prep_outliers(heart_disease, input=['age'], type='set_na', method='hampel')

# Bottom/top percentile
prep_outliers(heart_disease, input=['age'], type='stop',
              method='bottom_top', top_percent=0.01, bottom_percent=0.01)

---

Predictive Model Performance

gain_lift — Gain and lift curves

Higher values at the beginning of the population implies a better model.

gain_lift(data=scored_data, score='score', target='has_heart_disease')

coord_plot — Coordinate plot for clustering

Parallel coordinate plot for cluster profiling. Returns original and min-max normalized tables.

coord_plot(iris, group_var='species')

---

All Functions

Exploratory Data Analysis

Function	Description
status(data)	Health check: zeros, NAs, infinites, types, unique values (percentages rounded to 2 decimals)
profiling_num(data)	Numeric profiling: mean, std, percentiles, skewness, kurtosis
freq_tbl(data, input)	Frequency table for categorical variables
freq_plot(data, input)	Horizontal bar chart of frequencies
plot_num(data, bins)	Histograms for all numerical variables
corr_pair(data, method)	Pairwise correlation (R and R²)
num_vars(data)	Returns numeric column names
cat_vars(data)	Returns categorical column names

Data Preparation

Function	Description
equal_freq(var, n_bins)	Equal-frequency binning
tukey_outlier(x)	Tukey outlier thresholds (IQR × 3)
hampel_outlier(x, k_mad_value)	Hampel outlier thresholds (median ± k×MAD)
prep_outliers(data, input, type, method)	Handle outliers: set_na or stop (clip)
categ_analysis(data, input, target)	Profile categorical variable vs binary target
auto_grouping(data, input, target, n_groups)	Reduce cardinality via KMeans clustering
discretize_get_bins(data, input, n_bins)	Get bin thresholds for discretization (edges rounded to 2 decimals)
discretize_df(data, data_bins)	Apply discretization bins to a DataFrame
range01(x)	Min-max normalization to [0, 1]
convert_df_to_categoric(data, n_bins)	Convert all columns to categorical
todf(data)	Convert lists, arrays, Series to DataFrame

Target & Variable Analysis

Function	Description
cross_plot(data, input, target)	Stacked bar chart: input vs binary target
var_rank_info(data, target)	Variable ranking via Information Theory
plotar(data, input, target, plot_type)	Boxplot or density histogram grouped by target

Model Validation

Function	Description
gain_lift(data, score, target)	Cumulative gain and lift chart + table
coord_plot(data, group_var)	Coordinate (parallel) plot for cluster profiling

Dependencies

pandas, numpy, matplotlib, scikit-learn, scipy, seaborn