Clone this Git repository on E2 and switch to the release branch:
ssh chXXXXXX@e2.tch.harvard.edu
git clone git@github.com:jamesqo/AgePrediction.git
git checkout release
Create and activate a virtual environment for your project:
cd /path/to/my/project
module load anaconda3
conda create -n <env_name> python=3.6
conda activate <env_name>
Install the module into your virtual environment:
pip install -e /path/to/AgePrediction
After you've finished writing your script, create a file called slurm in your project directory with the following contents:
#!/bin/bash
#SBATCH --partition=bch-gpu
#SBATCH --time=120:00:00
#SBATCH --nodes=1
#SBATCH --gres=gpu:Titan_RTX:1
module load anaconda3
source activate <env_name>
python my_script.py "$@"from age_prediction.eval import predict_ages
"""
Parameters:
- filenames: List of filenames of the MRI images.
- architecture: CNN architecture to use. 'resnet18' or 'vgg8'
- age_range: The age range of the dataset that the model was trained on. '0-100'
- sampling_mode: The sampling strategy that was used during model training. 'none', 'over', 'under', 'scale-up', or 'scale-down'
- weighting: The reweighting strategy that was used during model training. 'none', 'inv', or 'sqrt_inv'
- lds: Whether or not label distribution smoothing (LDS) was used during model training. True or False
- device: Where the model should be evaluated. 'cuda' or 'cpu'
Returns: a Pandas DataFrame containing the predicted ages
"""
predict_ages(filenames,
architecture=...,
age_range=...,
sampling_mode=...,
weighting=...,
lds=...,
device=...)To submit a new job:
sbatch slurm <program args>
To view a list of running jobs:
squeue -u $USER
To cancel a job:
scancel <job_id>