Model Training Guide¶
This guide covers how to train machine learning models with Arc. Arc handles the complexity of model implementation while you focus on the high-level architecture and training strategy.
Overview¶
Training a model with Arc involves: 1. Describe your goal - Tell Arc what you want to predict 2. Review Arc-Graph - Arc generates a model architecture specification 3. Train - Arc builds and trains the PyTorch model 4. Monitor - View training progress in TensorBoard 5. Evaluate - Check model performance
Quick Start¶
The simplest way to train a model:
Arc will: - Analyze your data - Generate an Arc-Graph specification - Train the model - Launch TensorBoard - Report results
Using the /ml model Command¶
For more control, use the /ml model command:
/ml model --name diabetes_predictor
--instruction "Build a binary classifier to predict diabetes"
--data-table processed_diabetes_data
Understanding the Training Process¶
Arc generates an Arc-Graph specification defining inputs, model architecture, and training configuration. See Arc-Graph Specification for details.
Arc then builds the PyTorch model, loads data, trains for the specified epochs, and tracks metrics.
Monitoring Training¶
Arc automatically launches TensorBoard at http://localhost:6006 to view training progress, loss curves, and accuracy metrics.
Trained models are saved to ~/.arc/models/<model_name>/ with model weights, Arc-Graph specification, and metadata.
Check Training Jobs¶
View all training jobs:
View specific job details:
Example¶
/ml model --name spam_detector
--instruction "Binary classifier for spam detection with 2 hidden layers (64, 32)"
--data-table processed_emails
Next Steps¶
- Model Evaluation Guide - Evaluate your trained models
- Making Predictions - Use your model for inference
- Arc-Graph Specification - Understand the model spec format
Related Documentation¶
- Feature Engineering Guide - Prepare data for training
- TensorBoard Documentation - TensorBoard guide