DNN

Diffusion neural network

Setting up

To create a conda environment, cd to the directory, and
- run conda env create -f environment.yml on a local machine and
- run conda env create -f environment_cbica.yml on the CBICA cluster.
- To download conda, visit anaconda.com.
To activate environment in the command line,
- run conda activate dnn for a local machine and
- run conda activate dnn-cbica on the cluster.
To activate environment in PyCharm
- Open PyCharm. Open dnn directory as a project.
- Then, select "Preferences" > "Project: dnn" > ⚙️ > "Add..." > "Conda Environment" > "Existing environment" > "/Users/username/anaconda3/envs/dti-enn/bin/python"

Download the HCP covariates file (link), and save as dataset/hcp/res/hcp_covariates.csv.
Get AWS credentials to HCP (see link).
- Save to file ~/.aws/credentials in you home folder with conents
```
[default]
aws_access_key_id = (your HCP access key here) 
aws_secret_access_key = (your HCP secret access key here)
```
- Using CLI, run aws configure.

List the subjects to download and process in dataset/hcp/dti/scripts/conf/subjects.txt.
Go the root folder of the project.
Run python schedule (local/debug/cbica_cpu) dataset/hcp/dti/scripts/conf/args.ini.
- The downloaded and post-processed files will be stored in ~/.dnn/datasets/hcp, accessible from your home folder.
- On the UPenn CBICA cluster, you can use Cassiano's processed data in /cbica/home/beckerc/.dnn/datasets/hcp/processing. Set local_processing_directory in your .ini experiment configuration file in experiments/hcp/conf.

Select the arg.ini file describing your experiment.
Go to the root folder of the project.
Run python schedule (local/debug/cbica_cpu/cbica_gpu) experiments/hcp/conf/args.ini.
- The list of subjects for training and testing is located in experiments/hcp/conf/(test|train)_subjects.txt.
  - The results of you experiments will be stored in '~/.dnn/results' under a folder named according to the data and time when you then experiment was run.
- See list of covariates here.

See README.md on dataset/synth.