DynamicMap

[philippjfr]

This PR adds the new DynamicMap class, which allows generating frames on the fly when working with widgets. The DynamicMap has two operating modes or intervals:

• 'open': In open mode the ScrubberWidget allows the user to request a new frame via the play and next frame buttons. When creating the DynamicMap in open mode the user should supply an iterator that returns a new key and Element when next is called. This mode allows for streaming data for example from a simulator.
• 'closed': In closed mode the user should define the valid Dimension ranges or values. Using the SelectionWidget the user may then select any value in the space defined by the Dimensions, allowing much larger parameter spaces with near continuous sampling to be explored.

Both modes already work but there are a few to-do items left:

1. Currently the cache is infinite which means that all frames are buffered both server-side and client-side, there should be options to limit the buffer size and only buffering every n-th frame.
2. Currently nothing communicates a StopIteration to javascript so the scrubber will continue running and expand the slider even when the DynamicMap has run out of new frames.

[jlstevens]

I've been working through the proposal above and making good progress. One quick bug report for Dimensions though:

kdims=[hv.Dimension('size', range=(1,5), type=float)

This results in a slider that snaps to integers despite the explicit type declaration saying otherwise.

[jlstevens]

Ok, I've gone and rewritten DynamicMap according to the proposal above.

I am quite happy with it! Not having the specify the mode explicitly is very nice, making DynamicMap easier to use. It is currently working with all the tests I have for all the different modes.

As it is working with an improved API, I am happy to see it merged whenever you like. There are a few outstanding issues, most of them are minor:

Bug fixes

• The dimension type issue above.
• An odd bug where the first frame is the same as the last frame.

To implement

• When using __getitem__ with a key set, DynamicMap should regenerate the data (closed mode only).
• Caching needs to be improved on the client-side (i.e in the Javascript code).

These are all small jobs that I am happy to see implemented on master.

[philippjfr]

Okay I've fixed all the bugs you mentioned and have got the tests passing. The other two items are features and can be implemented on master (or as separate PRs). I'll go ahead and merge.

[jlstevens]

Ok great! We may still want to use this PR to decide how we want caching on the client-side to work...