teda-temporal 0.1.0

Temporal EDA for feature drift, PCA, breakpoints, and trends

TEDA - Temporal Exploratory Data Analysis

A Python package for temporal feature analysis and drift detection.

Features

1. Feature Overview Table

• Comprehensive statistics for all features
• Missing values, zero rates, unique percentages
• Volatility and PSI (Population Stability Index) tracking
• Correlation with target variable

2. Monthly PSI Tracking

• Track Population Stability Index over time
• Automatic threshold warnings

3. 3D PCA Visualization

• Monthly mean projection in PCA space
• Travel distance bar chart showing monthly movement

4. Structural Breakpoint Detection

• Uses Ruptures library for changepoint detection
• Monthly aggregation with PELT/Binseg/Dynp algorithms
• Visual breakpoint markers on time series

5. Feature Importance Evolution

• Track feature importance over time
• Multiple models: RandomForest & XGBoost

6. HP Filter Decomposition

• Hodrick-Prescott filter for trend extraction
• Seasonal pattern detection
• Residual analysis

Installation

pip install teda

Or install from source:

git clone https://github.com/tergelitu/teda.git
cd teda
pip install -e .

Requirements

• Python 3.8+
• pandas >= 1.3.0
• numpy >= 1.21.0
• plotly >= 5.0.0
• scikit-learn >= 1.0.0
• xgboost >= 1.5.0
• ruptures >= 1.1.0
• statsmodels >= 0.13.0
• ipywidgets >= 7.6.0

📖 Quick Start

See full example usage in teda-demo.ipynb

import pandas as pd
import teda as te

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

# 1. Feature Overview
te.overview_table(df, date_col='date', target_col='target', exclude_cols=['id'])

# 2. PSI Tracking
te.plot_monthly_psi(df, date_col='date', feature_col='col')

# 3. 3D PCA Visualization
te.plot_monthly_pca(df, date_col='date', exclude_cols=['id', 'target'])

# 4. Structural Breakpoints
te.detect_change_points(df, date_col='date', feature_col='col', model='l2')

# 5. Feature Importance Evolution
te.plot_monthly_importance(df, date_col='date', target_col='target', feature_name='col')

# 6. HP Filter Decomposition
te.plot_hp_decomposition(df, date_col='date', feature_col='col')

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Author

Tergel Munkhbaatar

• GitHub: @tergelitu

Use Cases

• Model Monitoring: Track feature drift in production ML systems
• Data Quality: Identify distribution shifts and anomalies
• Feature Engineering: Understand temporal patterns and breakpoints
• Exploratory Analysis: Quick overview of time-series feature behavior

---

Star ⭐ this repo if you find it useful!