vllama train — AutoML Training

Automatically trains and compares multiple ML models on preprocessed data with hyperparameter tuning, and produces a ranked leaderboard.

---

Syntax

vllama train --path <data_folder> --target <column>

---

Parameters

Parameter	Short	Description
--path	-p	Path to folder containing train_data.csv and test_data.csv (output from vllama data)
--target	-t	Name of the target column

---

What It Does

1. Auto-detects task type — classification or regression based on the target column
2. Trains all models with RandomizedSearchCV hyperparameter tuning on each
3. Evaluates every model on the test set with comprehensive metrics
4. Ranks models in a leaderboard
5. Saves all models and generates visualizations
6. Produces an HTML report with all results

---

Models Trained

ClassificationRegression

• Logistic Regression
• Random Forest
• XGBoost
• LightGBM
• CatBoost
• SVM
• KNN
• MLP (Neural Net)
• Naive Bayes

• Random Forest
• XGBoost
• LightGBM
• CatBoost
• SVR
• KNN
• MLP (Neural Net)

---

Examples

# Standard usage after vllama data
vllama train --path ./output_folder_20240101_120000 --target price

# Short form
vllama train -p ./output_folder_20240101_120000 -t label

---

Output Structure

results/
├── model_summary.csv           ← Leaderboard: all models ranked by metric
├── best_model.pkl              ← Best model, loadable with joblib
├── best_model.txt              ← Best model name and score
├── report.html                 ← Full interactive HTML report ← open this
└── per_model/
    ├── RandomForest/
    │   ├── RandomForest_best_model.pkl
    │   ├── RandomForest_tuning_results.csv
    │   ├── RandomForest_confusion_matrix.png  (classification)
    │   └── RandomForest_roc_curve.png         (binary classification)
    ├── XGBoost/
    │   └── ...
    └── ...

Open report.html

After training, open results/report.html in your browser. It contains the full leaderboard, per-model metrics, and all visualizations in one place.

---

Loading the Best Model

import joblib

model = joblib.load("results/best_model.pkl")
predictions = model.predict(X_test)

---

Full Pipeline

vllama data --path raw_data.csv --target price
vllama train --path ./output_folder_YYYYMMDD_HHMMSS --target price
# Open results/report.html

See the AutoML Pipeline Guide for a detailed walkthrough with a real dataset.