Anton Tsitsulin
Anton Tsitsulin
You can quite literally do np.fromfile('datafilepath', np.float32).reshape(num_vertices, dimensions) Please note that for performance I do not use custom node_ids in the data format (i.e. nodes are renamed). You can see...
In CSR node IDs are from 0 to 10312. I am not sure why do you have some "original" node IDs, blogcatalog was distributed as a matrix originally; the mapping...
In the arbitrary order: 2. Doing that will ruin portability for Windows, as there is not -openmp-simd flag that I would have used otherwise on gcc, clang and icc. Besides,...
2. It seems like OpenMP is supported by [most sensible](http://www.openmp.org/resources/openmp-compilers/) and few non-sensible compilers. I would not really agree that OpenMP is too much of an external dependency as so...
Hi, You can use `np.fromfile('embedding.bin', np.float32).reshape(num_nodes, embedding_dim)`. Note that this will load it into numpy array that has the same order as the converted graph, see https://github.com/xgfs/deepwalk-c/blob/master/deepwalk/cli.py#L69-L72 Best, Anton
Do you have an example graph to verify? Also, what other library you were using?
The directedness was tested on the node classification dataset. If you do not have a concrete example *and* the same method working for some other DeepWalk implementation, I am afraid...
Probably too late to answer, but I was able to reproduce the results with DeepWalk's evaluation code: https://github.com/phanein/deepwalk/blob/master/example_graphs/scoring.py Note that you are also feeding the correct number of classes to...
The shape of these two random arrays is (approximately) the same. The ~3-5 score you see is due to randomness of the approximation.