Assess Model Performance in Hydrogen Torch
- Predictive AI
- H2O Hydrogen Torch
This lesson walks through navigating the results of a completed object detection experiment in H2O Hydrogen Torch, starting with the Charts tab and the Validation mAP metric. It explains how mean average precision is calculated per class and then averaged, with higher scores indicating better object detection performance. It introduces hyperparameter tuning as the next step for improving that score, starting with the backbone, the pretrained model an experiment builds on, and noting that learning rate should typically be tuned alongside it. It sets up the following module's deeper look at grid search for testing hyperparameter combinations.
Key takeaways
- What the Validation mAP metric measures, averaging per class precision across an object detection model
- Why a higher mAP score indicates better object detection performance in Hydrogen Torch
- How the backbone, the pretrained model an experiment builds on, is usually the first hyperparameter to tune
- Why learning rate should be tuned alongside the backbone since different backbones respond differently
- The next module introduces grid search for testing hyperparameter combinations systematically
Start owning your AI