ypds_helpers 0.0.3

No project description provided

ypds_helpers

A Python toolkit for data analysis, visualization, and machine learning preprocessing

ypds_helpers is a collection of utility functions and tools designed to streamline common data science workflows. It provides convenient helpers for data exploration, correlation analysis, visualization, and machine learning model preprocessing pipelines.

Features • Installation • Usage • API Reference

Features

• Data Handling: Quick data exploration with comprehensive statistics and type-based column selection
• Correlation Analysis: Advanced correlation detection using Phik (φk) correlation coefficient for mixed data types
• Visualization: Ready-to-use plotting functions for distributions, categorical data, and model residuals
• ML Preprocessing: Pre-built pipelines for numerical and categorical data preprocessing
• Model Evaluation: Grid search utilities with automatic result tracking and comparison
• Jupyter Integration: Automatic detection and proper display in both Jupyter notebooks and regular Python environments

Installation

pip install ypds-helpers

The package requires Python ≤3.13 and automatically installs the following dependencies:

• pandas
• numpy
• scikit-learn
• seaborn
• matplotlib
• phik

Usage

Quick Data Exploration

import pandas as pd
from ypds_helpers.data_handling import show_df, get_num_cols, get_cat_cols

# Load your data
df = pd.read_csv('data.csv')

# Get comprehensive overview
show_df(df, n=10)  # Shows first 10 rows, statistics for numerical and categorical columns

# Get column lists by type
numerical_cols = get_num_cols(df, exclude_cols=['id'])
categorical_cols = get_cat_cols(df, exclude_cols=['target'])

Correlation Analysis

from ypds_helpers.data_handling import highest_corrs

# Find strongest correlations using Phik (works with mixed data types)
top_correlations = highest_corrs(
    df, 
    cols=['age', 'income', 'category', 'score'],
    interval_cols=['age', 'income', 'score'],
    num=15
)

Data Visualization

from ypds_helpers.plotting import plot_numeric, plot_cats, show_residues

# Visualize numerical features with histograms and boxplots
plot_numeric(
    df, 
    num_cols=['age', 'income', 'score'],
    hue='category',  # Split by category
    normalize=True,
    kde=True
)

# Visualize categorical distributions
plot_cats(
    df,
    cat_cols=['region', 'product_type'],
    hue='customer_segment',
    max_cats=10  # Group smaller categories
)

# Analyze model residuals
show_residues(y_true, y_pred, title='Model Performance')

Machine Learning Preprocessing

from ypds_helpers.models import (
    make_num_processor,
    make_ord_processor,
    make_typo_corrector,
    grid_search
)
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
from sklearn.ensemble import RandomForestRegressor

# Create preprocessing pipelines
num_pipeline = make_num_processor(min_val=0, max_val=100)
cat_pipeline = make_ord_processor(categories=['low', 'medium', 'high'])

# Combine transformers
preprocessor = ColumnTransformer([
    ('num', num_pipeline, numerical_cols),
    ('cat', cat_pipeline, categorical_cols)
])

# Create full pipeline
pipeline = Pipeline([
    ('preprocessor', preprocessor),
    ('model', RandomForestRegressor())
])

# Grid search with automatic result tracking
param_grid = {
    'model__n_estimators': [100, 200],
    'model__max_depth': [10, 20, None]
}

grid_search(
    pipeline=pipeline,
    grid=param_grid,
    X_train=X_train,
    y_train=y_train,
    model='random_forest',
    scoring='neg_mean_squared_error',
    cv_method=5,
    n_jobs=-1
)

API Reference

Data Handling Module

show_df(df, n=5)

Display comprehensive information about a DataFrame including head, statistics, and info.

Parameters:

• df: DataFrame to analyze
• n: Number of rows to display (default: 5)

get_num_cols(df, exclude_cols=None)

Returns list of numerical column names.

get_cat_cols(df, exclude_cols=None)

Returns list of categorical column names.

print_unique_cat_vals(dfs, exclude=None)

Print unique values for categorical features across one or multiple DataFrames.

highest_corrs(df, cols=None, interval_cols=None, num=10)

Calculate and return the highest correlations using Phik coefficient.

Parameters:

• df: DataFrame with data
• cols: Columns to analyze (default: all columns)
• interval_cols: Numerical columns for interval correlation
• num: Number of top correlations to return (default: 10)

Plotting Module

plot_numeric(df, num_cols=None, title='', hue=None, normalize=False, kde=True, ncols=2, scale=2.5, **kwargs)

Create histograms and boxplots for numerical features.

plot_cats(df, cat_cols=None, hue=None, title='', ncols=2, max_cats=10, max_cats_alias='all_other', **kwargs)

Create bar charts for categorical feature distributions.

show_residues(y_true, y_pred, title='', **kwargs)

Plot residual distribution and scatter plot for model evaluation.

Models Module

make_num_processor(min_val, max_val)

Create a preprocessing pipeline for numerical data with sanitization, imputation, and scaling.

make_ord_processor(categories)

Create a preprocessing pipeline for ordinal categorical data with typo correction, encoding, and imputation.

make_typo_corrector(correct_vals)

Create a transformer that corrects single-character typos using Hamming distance.

grid_search(pipeline, grid, X_train, y_train, model, scoring='roc_auc', cv_method=None, n_jobs=-1)

Perform grid search with automatic result tracking and model comparison.

Parameters:

• pipeline: Scikit-learn pipeline
• grid: Parameter grid (list or dict)
• X_train, y_train: Training data
• model: Model name for tracking
• scoring: Scoring metric (default: 'roc_auc')
• cv_method: Cross-validation method (default: 5-fold)
• n_jobs: Number of parallel jobs (default: -1 for all cores)

show_search_result(search, n_results=10)

Display formatted grid search results.

evaluate_params(grid)

Display maximum metric values for each hyperparameter.

Examples

Check out the examples directory for complete working examples including:

• Data exploration workflows
• Feature engineering pipelines
• Model training and evaluation
• Visualization galleries

Development Status

This package is currently in Beta (Development Status: 4 - Beta). The API may change in future releases.

License

ypds_helpers is distributed under the terms of the MIT license.