search

UMD-ARLIS / Graph-Benchmarking-Project

This repository contains work towards creating a comprehensive cross-platform graph processing benchmark for Community Detection and Subgraph Isomorphism problems. The repository is NOT YET COMPLETE and should not be used outside of a preliminary development perspective.
2 stars 0 forks source link

updated_gpu_script.py (and gpu_script.py) conflate experimentation and analysis. #20

Open osullik opened 8 months ago

osullik commented 8 months ago

We should aim to modularize each section of our system.

Currently updated_gpu_script.py both runs the experiments and generates the reports.

The experiment runner component should go as far as outputting the raw results from a run to a directory that identifies which experiment was run. Ideally, all experiments across all platforms output in the same (or very similar) format.

The data analysis code should start by looking in those directories and pulling the raw data in for processing, ending with report generation.

Splitting the functions across experiment framework, and data analysis respectively will make it easier to understand the code, and allow us to re-use the same analysis code across all experiment outputs.

osullik commented 8 months ago

Soon to be deprecated and merged with the "Experiment" class, which has basic unit test coverage built in.