PPPL deep learning disruption prediction package

module load anaconda cudatoolkit/7.5 cudann openmpi/intel-16.0/1.8.8/64
source activate environment
python setup.py install

Where environment should contain the Python packages as per requirements.txt file.

python guarantee_preprocessed.py

Use Slurm scheduler to perform batch or interactive analysis on Tiger cluster.

For batch analysis, make sure to allocate 1 process per GPU:

#SBATCH -N X
#SBATCH --ntasks-per-node=4
#SBATCH --ntasks-per-socket=2
#SBATCH --gres=gpu:4

where X is the number of nodes for distibuted data parallel training.

sbatch slurm.cmd

The workflow is to request an interactive session:

salloc -N [X] --ntasks-per-node=16 --ntasks-per-socket=8 --gres=gpu:4 -t 0-6:00

where the number of GPUs is X * 4.

Then launch the application from the command line:

cd plasma-python
mpirun -npernode 4 python examples/mpi_learn.py

Note: there is Theano compilation going on in the 1st epoch which will distort timing. It is recommended to perform testing setting num_epochs >= 2 in conf.py.