Experiment Framework

The experiment framework will become our way of building, monitoring, and executing experiments.

• Experiment
  • CPU_Experiment.py
  • Experiment.py
  • GPU_Experiment.py
• Experiment Runner
  • Experiment_Runner.py
• Tests
  • Test_Experiment.py
  • Test_Experiment_Runner.py
• Run_SGM_CPU_Medium.py
  • Example Text
• Run_SGM_CPU_Small.py
  • Example Text
• Run_SGM_GPU_Small.py
  • Example Text

Getting started

Activate the environment, which should be done with something like:

source ../../graph_benchmark_env/bin/activate

Create it at this location if you have to.

If not part of the environment already, install NetworkX:

pip install networkx

Change into the test directory:

cd tests

Run the current tests:

python test_experiment.py

Current Functionality

The experiment class tracks the common metadata across all types of experiments. It is capable of:

1. Tracking all relevant metadata.
2. Loading a graph from a file and generating summary metadata.
3. Logging timing information.
4. Profiling the operating system for Memory, CPU, Operating system, etc.
5. Dynamically creating required output directories for experiments and dumping experiment metadata to .json.

Future Work

1. Add Subclasses for CPU_Experiment and GPU_Experiment.
2. Add an “experiment runner” class that, given a .JSON of all required experiments, executes them according to provided parameters.

Testing Policy

No new features without a failing test.

No merging to main without all tests passing.