Release after major change with recreated biweekly renku datasets
  • Recreate biweekly renku datasets with float32 single precision and newly uploaded data to climetlab_s2s_ai_challenge. Before lead_time in tp had a one day shift. These new renku datasets will be used by the s2saichallengescorer. (#3, #5, !14, s2s-ai-challenge#22, Aaron Spring). Averaged recalibrated ECMWF RPSS skill value to beat at least: RPSS = -0.043, see RPSS_verification.ipynb. You should also beat climatology, i.e. RPSS = 0.
  • Order of processing gridded RPSS to final score: (#7, !9, s2s-ai-competition-scoring-image!2, Aaron Spring)
    1. RPSS
    2. penalize where NaN submitted but value expected #7
    3. clip(-10,1): prevent too negative values
    4. mean over forecast_time
    5. spatially weighted mean [90N-60S]
    6. mean over lead_time and data_vars
  • Dont forget to git add current_notebook.ipynb also to ensure that consistent training pipeline and submission file are tagged, added to notebooks. (!9, Aaron Spring)
  • Rerun ML_train_and_predict.ipynb (!9, Aaron Spring)
  • Fix typo in safeguards in ML_forecast_template.ipynb: "We did NOT use test explicitly in training or implicitly in incrementally adjusting parameters." (!8, Aaron Spring)
  • Add notebooks showcasing accessing output of different models from different sources: (!2, Aaron Spring)
  • fix netcdf4 version to 1.5.4 for opendap to work lazily with xarray (!2, !7, Aaron Spring)