Comparing AI model performance across multiple analysis files
The Evaluate Models screen lets you compare the performance of AI models by selecting multiple analysis files and viewing comparative metrics. This is useful for validating that a new model version performs as well as or better than the previous one.
Two or more analysis files (.hdf5) that have been analyzed with AI models.
Click Add Files to browse for analysis files. Select multiple .hdf5 files to include in the evaluation. The application reads each file's metadata and extracts:
Model name used for analysis.
File name and path.
Number of seeds analyzed.
The selected files appear in a list. Click the remove icon next to any file to deselect it.
The application checks that selected files are compatible for comparison:
If files were analyzed with different models, a warning appears. While cross-model comparison is possible, results are most meaningful when comparing files analyzed with the same model or successive model versions.
Click Compare to navigate to the evaluation dashboard. The application generates comparative metrics including:
Class distribution comparison across files.
Agreement metrics between model predictions.
Per-class accuracy comparisons.
Statistical summaries highlighting significant differences.
Clear All: Remove all selected files and start fresh.
Regenerate: Reload metadata from all selected files.
Compare button is disabled
Fewer than 2 files selected
Select at least two analysis files
Model name shows "Unknown"
File was analyzed with an older application version
Re-run AI Classification on the file
Comparison metrics seem wrong
Files have very different sample sizes
Consider comparing files with similar seed counts
File cannot be added
File has no AI results
Run AI Classification on the file first
Last updated