Shell scripts

We provide several shell scripts for easy managing the experiments.

File	Description
download_data.sh	Download the training data
process_data.sh	Save the training data to shared memory
download_models.sh	Download the pretrained models
setup_exp.sh	Set up a new experiment with default settings
run_train.sh	Train a model
run_inference.sh	Run inference from a trained model
run_interpolation.sh	Run interpolation from a trained model
run_exp.sh	Run an experiment (training + inference + interpolation)
rerun_exp.sh	Rerun an experiment (training + inference + interpolation)

Below we assume the working directory is the repository root.

Download the training data

./scripts/download_data.sh

This command will download the training data to the default data directory (./data/).

Save the training data to shared memory

./scripts/process_data.sh

This command will store the training data to shared memory using SharedArray package.

Download the pretrained models

./scripts/download_models.sh

This command will download the pretrained models to the default experiment directory (./exp/).

Set up a new experiment with default settings

./scripts/setup_exp.sh "./exp/my_experiment/" "Some notes"

This command will create a new experiment directory at the given path ("./exp/my_experiment/"), copy the default configuration and model parameter files to that folder and save the experiment note ("Some notes") as a text file in that folder.

Train a model

./scripts/run_train.sh "./exp/my_experiment/" "0"

This command will look for the configuration and model parameter files in the given experiment directory ("./exp/my_experiment/") and train a model according to the configurations and parameters on the specified GPU ("0").

Run inference from a trained model

./scripts/run_inference.sh "./exp/my_experiment/" "0"

This command will look for the configuration and model parameter files in the given experiment directory ("./exp/my_experiment/") and run inference from the trained model on the specified GPU ("0").

Run interpolation from a trained model

./scripts/run_interpolation.sh "./exp/my_experiment/" "0"

This command will look for the configuration and model parameter files in the given experiment directory ("./exp/my_experiment/") and run inference from the trained model on the specified GPU ("0").

Run an experiment (train + inference + interpolation)

We provide a simple wrapper for a typical experiment scenario—first we train a model and then we run inference and interpolation from the trained model. In this case, you can simply run the following command.

./scripts/run_exp.sh "./exp/my_experiment/" "0"

This command will look for the configuration and model parameter files in the given experiment directory ("./exp/my_experiment/") and run the experiment (train, inference and interpolation) on the specified GPU ("0").

Rerun an experiment (train + inference + interpolation)

./scripts/rerun_exp.sh "./exp/my_experiment/" "0"

This command will remove everything in the experiment directory except the configuration and model parameter files and then rerun the experiment (train, inference and interpolation) on the specified GPU ("0").