Detection of Mesoscale eddies using Convolutional Neural Network on Simulated Data

File Structure

\--CNN_eddy_detection
        \--static # All python scripts 
        \--notebooks # All jupyter notebooks    
        \--test #test notebooks or scripts
        -- interpolator.sh                    # interpolator slurm script
        -- README.md
        -- requirements.txt          # Required dependencies to run this program
        -- .gitignore    

Requirement Installation guide

# Clone the repo.
git clone https://github.com/LegoCreation/CNN_eddy_detection
cd CNN_eddy_detection

# Create environment to work in and activate it
$ conda create -n eddy-tracking python=3.8.16
$ conda activate eddy-tracking


# use pip to install PyEddyTracker
$ pip install pyEddyTracker==3.6.1 --no-cache-dir

# manually install/uninstall a couple of dependencies
$ pip install polygon3==3.0.9.1 --no-cache-dir # This step is cruial as previouly compiled polygon3 package will not align with the numpy version required by py-eddy-tracker
$ pip install xarray==2022.11.0
$ pip install dask==2023.2.0

## For tensorflow numpy==1.24.3 is required but for py-eddy-tracker numpy==1.22.4 is required. Hence cannot be used simultaneouly. Please switch the numpy version which using py-eddy-tracker 

$ pip install tensorflow==2.13.1


# Create a Kernel for jupyter notebook
$ conda install ipykernel
$ python -m ipykernel install --user --name eddy-tracking --display-name="eddy-tracking"

For usage of package please refer to scripts.

As notebooks are updated frequently and contain more scratch work, hence might create confusion. However, going through the notebook once is recommended for gaining better insight of overall process.