Steps to replicate the results of the paper:
1- Create the Docker image and enter inside of the container:
-
Give execution permissions to the installation file:
chmod +x setup.sh -
Run the setup file, this steps can be very slow:
./setup.sh -
See the IDs of the running containers:
docker ps -
Enter inside of the created container:
docker exec -it {created_container_id} bash
2- Replicate the results obtained in the paper:
The experiments are configured with the Hydra library. To run an experiment, simply execute the following command:
python ./scripts/run_average_experiment.py --config-name {hydra_file}
Currently, the following files are implemented:
- aqi36.yaml
- metr-la_point.yaml
- metr-la_block.yaml
- pems-bay_point.yaml
- pems-bay_block.yaml
So, for example, if you want to run the experiment with the Metr-La dataset in the Point Missing scenario, you should execute the following command line:
python ./scripts/run_average_experiment.py --config-name metr-la_point
If you want to change the size of the time windows, simply go to the yaml file of the experiment you want to run and change the parameter scale_window_factor.
By default, experiments run with the TIMBA model, which is the one proposed in this paper, but you can test CSDI and PriSTI by executing the commands as follows:
- For CSDI:
python ./scripts/run_average_experiment.py --config-name {hydra_file} model_name=csdi
- For PriSTI:
python ./scripts/run_average_experiment.py --config-name {hydra_file} model_name=pristi