Ensure no leaking between train and validate/test dataset

informatics-lab / precip_rediagnosis

Project to use ML to re-diagnose precipitation fields from ensemble model fields

0 stars 0 forks source link

Ensure no leaking between train and validate/test dataset #82

Closed hannahbrown7 closed 1 year ago

hannahbrown7 commented 1 year ago

All ensemble members and adjacent data point should be in the same sample, so best to sample over time. Taking last 20% of timesteps from each event perhaps.