Comparing Experiments in H2O Hydrogen Torch by mAP
- Predictive AI
- H2O Hydrogen Torch
This lesson shows how to compare experiments in H2O Hydrogen Torch to find the model with the best mean average precision, or mAP, score. It covers filtering the View experiments table by project name to narrow down a set of related runs, then selecting two or more experiments and opening the Compare experiments view, which overlays their training charts in different colors. It also covers the Config tab, where the 'Show differences only' option hides identical hyperparameter values so you can quickly see what changed, such as the backbone model, embedding size, or learning rate, between two runs.
Key takeaways
- How to filter the View experiments table by project name to focus on a related set of runs
- What the Compare experiments view shows when you select two or more experiments to overlay
- How the 'Show differences only' option isolates the hyperparameters that changed between two experiments
- Why adjusting the backbone architecture and learning rate can meaningfully improve a model's mAP score
Start owning your AI