Speed issues with HV -> Bokeh -> HTML about 2ioa0-dbl-hti-webtech

[tomudding]

It might actually be due to my setup. I'll test conversion of the bokeh.document.document.Document to JSON and directly pass it to the page.

If that doesn't work I'll try testing an API which we can request data from using AJAX (XHR) such that we can load the page and then once it is loaded start loading/retrieving the actual plots.

[Yvee1]

I tried your code and I have a correction, though maybe you know it already. You mention that it takes 60s to generate the HTML, but actually if you place the timer after the datashading and edge bundling you see it takes (only) 2 seconds. So in that case doing the datashading and edgebundling is what takes long. If you remove datashading and edgebundling however the generating of the html goes up to 8.2 seconds, so in total it then takes about 9 seconds to load the page. Though rendering 1000 nodes, with so many edges is also not that useful for data analysis so maybe we should only a subset of the nodes/edges, and all the data is only shown when filtered down to an amount where it is useful.

[tomudding]

Dammit, didn't think about that. I was just using it because "it should improve the running time".

And we indeed should not render 1000 nodes, probably only the nodes and edges with weight (connections) >= some number derived from the data-set.

[TechWizzart]

Yeah rendering all nodes and edges in a radial layout is far from useful, therefore I wanted to implement a slider that adjusts the value that @tomudding mentions. The edgebundling and datashading are things I have not looked at yet so it could very well be that these algorithms are far from optimised just as Bokeh's arrows are far from optimised.

[tomudding]

Closing for now until we have properly implemented the nginx/gunicorn/flask/bokeh part and python bokeh callbacks.