AI GitHub
Aim for comparing training runs
Log experiment metrics and inspect curves and metadata across runs.
Key Takeaways
- Aim records tracked values in a run and offers a UI for exploring them.
- Add instrumentation around the key points in the training loop.
- Track two short training runs with identical metric names and distinct contexts.
- A plotting interface can only show values the experiment recorded.
How it works
Aim records tracked values in a run and offers a UI for exploring them. Context labels can distinguish signals such as training and validation loss.
What you need
Add instrumentation around the key points in the training loop. Use consistent metric names and dimensions so the UI can compare runs without an implicit mapping.
What to try first
Track two short training runs with identical metric names and distinct contexts. Compare the curves at equal step counts and review which settings were captured.
Record an evaluation curve at a different interval from the training curve. In comparisons, check whether the UI aligns points by step, time or another index.
Where to look for failures
A plotting interface can only show values the experiment recorded. Keep dataset identity and configuration beside metrics if a run must be reproduced.
Related Franklin pages
Choose what needs versioning and what needs tracking. A run's curves are easier to reproduce when they link to the exact data, configuration and saved artifact.
- Connect Python experiment and data workflows to a DagsHub repository. DagsHub Client for connecting ML projects to DagsHub
- Track large artifacts and dependency-aware stages alongside a Git project. DVC for versioning datasets and ML pipelines
For the steps that prepare raw data for training or evaluation, see Franklin’s glossary. Data preprocessing
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗