The sequence is fixed and sensible — inspect, preprocess, prepare, compare, tune — and each step is its own call, so you can stop at the correlation or outlier plots before committing to the next one. Regression covers Linear, Ridge, Lasso, ElasticNet, Random Forest, XGBoost, SVR, KNN and CatBoost; classification covers Logistic Regression, Ridge Classifier, Random Forest, XGBoost, SVM, KNN, Decision Tree, Naive Bayes and CatBoost.
A Python server wrapping an automated machine-learning workflow: dataset inspection, correlation and outlier plots, preprocessing, model comparison for regression or classification, hyperparameter tuning and prediction.
- First look — `information_about_data` for shape, memory use, dtypes and missing values; `reading_csv` to load the file
- Visual checks — `visualize_correlation_num`, `visualize_correlation_cat` and `visualize_outliers`, plus `visualize_correlation_final` and `visualize_outliers_final` for the post-preprocessing view
- Preparation — `preprocessing_data` handles outliers and nulls, `prepare_data` encodes and scales for the problem type you name
- Model comparison — `models` evaluates a set of algorithms and reports R², MAE and MSE for regression, or accuracy and F1 for classification
- `best_model_hyperparameter` tunes the chosen model over a trial count and scoring metric you set
- `test_external_data` scores a separate file, `predict_value` predicts for new input, `feature_importance_analysis` ranks features with XGBoost, `visualize_accuracy_matrix` draws the confusion matrix
Python 3.8 or higher and `pip install -r requirements.txt` — pandas, pyarrow, numpy, scikit-learn, xgboost, lightgbm and catboost. Set the data directory path in `utils/read_csv_file.py` to match your machine, then point your client at `uv --directory C:\YOUR\PROJECT\PATH\AutoML run main.py`.
One command — uvx auto