Skip to content

Milestones

List view

  • Sprint 4: Use case for Europe

    - Include E-OBS, PEP725, Phenocam, MODIS - Area around Germany (?) - Scientifically meaningful experimental setup, - Brainstorm on paper outline - MERF, EBM, MERF+EBM, SKLearn, Pycaret builtins, pyphenology - *Also include physical models* (should adhere to sklearn API) - Update of documentation Follow up: - Mini sprint (wrap up stuff) in August/September - Sprint 5 (~end of October): - Evaluation metrics / strategy - Refactoring/generalizing MERF - Explainability of ME-EBM - Dataset loading -> stac/intake/... - Paper outline

    No due date
    •16/16 issues closed
    100% complete0 open16 closed

  • Sprint 3: Complete ML workflow from start to finish

    By the end of this sprint, we'd like to be able to run a workflow from start to finish, i.e. downloading, loading, transposing, dimensionality reduction (by time averaging), machine learning, and cross-validation. Suggested workflow: Datasets: - NPN (which species?) - MODIS (which variables/bands/products?) - Daymet (which variables?) Area: - BBOX with ~15 stations inside it and several modis pixels (Dakota?) - Specify bbox --> get NPN stations --> get modis and daymet pixels for those stations Time: - 2015 - 2020 - Monthly averages (for now) Train/test splitting: - Random / shuffle split (70% train 30% test) ML (hardcoded/default hyperparameters): - scikit learn (linear regression) - scikit learn (random forest) - merf - EBM Output: - Score: RMSE, MAE - Print parameters or persist model to disk

    Overdue by 2 year(s)
    •
    Due by May 4, 2023
    •6/7 issues closed
    85% complete1 open6 closed

  • Sprint 2: ML with existing algorithms

    Sprint 2: 27-31 March Sprint goal: MVP of phenological modelling workflow 1. Review previous sprint and modify/improve data tools: - Fix remaining issues - Better documentation for the supported data sources - Clear and working installation instructions for different use cases (pc/mac, crib, etc.) 2. Add ML workflows to the package - Develop needed data preparation tools for modelling (cleaning, merging, splitting,…) - Modelling (fit+predict) using various data sources and algorithms: - scikit-learn - pyPhenology - MERFs - EBM - Two example cases based on [paper](https://www.sciencedirect.com/science/article/pii/S1470160X21007913): - Predict DOY of first bloom of network (PPO, NPN, PEP725, Phenocam) observations based on ... - Predict MODIS greenness index based on temperature data (point based, perhaps later extension to raster based). - An example notebook (paper demo?) - Model evaluation

    Overdue by 2 year(s)
    •
    Due by March 31, 2023
    •17/17 issues closed
    100% complete0 open17 closed

  • Sprint one: data retrieval

    First sprint will focus on streamlining data download from various sources

    Overdue by 3 year(s)
    •
    Due by February 17, 2023
    •15/17 issues closed
    88% complete2 open15 closed