census-scvi

Instructions for running in AWS

1. Launch an instance with the Deep Learning Base GPU AMI. Note that this AMI currently only supports P5, P4de, P4d, P3, G5, G3, and G4dn instances.

2. Log into the instance and configure GPUs to run inside Docker containers

   sudo nvidia-ctk runtime configure --runtime=docker
   sudo systemctl restart docker
   

3. Pull the image:

   docker pull martinkim0/scvi-tools:py3.11-cu11-autotune-main
   

4. Mount the AWS drive to a directory

   sudo mkdir /data
   sudo mount /dev/mapper/name-of-drive /data
   

5. Clone this repository

   git clone https://github.com/YosefLab/census-scvi.git
   

6. Run the container in detached mode with a mounted volume

   docker run --name autotune --rm --gpus all --volume /data:/data -dit martinkim0/scvi-tools:py3.11-cu11-autotune-main /bin/bash
   

7. Copy the repository into the container

   docker cp census-scvi autotune:census-scvi
   

8. Execute the autotune script in the container

   docker exec -d autotune python /census-scvi/bin/autotune_scvi_v2.py --adata_path path_to_adata --batch_key batch_key --num_cpus num_cpus --num_gpus num_gpus --experiment_name experiment_name --save_dir /data
   

9. After the experiment finishes, you can stop the container

   docker stop autotune
   

10. All logs are stored in the save_dir argument passed into (6)