🕘 Meeting notes: Machine Learning WG
June 1, 2020
- Blog post
- rechunking-reshaping data
- file format
- dask-ml’s blockwise ensemble methods.
- Data Engineering stuff, Apache Beam pipelines, kubernetes stuff
- Drafted a blogpost
- Looking for postdoc to join Jupyter meets earth group (on Pangeo discourse)
- Student making progress on surface currents project again. Could use a GPU on the cloud.
- rechunker package: https://github.com/pangeo-data/rechunker
- Working on weatherbench dataset, building models.
- Hitting RAM limits (700 GB). Looking into converting the problem (possibly TFRecords)
- Management, hiring / onboarding data scientists.
- Prepping for AI for earth systems summer school. All virtual, webinars, hackathon.
- NOAA hazardous weather testbed. Storm mode analyzer. https://www2.mmm.ucar.edu/projects/ncar_ensemble/camviewer/images.php?d=2020053000&f=cnn_storm_mode&r=CONUS&i=1
- NN emulators fortran-neural network interface
- transfer learning for streamflow prediction (EGU presentation link below): https://doi.org/10.5194/egusphere-egu2020-11332
- pangeo gpu article. attached gpu to notebooks node for deep learning.
- snakemake workflow for hyperparameter search
- docker image to start jupyter lab session