In this project, we aim to analyze and extrapolate the trends in land surface temperature over the past 20 years to accurately forecast surface temperature maps in near future.
Install the dependencies via:
pip install -r requirements.txt
For this project, we use the MOD11A2 dataset. Put all the raw .tif images together as shown below:
temperature-map-prediction (project root)
├── datasets
│ └── MOD11A2
│ └── <image1>
│ └── <image2>
│ └── <image1>
│ ...
...
Run the following command to train the model
python train.py --config <config_name>
where <config_name> is the config file name like transformer.json.
The default config has been set to transformer.json so no config argument is needed if training the Transformer model.
Run the following command for auto-reformatting:
bash ./scripts/auto_format.sh
Every commit made to the repository also goes through the same auto-reformatting on GitHub. If your remote branch is auto-reformatted, your new local commits may be rejected. Make sure to "pull the remote branch with rebase to your local branch" before pushing new local commits.
Add new dependencies to requirements.in.
requirements.in keeps track of only the top level dependencies.
requirements.txt is generated with the following command using the pip-tools package:
pip-compile requirements.in
Run this command whenever a dependency in requirements.in is updated. It automatically writes all the necessary
and compatible packages to requirements.txt.