Jon McLean

I was confused by the documentation. The term "startup" appears to be overloaded. I assumed that the "initial startup run" mentioned in the docs above referred to the overall function...

When I set n_components to None so it would use the Zhu and Ghodsie selection procedure I got this error: "n_components must be

I am calculating the latent distribution test for the cross-product of the following graphs: GraphA has 1695391 edges GraphB has 463153 edges GraphC has 2101995 edges GraphD has 161548 edges...

This is the code I am running. "graphs" is a dictionary of id to NetworkX graph object. ``` embeddedGraphs = {} for id, graph in graphs.items(): logger.info(f'{id} contains {graph.number_of_edges()} edges....

@alyakin314 here are the vertex counts for each graph: GraphA contains 1695391 edges and 13106 nodes. GraphB contains 463153 edges and 7640 nodes. GraphC contains 2101995 edges and 12094 nodes....

I have ignored the smaller graphs. I think this is mostly a documentation & error message concern. I propose the following documentation changes: * Make the relationship between n_components and...

Similar PR https://github.com/openai/tiktoken/pull/407 that also includes gpt-4.5