ClimateBench-Plus

Website

To see what's our project about, here is the link to website: (https://jackljk.github.io/DSC180B-website/)

Data

The preprocessed training and test data can be obtained from Zenodo: (https://zenodo.org/records/7064308).

DKL Gaussian Process Regression

Setting up the environment for DKL Gaussian Process

To run the code, first you need to have anaconda install on your machine. then navigate to the DKL Gaussian Process folder (or the directory that contains the environment.yml file) and run the following command:

conda env create -f environment.yml

This will create a conda environment in the name CB2 with all the required packages. Then, activate the environment by running the following command:

conda activate CB2

Running the DKL Gaussian Process

After activating the environment, in the terminal cd to the DKL-Run-Model-Script folder. In the config file config.yaml you can set some of the parameters of the model and whether you want to train a model(train) with a set of parameters or perform a hyperparameter search (search). More information about the parameters and the settings to run the model can be found in the config.yaml file. Then to run the model, run the following command:

python main.py --config configs/config.yaml 

If set to search the model will setup a hyperparameter search using the raytune and return a csv file for all the trials performed in the directory specified in the config.yaml file (default is hyperparam_results/). It will also save the predictions of the best model in the directory specified in the config.yaml file (default is model_results/). If plot is set to True in the config.yaml file, the model will also plot the predictions of the best model and save them in the directory specified in the config.yaml file (default is model_results/).

Extras

• Some of my results and predictions from models that I have trained can be found in the results directory including the final results that I used for the validation of the model.
• The notebooks directory contains the notebooks for each of the Final individual models which I trained and performed my hyperparameter search on.
• The tests directory contains my other attempts to implement the Deep Kernel Learning model and which did not work out as expected.

XGBoost Regression

To run the code, after creating a basic conda environment, go to the requirement.txt file in XGBoost folder and run the following command:

pip install -r requirements.txt

You need the utils.py file in XGBoost folder to prepare the preprocessed data and save the training and test data by a certain output path.

Running XGBoost

Replace the parameters, and run the following command to run the model:

python xgboost_main.py

CNN LSTM with Physics-Informed Loss

To run the code, first you need to have anaconda install on your machine. then navigate to the PINN folder and run the following command:

conda env create -f environment.yml

This will create a conda environment in the name ClimateBenchPlus with all the required packages. Then, activate the environment by running the following command:

conda activate ClimateBenchPlus

After activating the environment, you can change the config file config.yaml to change what variable you want to predict for. Then to run the model, run the following command:

python main.py --config config.yaml