DeepLabCut Workflow Cont.

Create a Python Script

• Create a python script for your DLC project.

$ nano <project_name>.py

• Copy and paste the following script, then make sure to update cfg and videofile_path with the appropiate paths.

import tensorflow as tf

import os

os.environ["DLClight"] = "True"

import deeplabcut

cfg = '<project_path>/config.yaml' # update

deeplabcut.create_training_dataset(cfg, Shuffles=[1])

deeplabcut.train_network(cfg,

shuffle=1,

saveiters=1000,

displayiters=1000,

maxiters=200000)

deeplabcut.evaluate_network(cfg, Shuffles=[1], plotting=True)

videofile_path = ['<video_file_path>'] # update

deeplabcut.analyze_videos(cfg, videofile_path, videotype='mp4')

deeplabcut.create_labeled_video(cfg, videofile_path, filtered=False)

deeplabcut.plot_trajectories(cfg,

videofile_path,

videotype='.mp4',

filtered=False)

prj_path = cfg.replace('/config.yaml', '/videos/')

os.rename(prj_path + 'plot-poses', prj_path + 'plot-poses-unfiltered')

deeplabcut.filterpredictions(cfg,

videofile_path,

shuffle=1,

trainingsetindex=0,

filtertype='arima',

p_bound=0.01,

ARdegree=3,

MAdegree=1,

alpha=0.01)

deeplabcut.create_labeled_video(cfg,

videofile_path,

videotype='.mp4',

filtered=True)

deeplabcut.plot_trajectories(cfg,

videofile_path,

videotype='.mp4',

filtered=True)

os.rename(prj_path + 'plot-poses', prj_path + 'plot-poses-filtered')

• When you are done, hit CTRL+O, ENTER, CTRL+X to save and exit.

Create a GPU Job

• Now we will submit the python script you created earlier as a GPU job. To do this, create a new file <project_name>.slurm.

$ nano <project_name>.slurm

• Copy and paste the following script, then update <username> and <project_name> to your username and project folder name.

#!/bin/bash

#SBATCH --time=08:00:00

#SBATCH --mem-per-cpu=1024

#SBATCH --job-name=TOXO-DLC

#SBATCH --partition=gpu

#SBATCH --gres=gpu

#SBATCH --constraint='gpu_v100'

#SBATCH --mem=16gb

#SBATCH --ntasks-per-node=21

#SBATCH --nodes=1

#SBATCH --mail-type=ALL

#SBATCH --mail-user=<email_address>

#SBATCH --error=/work/chaselab/<username>/dlc-chaselab/<project_name>/%J.err

#SBATCH --output=/work/chaselab/<username>/dlc-chaselab/<project_name>/%J.out

cd /home/chaselab/<username>

export HOME=/work/chaselab/<username>

export WORK=/work/chaselab/<username>

cd $WORK

module load rclone

module load cuda

module load anaconda

conda activate DLC-GPU

yes | python $WORK/dlc-chaselab/<project_name>/<project_name>.py

rclone copy $WORK/dlc-chaselab/<project_name> GoogleDrive:/dlc-chaselab/<project_name>

• Save CTRL+O ENTER and exit CTRL+X.

Submit Your GPU Job

• To submit the job, run:

$ sbatch <project_name>.slurm

Check Job Status

• To check your submitted job status, run:

$ squeue -u <username>

Job state (ST) codes: R - Running, PD - Pending: Job is awaiting resource allocation.

If you want to monitor the job, you can add watch before the squeue command (i.e., $ watch squeue -u <username>).