![]() ![]() However, for the shortest path problem (not analysed in their paper) it lags behind all other packages. That is consistent with the findings of their research paper where they claim that using some of the latest state of the art algorithms led to their processing speed being faster by an order of magnitude. For the k-core decomposition it is also 10 times faster than all other competitors or 2000 times networkx. On the pokec dataset it takes just 0.2s to run the page rank algorithm (graph-tool: 1.7s, igraph: 59.6s, snap: 19.5s). When networkit is fast, it is extremely fast. Networkit and graph-tool takes the top spot in most of the tests with graph-tool having the shortest run time for the single source shortest path and connected components problems and networkit winning the race for k-core and page rank. The other 3 packages should be using C libraries to read the files which result in better performance. I was reading the datasets as a tab delimited file and graph-tool basically uses a Python code to parse the input. ![]() Looking at the plots above, graph-tool and networkit loads data much more slowly than the other two libraries. Here are the run times of the remaining four packages:įull results can be seen from the table below: dataset Hence, I left it out of the comparison plots. Page rank took more than 10 minutes to run compared to 1 minute for igraph. 2 For example, it took 67s to run the single source shortest path problem on the Pokec dataset compared to 6.8s for networkit (the next slowest). Across all computation tasks and for all datasets it is around 10 times slower than the slowest library. Networkx is much slower than any of the other libraries. ResultsĪll timings reported are normalised to reflect the run time for a single run of the task. Most of them were written in 2015/2016 and it will be interesting to see if anything has changed since then.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |