Current weight plot method is time consuming for a large data set

[weikang9009]

The current plot method for W is quite time consuming for a large data set. An example is given in the notebook.

[ljwolf]

Send a PR! What’s there is real naive, and relies mainly on the data frame indexer rather than the precomputed id2i stuff. I was hoping to replace what’s there with the splot function @slumnitz wrote, but what you’ve got there is fine too.

On Wed, 31 Oct 2018 at 00:24 Wei Kang ***@***.***> wrote: The current plot method for W <https://github.com/pysal/libpysal/blob/master/libpysal/weights/weights.py#L1201> is quite time consuming for a large data set. An example is given in the notebook <https://gist.github.com/weikang9009/503be3a551c84d92f25f26425b461785>. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#123>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACJY805ijJpN6ayI25pbaiwLx3QsqKzsks5uqO3EgaJpZM4YDH4G> .

-- Levi John Wolf Lecturer in Quantitative Human Geography | University of Bristol Fellow | Center for Spatial Data Science, University of Chicago Steering Committee | Jean Golding Institute for Data Science, University of Bristol Director | Q-Step Program, School of Geographical Sciences, University of Bristol ljwolf.org

[darribas]

An alternative example with datashader can be found here:

https://gist.github.com/darribas/8d8f931a4414a52d82480db4f37c183e

IIRC their method is pretty quick, as it relies on numba, but it's an extra dependency. Maybe we could offer this as a first option and, if not present (or the user wants), fall back to the current one?

[weikang9009]

A quicker plotting method is given in splot, but not integrated into libpysal yet.

I think it is a great idea to offer the datashader functionality as the default option and falls back to the current splot function. In addition, I am thinking maybe we should move this (datashader function) to splot and deal with the logic/order over there?

[slumnitz]

The data shader visualisation looks awesome! Since it is static, we could definitely integrate it into splot.libpysal as is. I would suggest to try and keep the current plot_spatial_weights() API if possible. @weikang9009 in which cases would you suggest the current plot function should be used, compared to the data shader function? Maybe it would be best to completely replace it?

[weikang9009]

I think providing the fast plotting of spatial weights is great! But note that the function requires extra dependencies: datashader and numba - I think neither of them is a dependency for splot or libpysal at the moment? It might be more reasonable to treating these two packages as soft dependencies - if they are not installed, the current plot function will be used. This is also the strategy suggested by @darribas?

[slumnitz]

ok, I am happy to give it a go and play with both functions on the weekend. Would you prefer an automatic switch with a warning as soon as numba is not present to the current splot functionality? Or should we include an argument in the API to choose between the two options?

[github-actions]

Stale issue message