aabragan/automl-model-training

How aabragan/automl-model-training works

This is a Python AutoML training and prediction pipeline built on AutoGluon. Its main workflow takes a CSV through optional profiling and feature engineering, trains and evaluates models, then writes run artifacts for later prediction. Separate capabilities support backtesting, experiment comparison, and iterative training through an autonomous or LLM-driven agent. The graph follows documented workflows and sampled source; some subsystem wiring is omitted where implementation excerpts were unavailable.

Main entry points: train.py Model training · predict.py Prediction · run_artifacts.py Run artifacts · agent.py Autonomous training

Components 20

Data preparation

Dataset profiling src/automl_model_training/profile.py
Load and split data src/automl_model_training/data.py
Feature transforms src/automl_model_training/feature_engineering.py

Training and evaluation

Model training src/automl_model_training/train.py
Evaluation and analysis src/automl_model_training/evaluate
SHAP explanations src/automl_model_training/evaluate/explain.py
Ensemble pruning src/automl_model_training/evaluate/prune.py

Prediction and backtesting

Prediction src/automl_model_training/predict.py
Drift detection src/automl_model_training/drift.py
Temporal backtesting src/automl_model_training/backtest.py

Agents and tools

Autonomous training src/automl_model_training/agent.py
Ollama agent src/automl_model_training/ollama_agent.py
Agent tool layer src/automl_model_training/tools

Run management

Experiment tracking src/automl_model_training/experiment.py
Run comparison src/automl_model_training/compare.py
Run artifacts src/automl_model_training/run_artifacts.py

Other components

User
CSV dataset
AutoGluon
Ollama service

Connections 35

User → CSV dataset: provides
User → Dataset profiling: requests profile
Dataset profiling → CSV dataset: analyzes
User → Feature transforms: requests transforms
Feature transforms → CSV dataset: reads
Feature transforms → Model training: engineered CSV
User → Model training: starts training
Model training → Load and split data: loads and splits
Load and split data → CSV dataset: reads
Model training → AutoGluon: trains models
Model training → Evaluation and analysis: evaluates run
Model training → SHAP explanations: generates explanations
Model training → Ensemble pruning: may prune models
Model training → Run artifacts: writes run outputs
User → Prediction: requests predictions
Prediction → AutoGluon: loads model and predicts
Prediction → Drift detection: may check drift
Prediction → Run artifacts: writes outputs
User → Temporal backtesting: starts backtest
Temporal backtesting → Model training: runs temporal training
Temporal backtesting → Run artifacts: writes results
User → Autonomous training: starts iteration
Autonomous training → Dataset profiling: profiles data
Autonomous training → Model training: trains candidate
Autonomous training → Evaluation and analysis: reviews results
Ollama agent → Ollama service: requests tool calls
Ollama agent → Agent tool layer: dispatches tools
Agent tool layer → Model training: runs training tools
Agent tool layer → Feature transforms: applies transforms
Agent tool layer → Prediction: runs prediction tools
User → Ollama agent: starts LLM workflow
User → Experiment tracking: lists runs
Experiment tracking → Run artifacts: records run data
User → Run comparison: compares runs
Run comparison → Run artifacts: reads run data

More Python diagrams: dynatrace/dynatrace-for-ai · hitsz-tmg/lychee-fd · hvasbath/beat · kosasih/quantum-pi-network · ntt123/light-speed

Drawn by GitDiagram on September 25, 2026 from the repository's file tree, README and source files. Read as Markdown ·