requirements.txt

# status 06/2024: tested on Windows 10; WSL 2.0; Ubuntu 18.04 with version in comments.
numpy # 1.26.4
matplotlib # 3.9.0
tqdm # 4.66.4
cartopy # 0.23.0
xarray # 2024.6.0
notebook # 7.2.1
netcdf4 # 1.7.1.post1
dask # 2024.6.2
bottleneck # 1.4.0
torch # 2.3.1; deep learning library. Will install CUDA if GPU is available. Might take a while >1GB packages. Need torch>=2.1.0, <2.3.0 for keras to work.
torchvision # 0.18.1
ipywidgets # 8.1.3
# hvplot # uncomment to create interactive plots nicely integrated with xarray
# jupyter_bokeh # uncomment to make hvplot run
# geoviews # uncomment to create geospatial plots in hvplot
scikit-learn # 1.5.0; for linear regression model and some data preprocessing
regionmask # 0.12.01; for plotting regional boundaries
streamlit # 1.36.0; for webdemo
xskillscore # 0.0.26; for error metrics on xarrays
nc-time-axis # 1.4.1; to plot monthly xarray data easily
line_profiler # 4.1.3; to prfile runtimes line by line
# climateset downloader
xmip # 0.7.2 
esgf-pyclient # 0.3.1
h5netcdf # 1.3.0
# zipfile 
flox # 0.9.8; for fast xr.Dataset.groupby() operations
keras # 3.4.0; for training cnn-lstm
wandb # 0.17.3; for logging experiments
huggingface_hub # 0.23.4; for downloading dataset
tensorflow # 2.16.1; Some components related Keras 3.0 with torch backend still rely on tf (here wandblogger and Dataset() class). Todo: remove this requirement after wandb and keras update.
ray # for parallel processing on cpu