Solar-Irradiance-Prediction

Evaluator

To evaluate the best model on the hidden test set on helios, the script ./scripts/evalulation.sh can be run.

./scripts/evalulation.sh {output_file} {config_file}

Setup

To keep the code clean and consistent, some linters are in place.

• flake8 Ensure pep8 standards and check some syntax errors.
• mypy Ensure there is no type error.
• pydocstyle Ensure the same documentation format is used across the project.
• black Ensure the same code formatting is used across the project.

Vs Code User

Defaults VsCode settings can be installed.

cp .vscode/settings.default.json .vscode/settings.json

Image dimension

All the images have dimension 650x1500

Stations Pixel Location (Mauvaise localisation?)

• BND: X = 688, Y = 188
• TBL: X = 371, Y = 186
• DRA: X = 169, Y = 252
• FPK: X = 374, Y = 33
• GWN: X = 660, Y = 296
• PSU: X = 883, Y = 174
• SXF: X = 533, Y = 118

Important paths.

Samples

Running

Training

To train a model, you can use the script run_model.py. On helios, you can either use ./scripts/run_model_cached.sh or ./scripts/run_model.sh with the same arguments.

A few examples:

python run_model.py --model Conv2D --train --lr 0.001 --skip_non_cached 
python run_model.py --model Conv3D --train --epochs 2 --seed 1234

Testing

To test a model on the test set, the same script can be used. The only difference is that you have to pass to argument --test {checkpoint}

An example while testing the Conv2D model with the 4th checkpoint.

python run_model.py --model Conv2D --test 4

Samples

Each gif has 10 images with 30 minute intervals between them on all channels.