CSED499I

• 24' Fall Research Project I @ POSETCH MLLab
• Jiwoo Hong, POSTECH CSE 22
• Research Advisor: Prof. Dongwoo Kim

Overview

(To be written)

Instructions

[Optional] graphcast

• You can skip this entire step if you want to run this repository right away.
• Use graphcast_prediction_saver.ipynb (recommended to run on Google Colab)
  • On cell Choose the model, select Mesh size: 4, 5, 6 for each run, keeping other values as default
  • On cell Get and filter the list of available example datasets, select source: era5, date: 2022-01-01, res: 0.25, levels: 13, steps: 01 (default choice)
  • On cell Choose training and eval data to extract, continue with the default choice
  • Keep running the cells until the end of the notebook, it might take some time
• Download exported .nc datasets on Colab to ./ directory

GINR

• [Optional] Open ./npy_processor.ipynb to make dataset files (.npy/.npz) and place them as below

  ./GINR/dataset/
    ├── gcm4/
    │   ├── fourier_m4.npy, points_m4.npy
    ├── gcm4to5/
    │   ├── fourier_m4.npy, points_m4.npy
    │   └── npz_files
    │       └── data_m5.npz
    ├── gcm4to6/
    │   ├── fourier_m4.npy, points_m4.npy
    │   └── npz_files
    │       └── data_m6.npz
    ├── gcm5/
    │   └── fourier_m5.npy, points_m5.npy
    ├── gcm5to6/
    │   ├── fourier_m5.npy, points_m5.npy
    │   └── npz_files
    │       └── data_m6.npz
    ├── gcm6/
    │   └── fourier_m6.npy, points_m6.npy
    └── targets/
        └── data_m5.npz, data_m6.npz
  

  • rename all _m4, _m5, _m6 to make files fourier.npy, points.npy, data.npz, except for targets/ directory
  • These are already provided in the repository.
• Install dependencies (Linux is required, virtual environment is recommended)

  cd ./GINR
  # Create your virtual environment here
  pip install -r requirements.txt

• Build and install PyMesh (Again, Linux is required, virtual environment is recommended)

  git clone https://github.com/PyMesh/PyMesh.git
  cd PyMesh
  git submodule update --init
  sudo apt-get install libmpfr-dev libgmp-dev libboost-all-dev
  echo "SET(CMAKE_C_FLAGS " -fcommon ${CMAKE_C_FLAGS}")" >> ./GINR/PyMesh/third_party/mmg/CMakeLists.txt # Handle CMake error
  ./setup.py build
  ./setup.py install

• Train models and evaluate them with commands inside ./GINR/run.sh
  • Check ./GINR/run.sh for more information
  • Do not run it on the shell directly
• Evaluate and generate MSE comparison plots by executing ./GINR/eval_mse.py

Based Researches

• Graphcast: Learning skillful medium-range global weather forecasting
• GINR: Generalised Implicit Neural Representations