t3hseus/ariadne

ariadne

Setup environment:

1. Install miniconda
2. Choose the needed environment, with (environment_gpu.yml) or without CUDA (environment_cpu.yml)
3. If you are installing with CUDA, check your drivers first (here)
4. Run

conda env update --file environment_cpu.yml --name ariadne_cpu
conda activate ariadne_cpu

note: to delete the conda environment run

conda remove --name %NAME% --all

Training

To start training procedure execute train.py script and pass to it a path to the training configuration file

python train.py --config resources/gin/tracknet_v2_train.cfg

Ariadne uses logging, so to specify the log level one should use --log parameter. E.g.:

python train.py --config resources/gin/tracknet_v2_train.cfg --log DEBUG

The default loggin level is INFO.

Run scripts on HybriLIT

There are several utility scripts to facilitate ariadne execution on the hydra JINR cluster:

1. scripts/hydra_slurm/hydra_cpu.sh
2. scripts/hydra_slurm/hydra_gpu.sh
3. scripts/hydra_slurm/govorun_gpu.sh

The main syntax is:

sbatch $SCRIPT_PATH $command_to_be_executed

Hydra

For example, to execute training script on the GPU queue of hydra cluster:

1. Verify that the miniconda has installed in the ~/miniconda3 or manually change the path in the script you want to execute. Row with source ~/miniconda3/etc/profile.d/conda.sh command

2. Run scripts/hydra_slurm/hydra_gpu.sh script

sbatch scripts/hydra_slurm/hydra_gpu.sh python train.py --config resources/gin/tracknet_v2_train.cfg

3. The slurm-jobid.out file with stdout will appear in the root directory.

GOVORUN

Executing a command on GOVORUN differs from executing Hydra commands only in the need to add a module for working with the supercomputer.

module add GVR/v1.0-1 && \
sbatch scripts/hydra_slurm/govorun_gpu.sh python train.py --config resources/gin/tracknet_v2_train.cfg