mljar / docs Goto Github PK

Needed tutorials on how to use different types of modes available for now.

Please add documentation about Optuna integration.

On the left side menu there's a mistake in word explainaility, "x" is missing.
Next, under "Decision tree visualization" there's a word "craeted" while it should be created.

http://supervised.mljar.com/features/explain/

There are 3 modes available:

• Explain - To be used when the user wants to explain and understand the data.
• Perform - To be used when the user wants to train a model that will be used in real-life use cases
• Compete- To be used for machine learning competitions (maximum performance)

The Explain mode:

• Uses algorithms: Baseline, Linear, Decision Tree, Random Forest, XGBoost, Neural Network, and Ensemble
• Uses 75%/25% train/test split
• Has full explanations in reports: learning curves, importance plots, and SHAP plots. (explain_level=2)
• Steps: simple_algorithms, default_algorithms, ensemble

The Perform mode:

• Uses the following models: Linear, Random Forest, LightGBM, XGBoost, CatBoost, Neural Network, and Ensemble.
• Uses a 5-fold CV (Cross-Validation) with shuffle and stratification.
• Has learning curves and importance plots in reports. (explain_level=1)
• Steps: simple_algorithms, default_algorithms, not_so_random, golden_features, insert_random_feature, feature_selection, hill_climbing_1, hill_climbing_2, ensemble
• There are 4 models created for each algorithm in not_so_random step.
• There are 2 top models for each algorithm tuned in hill_climbing step.

The Compete mode:

• Uses the following models: Linear, DecisionTree, Random Forest, Extra Trees, XGBoost, CatBoost, Neural Network, Nearest Neighbors, Ensemble, and Stacking.
• Uses 10-fold CV (Cross-Validation) with shuffle and stratification.
• It has only learning curves in the reports. (explain_level=0)
• Steps: simple_algorithms, default_algorithms, not_so_random, golden_features, insert_random_feature, feature_selection, hill_climbing_1, hill_climbing_2, ensemble, stack, ensemble_stacked
• There are 9 models created for each algorithm in not_so_random step.
• There are 3 top models for each algorithm tuned in hill_climbing step.