This project aims to classify climber motions based on sensor data. The script provides functionality to train a model, pretrain a model, or finetune a model using pretraining.

• Python 3.7 or later
• PyTorch
• pandas
• numpy

1. Clone the repository

gh repo clone LucVermeer/gru-d

2. Navigate to the directory

cd gru-d 

3. Install the requirements

pip install -r requirements.txt

This script accepts a variety of command line arguments:

• --mode: Specifies whether to train, pretrain, coarse_pretrain or finetune the model (default is train)
• --seed: The seed for reproducibility (default is 42)
• --batch_size: Batch size for training (default is 32)
• --num_workers: Number of workers for the DataLoader (default is 0)
• --data_path: Path to the preprocessed data (default is grud/data.csv)
• --seq_len: Sequence length for training (default is 10)
• --step_size: Step size for training (default is 1)
• --model_name: Name of the pretrained model (default is coarse_model_42)
• --freeze_gru: Whether to freeze the weights of the GRU layers (default is False)

Run script from the parent directory:

cd ..

To run the script with the default parameters, use:

python -m grud.train

Example:

python -m grud.train --mode=finetune --seed=0 --batch_size=64

Evaluation in experiments.ipynb experiments.ipynb primarily focuses on the evaluation of our model's performance.

Luc Vermeer