AutoML Workbench (Streamlit)

A practical no-code/low-code ML app for:

• Uploading tabular data
• Exploring data quality and leakage risks
• Configuring preprocessing
• Training and comparing models
• Exporting predictions and model artifacts

doc_2026-02-13_16-54-21.mp4

Quick Start

git clone https://github.com/bewaffnete/Streamlit-ML-Workbench.git
pip install -r requirements.txt
streamlit run app.py

Open the URL shown by Streamlit (usually http://localhost:8501).

What This App Does

The UI follows a guided flow:

1. Data Upload: CSV, Excel, Parquet + validation and dataset summary
2. Target & EDA: choose target/features and inspect distributions/correlations
3. Warnings: automatic alerts (leakage, imbalance, missingness, high-cardinality)
4. Preprocessing: imputation, encoding, scaling, outlier handling, optional polynomial features
5. Model Config: task type + model family + CV/split + tuning controls
6. Train & Evaluate: launch background training jobs, compare results, inspect metrics
7. Predict & Export: download predictions, metadata, and trusted model artifacts

Key Features

• Clean separation: UI layer, service/orchestration layer, training/evaluation utilities
• Background training via ProcessPoolExecutor (non-blocking Streamlit flow)
• Fingerprint-based caching for heavy dataset summaries
• Smart data warnings with configurable thresholds
• Optional hyperparameter tuning (RandomizedSearchCV)
• Safe project state import/export (JSON/YAML) with strict schema checks

Project Structure

• app.py: app entry point and composition root
• automl_gui/ui/: Streamlit rendering modules (upload, EDA, training, export, sidebar)
• automl_gui/services.py: business orchestration (DataService, WarningService, TrainingService)
• automl_gui/core/jobs.py: background job manager
• automl_gui/data_utils.py: file loading, validation, dataset fingerprinting, cached summaries
• automl_gui/preprocessing.py: preprocessing config + factory
• automl_gui/training.py: model trainer + optimization hooks
• automl_gui/evaluation.py: metrics and evaluation helpers
• automl_gui/models.py: extensible model registry
• automl_gui/state.py: session facade + strict config schema validation
• automl_gui/warnings_utils.py: warning generation logic
• automl_gui/visualization.py: plotting helpers

Configuration

Environment variables:

• AUTOML_MAX_UPLOAD_MB: max upload size in MB (default 200)
• AUTOML_MAX_N_JOBS: max CPU jobs for supported models/tuning (default 2)
• AUTOML_BG_WORKERS: background worker processes (default 1)
• AUTOML_LOG_LEVEL: logger level (INFO, DEBUG, ...)
• AUTOML_LOG_TO_FILE: set 1 to enable rotating file logs
• AUTOML_LOG_FILE: log filename (default automl_gui.log)

Security Notes

• Do not load .joblib/.pkl files from untrusted sources.
• Model loading in the UI requires explicit trust confirmation.
• Project settings import uses strict schema validation and safe parsing (yaml.safe_load).

Troubleshooting

• ModuleNotFoundError: verify venv is activated and pip install -r requirements.txt completed.
• App feels slow on very large data: enable sampling in Upload tab and reduce selected features/models.
• Training takes too long: disable tuning, lower iterations, and reduce model count.