Reading Model Performance Charts in Hydrogen Torch
- Predictive AI
- H2O Hydrogen Torch
This lesson walks through evaluating a completed image regression experiment in H2O Hydrogen Torch. It covers refreshing the experiment's charts and reading the Validation MAE chart, where a value closer to zero signals better performance. It then explores the Validation prediction insights tab, which surfaces the best and worst validation samples at each epoch along with each sample's predicted sum, actual sum, and loss value, giving a way to judge how well the model is learning over time.
Key takeaways
- Why a Validation MAE value closer to zero signals more accurate predictions in an image regression model
- How the Validation prediction insights tab surfaces the best and worst samples at each training epoch
- How comparing a sample's predicted sum, actual sum, and loss value shows where the model is struggling
Start owning your AI