tlux_lazy

An open-source Python library that automatically compares different Machine Learning models along with different preprocessing and scaling techniques.

Instead of manually writing "model → choose scaler → choose columns → transform → train → predict → calculate metrics" for every combination, this library runs all the experiments for you and returns a model leaderboard.

---

Features

• Explicit problem type — you tell it classification or regression via the required is_classification=True/False flag (no guesswork). The flag is validated against your data — a mismatch raises an error instead of running.
• 21 classifiers + 21 regressors — including XGBoost (optional).
• 8 scaling techniques — Standard, MinMax, Robust, MaxAbs, Normalizer, Power, Quantile, plus No Scaling. Runs on raw data by default; opt in with scale=True, a scaler name, or a list of names.
• User-controlled column scaling — pick exactly which columns get scaled via scale_columns (used together with scaling).
• Automated leaderboard — sorted by the metric you choose (default: Accuracy for classification, R² for regression).
• 7 default metrics — Accuracy / Precision / Recall / F1 (classification), R² / MAE / MSE / RMSE (regression). Fully customisable.
• Run only the models you want — pass a name, list, class, or dict.
• Parallel training via n_jobs (joblib).
• Clean result — returns a ready-to-use leaderboard DataFrame; errors are printed to stderr.

---

Installation

pip install tlux_lazy

For development (includes tests):

pip install -e ".[dev]"

For XGBoost support (optional):

pip install -e ".[xgboost]"

Or install everything at once:

pip install -e ".[dev,xgboost]"

---

Quick Start

import numpy as np
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split

from tlux import compare_models

# Load data
iris = load_iris()
X_train, X_test, y_train, y_test = train_test_split(
    iris.data, iris.target, test_size=0.3, random_state=42
)

# Raw data only (default — no scaling)
leaderboard = compare_models(
    X_train, y_train, X_test, y_test,
    is_classification=True,   # required: True=classification, False=regression
)

Scaling is opt-in with scale. You can enable all techniques, scale all columns, or pick specific scaling techniques and specific columns:

# Raw data only (default) — no scaling
compare_models(..., is_classification=True)

# Try EVERY scaling technique on ALL columns
compare_models(..., is_classification=True, scale=True)

# Try specific scaling technique(s)
compare_models(..., is_classification=True, scale="StandardScaler")
compare_models(..., is_classification=True, scale=["MinMaxScaler", "RobustScaler"])

# Restrict which columns get scaled (used with any of the above)
compare_models(..., is_classification=True, scale="RobustScaler", scale_columns=[0, 2])

Output is a pandas.DataFrame (the leaderboard) and a printed table:

Model	Scaling	Scaled Columns	Accuracy	Precision	Recall	F1
SVC	StandardScaler	0, 2	1.00	1.00	1.00	1.00
XGBoost	No Scaling	0, 2	1.00	1.00	1.00	1.00
Random Forest	RobustScaler	0, 2	1.00	1.00	1.00	1.00
...	...	...	...	...	...	...

---

Run Only Specific Models

# Single model by name
compare_models(..., models="XGBoost")

# A few models by name
compare_models(..., models=["XGBoost", "Random Forest", "KNN"])

# Using sklearn estimator classes directly
from sklearn.svm import SVC
compare_models(..., models=[SVC, RandomForestClassifier])

# Custom-named dict (great for custom estimators)
compare_models(..., models={"My Custom Model": SVC})

List available model names first:

from tlux import get_model_names

print(get_model_names("classification"))
print(get_model_names("regression"))

---

Sort by Any Metric

# Sort the leaderboard by F1-score
compare_models(..., sort_by="F1")

# Sort by MAE (ascending — lower is better)
compare_models(..., sort_by="MAE", ascending=True)

# Regression defaults to R2, classification defaults to Accuracy

---

Workflow

X_train, y_train, X_test, y_test
              ↓
       Detect Problem Type
          ↙          ↘
 Classification     Regression
       ↓                ↓
 Try Classifiers    Try Regressors
       ↓                ↓
 Try Preprocessing / Scaling
       ↓
 Train & Predict
       ↓
 Calculate Metrics
       ↓
 Compare Results
       ↓
      Leaderboard

---

Project Structure

Tlux/
├── pyproject.toml
├── README.md
├── DOCUMENTATION.md
├── WHY.md
├── Q&A.md
├── tlux/
│   ├── __init__.py      # Public exports
│   ├── api.py           # compare_models() entry point
│   ├── comparator.py    # Core engine: runs experiments
│   ├── detector.py      # Validate classification vs regression
│   ├── leaderboard.py   # Build + print leaderboard
│   ├── models.py        # Model catalogues + model resolution
│   └── scalers.py       # Scaling techniques + ColumnTransformer
└── tests/
    └── test_tlux.py     # Test suite (pytest)

---

Public API

from tlux import (
    compare_models,       # Main entry point
    run_experiments,      # Run experiments manually
    is_classification,    # Detect problem type
    build_leaderboard,    # Build leaderboard from results
    CLASSIFICATION_MODELS, # List of classification model names
    REGRESSION_MODELS,     # List of regression model names
    get_model_names,       # Get model names by problem type
    SCALERS,               # List of scaler names
    get_scaler_names,      # Get available scaler names
)

---

Future Features

• Automatic preprocessing
• Missing-value handling
• Categorical encoding
• Cross-validation
• Hyperparameter tuning
• Feature selection / importance
• Confusion matrix, ROC-AUC, Precision-Recall
• Model explainability
• Best-model recommendation
• Experiment reports
• Model saving / loading
• Custom scalers & custom metrics

---

License

MIT License. Free for the Python and Machine Learning community.

---

See DOCUMENTATION.md for the full API reference, Q&A.md for common questions, and WHY.md for the design rationale.