A run tracker, you say? What is that? Simply put, it’s a tool that keeps a record of each run of your Machine Learning experiments. But what’s it good for? Read on to find out.
If you are doing the following, it’s a sign that you could benefit from using a run tracker.
- You are using a spreadsheet to track your parameters, hyperparameters and summary statistics.
- You can’t be certain which version of the data you were using when you got a particular result.
- Model accuracies sometimes change for no discernable reason.
- Picking up someone else’s work involves emailing things around and having several meetings.
- Sometimes it’s hard to be sure of the differences between how two models were created.
- It’s impossible to tell whether a change in output was caused by your runtime environment.
- You can’t easily go back to a previous state in your experiment.
Dotscience is a run tracker that keeps automatic track of your work as you go, allowing you to:
- Track runs in Jupyter and scripts.
- Compare and share run metrics
- Track provenance of data and models
- Deploy and statistically monitor models