finkandreas/dcomex-framework

The DComEX Framework

Getting started

To run Msolve and korali integration example use the docker image,

$ docker build github.com/DComEX/dcomex-prototype --tag dcomex
$ docker run -it dcomex bash

inside the container run

$ cd /src/tests/units
$ ./run.sh
$ cd /src/tests/validation/inference_heat/
$ OMP_NUM_THREADS=1 ./run_inference.py --num-cores 12 --num-samples 200

To run one Msolve simulation

$ bio 1 2 3
1.0851841319006034E-05
$ bio -h
Usage: bio [-v] [-c] [-s] k1 mu time

MSolve simulation of tumor growth.

Positional arguments:
  k1                   The growth rate of the tumor, in units of 1/second.
  mu                   The shear modulus of the tumor, in units of kPa.
  time                 The timestep to use in the simulation, in units of days.

Options:
  -v                   Print verbose output during the simulation.
  -c                   Output the MSolve configuration file and exit.
  -s                   Use a surrogate instead of the full simulation.
  -h                   Display this help message and exit.
Environment Variables:

  mph                  The path to the mesh of the simulated domain,
		       in .mph format.
Returns:
  The volume of the tumor, in units of cubic millimeters.

The online documentation is at ReadTheDocs.

Install as unprivileged user

Install korali, integration tools, and Msolve application

$ make 'USER = 1' 'PREFIX = $(HOME)/.local' lkorali
$ make 'USER = 1' 'PREFIX = $(HOME)/.local' lbin lib
$ make 'USER = 1' 'PREFIX = $(HOME)/.local' lmsolve

Run examples

$ python examples/graph/korali0.py
log evidence:  -2.490052761730272
1.0851841319006034E-05
$ bio -v 1 2 3
1.0851841319006034E-05
$ python examples/graph/analitical.py
graph.metropolis: accept = 0.724
graph.metropolis: accept = 0.684
graph.metropolis: accept = 0.7068

Open analitical.vis.png file

Running on Piz Daint

To replicate CI runs manually it is possible to pull the containers by logging in to Piz Daint and execute the commands

$ module load sarus
$ sarus pull IMAGE_NAME
$ srun --pty -C gpu -A GROUP_ID -N1 -n1 sarus run --mpi --tty IMAGE_NAME bash

This will drop you to a shell on the compute node inside the container. From there you can replicate running the commands as in ci/prototype.yml. It is important to make sure that the container image names match the naming */public/*, i.e. they must reside in a directory named public, only then anonymous access is possible.

Directory structure

• bin: integration utilities
• CI: definition of containerised build, test and deployment pipelines via CI/CD at CSCS
• docs: the source of the documentation
• examples: tutorials and examples
• graph.py: sample Bayesian graphs
• follow.py: trace function evalution and detect loops
• kahan.py: Kahan summation or compensated summation algorithms
• korali: a directory to build korali
• tests: unit and integration tests

Application

• msolve: Msolve application
• integration: integration of Msolve and korali

Results