Automating Workflows

This section covers advanced techniques for automating and structuring multiple experiments in HydraFlow. It provides tools for defining complex parameter spaces and reusable experiment definitions.

Overview

After creating your basic HydraFlow applications, the next step is to automate your experiment workflows. This includes:

• Creating parameter sweeps across complex combinations
• Defining reusable experiment configurations
• Organizing large-scale experiment campaigns

Key Components

The main components for workflow automation in HydraFlow are:

1. Extended Sweep Syntax: A powerful syntax for defining parameter spaces beyond simple comma-separated values.

2. Job Configuration: A YAML-based definition system for creating reusable experiment workflows.

Practical Examples

For hands-on examples of workflow automation, see our Practical Tutorials section, specifically:

• Automating Complex Workflows: A tutorial that demonstrates how to use hydraflow.yaml to define and execute various types of workflows
• Analyzing Experiment Results: Learn how to work with results from automated experiment runs

Extended Sweep Syntax

HydraFlow extends Hydra's sweep syntax to provide more powerful ways to define parameter spaces:

# Range of values (inclusive)
python train.py -m "learning_rate=0.001:0.1:0.001"  # start:end:step

# SI prefixes
python train.py -m "batch_size=1k,2k,4k"  # 1000, 2000, 4000

# Logarithmic spacing
python train.py -m "learning_rate=log(0.0001:0.1:10)"  # 10 points log-spaced

Learn more about these capabilities in Sweep Syntax.

Job Configuration

For more complex experiment workflows, you can use HydraFlow's job configuration system:

jobs:
  train_models:
    run: python train.py
    sets:
      - each: model=small,medium,large
        all: seed=42 epochs=10

  evaluate_models:
    run: python evaluate.py
    sets:
      - each: model=small,medium,large
        all: test_data=validation

This approach allows you to define reusable experiment definitions that can be executed with a single command. Learn more in Job Configuration.

Executing Workflows

Once defined, workflows can be executed using the hydraflow run command:

# Execute a job defined in hydraflow.yaml
hydraflow run train_models

# Preview execution with dry run
hydraflow run train_models --dry-run

# Run a job with additional overrides
hydraflow run train_models seed=123

What's Next

In the following pages, we'll explore workflow automation in detail:

• Sweep Syntax: Learn about HydraFlow's extended syntax for defining parameter spaces.
• Job Configuration: Discover how to create reusable job definitions for your experiments.

After automating your experiments, you'll want to analyze the results using the tools covered in Part 3: Analyzing Results.