model-track-cr 1.1.1

Feature selection and model monitoring toolkit for credit and risk modeling.

🚀 model-track-cr

Read this in other languages: English, Português

model-track-cr is a professional Python toolkit designed to structure, standardize, and operationalize the full statistical and machine learning modeling workflow.

It is purpose-built for credit, risk, and supervised modeling (Binary, Multiclass, and Regression). Instead of fragmented notebooks, model-track-cr provides cohesive, Pandas-first components that work seamlessly together—from data diagnostic to model deployment.

---

📦 Installation

Install via pip:

pip install model-track-cr

For advanced features (like Bayesian Tuning and LightGBM support):

pip install "model-track-cr[tuning]"

---

⚡ Quickstart

Here's how easy it is to build a full feature engineering pipeline:

import pandas as pd
from model_track.preprocessing import DataOptimizer
from model_track.binning import TreeBinner
from model_track.woe import WoeCalculator
from model_track.stats import StatisticalSelector

# 1. Optimize memory
df = DataOptimizer.reduce_mem_usage(df)

# 2. Supervised Binning
binner = TreeBinner(max_depth=3)
binner.fit(df, column="feature", target="target")
df["feature_binned"] = binner.transform(df, column="feature")

# 3. Weight of Evidence (WoE) Transformation
woe_calc = WoeCalculator()
woe_calc.fit(df, target="target", columns=["feature_binned"])
df_woe = woe_calc.transform(df, columns=["feature_binned"])

# 4. Feature Selection (Information Value & Cramer's V)
selector = StatisticalSelector(iv_threshold=0.02)
selector.fit(df_woe, target="target", features=["feature_binned"])
df_selected = selector.transform(df_woe)

---

🛠️ Core Capabilities

• 📊 Diagnostics & Optimization: Memory reduction (DataOptimizer), missing value auditing (DataAuditor), and data schema extraction.
• 🪜 Binning: Supervised (TreeBinner) and Unsupervised (QuantileBinner) binning strategies.
• 🧮 WoE & IV: Weight of Evidence calculators (WoeCalculator) and Information Value adapters for Binary, Multiclass, and Regression tasks.
• 🎯 Feature Selection: Automated selection using IV, Variance, Spearman correlation, and Cramer's V (StatisticalSelector, RegressionSelector, MulticlassSelector).
• 📈 Stability Monitoring: Population Stability Index (PSI) and Temporal WoE stability matrices (WoeStability) to track data drift.
• 🧠 Hyperparameter Tuning: Model-agnostic Bayesian optimization with built-in LightGBM presets (BayesianTuner, LGBMTuner).
• 📏 Evaluation: Standardized metrics and reports for all task types.
• 💾 Project Context: Serialize your entire pipeline (bins, WoE maps, metadata) for production deployment (ProjectContext).

---

📓 Example Notebooks

Want to see it in action? Check out our end-to-end examples:

• Multiclass Pipeline (Wine Dataset): Binning → MulticlassSelector → OvRWoeAdapter → LightGBM → Evaluation.
• Regression Pipeline (California Housing): Audit → RegressionSelector → LightGBM → Stability Report.

---

🧩 Architecture & Philosophy

The library is built around a Pandas-first philosophy, where every component follows a fit/transform interface but expects and returns Pandas DataFrames. This ensures metadata (like column names) is preserved throughout the pipeline.

classDiagram
    BaseTransformer <|-- TreeBinner
    BaseTransformer <|-- WoeCalculator
    BaseTransformer <|-- StatisticalSelector
    
    class BaseTransformer {
        <<abstract>>
        +fit(df, target)
        +transform(df)
    }

🤝 Contributing

The project follows strict Test-Driven Development (TDD) with 100% coverage. See CONTRIBUTING.md and AGENTS.md for local setup and Gitflow guidelines.

📄 License

MIT License