Batch property updates fail

[wxd]

Python version: 3.8
ipyleaflet version: 0.13.1 used in a vanilla Jupyter notebook

A question, not a bug: I have a layer group of ~3000 polylines, which I need to re-style in response to user input. I'm only learning ipyleaflet, so I do it naïvely in a loop:

for obj_id, obj_data in data:  # len(data) > 3000
    line: ipyleaflet.Polyline = self._obj_lines[obj_id]

    if obj_data["x"] > user_input:
    	line.color = 'red'
    	line.weight = 10
    else:
    	line.color = 'black'
    	line.weight = 1

Sometimes (not always) these updates get stuck half-way (see the GIF; no line is supposed to stay blue) and nothing short of recreating the whole map object helps, certainly not re-executing the update code.

My guess is that this approach generates too many individual update messages between the kernel and the front-end which leads to some communication breakdown or state discrepancy. If that is correct, what are the better options from the Python side?

• is it possible to manually trigger "state synchronization" to ensure update completion?
• can/should I manually batch updates?
• should I switch to a GeoJSON layer? (GeoData doesn't seem to support dynamic styling, if I'm correct)

[martinRenou]

My guess is that this approach generates too many individual update messages between the kernel and the front-end which leads to some communication breakdown

That sounds very much correct. That would have been my first guess as well. There is a message rate limit in Jupyter comms, which is something like 1000 msgs/second I think.

You can change this rate limit using the following argument to Jupyter Notebook: https://stackoverflow.com/questions/43490495/how-to-set-notebookapp-iopub-data-rate-limit-and-others-notebookapp-settings-in

I am not sure I would recommend doing it though, but I have no best idea for this (apart from putting a time.sleep call somewhere in your code in order to reduce the rate manually, but that is no better solution).

Another solution might be to represent all your data as a single GeoJSON widget. Each feature would have its own style in the JSON structure. Then you only send one single message to update the entire JSON which contains all the lines. Not only that would fix your issue, but it might as well be faster in terms of performances (you send only one update message instead of 3000).

Edit: Sorry, I did not see you were mentioning GeoJSON. Actually I think that would be the best approach for fixing this. GeoData should support dynamic styling as it inherits from GeoJSON, but to be confirmed.

[wxd]

Thank you for a quick reply!

Is changing style_callback supposed to have any effect on the existing map layer? It doesn't seem to do anything in my case: layer.style_callback = new_cb does update layer.data but the new style does not get applied.

Does it mean that I should update the style properties in the data and do layer.data = updated_data?

[martinRenou]

The fact that it does not update itself when you assign the style_callback might be a bug. Could you provide some code that I could easily run to reproduce your issue?

[wxd]

https://gist.github.com/wxd/8142eb1fb21a07d91073499263e06e0a

After [5] the blue lines are drawn; after [6] no visual changes.

Versions (all latest versions I got from pip install two weeks ago or so):

• Python 3.8.3
• ipykernel==5.3.2
• ipyleaflet==0.13.1
• ipython==7.16.1
• traitlets==4.3.3
• Chrome 83.0.4103.116
• macOS Catalina 10.5.5

[martinRenou]

This piece of code should work indeed. The following PR should fix it: #658. I tried locally and could see it working but if you have time to try it out too that would be greatly appreciated :)

[wxd]

I can confirm that it works. Thank you!

[martinRenou]

Sure! Let's merge this PR and include it in the next release then :)

[martinRenou]

ipyleaflet 0.13.2 is out with the fix. You should be able to install it with pip now. If you are using conda you will need to wait a little bit.

[wxd]

Re-opening the issue: in 0.13.2 as well as 0.13.2, updating the style callback does something weird; see the GIF. It appears that a "new layer" with the new style gets drawn on the map, while the "old layer" gets hidden. However, the "new layer" seems to exist only in the JavaScript world, not in Python.

I don't think that just applying #658 on top of 0.13.1 had the same effect, but I'm not 100% sure.

[martinRenou]

This looks like a different issue. Would you mind opening another issue and provide some code to reproduce?