This phase largely troubleshoots issues with the training and deployment pipelines related to issues with the input data, preprocessing steps and discrepancies between model training/deployment.
Specifically:
different scaling approaches were tested
missing preprocessing steps for sentinel 1 and sentinel 2 data were added
fixes to adjust for swapped features during TML feature extraction were added
a validation script was added to check assumptions on input data
the training pipeline was adapted to allow testing of multi-class classification approaches
This phase largely troubleshoots issues with the training and deployment pipelines related to issues with the input data, preprocessing steps and discrepancies between model training/deployment.
Specifically: