"TypeError: invalid type promotion" when plotting streamz.DataFrame with recent bokeh

[dcferreira]

I'm getting a very non-informative error when trying to do a simple plot from one of the streaming examples (slightly modified) from the documentation.
I have a notebook with 2 cells, the first one:

import pandas as pd
from streamz.dataframe import Random
import holoviews as hv
from holoviews.streams import Buffer
hv.extension('bokeh')

sdf = Random()

and the 2nd:

%%opts Curve [width=500 show_grid=True]

raw_dmap = hv.DynamicMap(hv.Curve, streams=[Buffer(sdf.x)])
raw_dmap

When executing the 2nd cell, I get the following traceback:

ERROR:tornado.application:Exception in callback functools.partial(<function wrap.<locals>.null_wrapper at 0x7f8a04257048>, <Future finished exception=TypeError('invalid type promotion',)>)
Traceback (most recent call last):
  File "/tmp/test/lib/python3.6/site-packages/tornado/ioloop.py", line 760, in _run_callback
    ret = callback()
  File "/tmp/test/lib/python3.6/site-packages/tornado/stack_context.py", line 276, in null_wrapper
    return fn(*args, **kwargs)
  File "/tmp/test/lib/python3.6/site-packages/tornado/ioloop.py", line 781, in _discard_future_result
    future.result()
  File "/tmp/test/lib/python3.6/site-packages/tornado/gen.py", line 1113, in run
    yielded = self.gen.send(value)
  File "/tmp/test/lib/python3.6/site-packages/streamz/dataframe/core.py", line 831, in _cb
    yield source._emit((last, now, freq))
  File "/tmp/test/lib/python3.6/site-packages/streamz/core.py", line 226, in _emit
    r = downstream.update(x, who=self)
  File "/tmp/test/lib/python3.6/site-packages/streamz/core.py", line 490, in update
    return self._emit(result)
  File "/tmp/test/lib/python3.6/site-packages/streamz/core.py", line 226, in _emit
    r = downstream.update(x, who=self)
  File "/tmp/test/lib/python3.6/site-packages/streamz/core.py", line 490, in update
    return self._emit(result)
  File "/tmp/test/lib/python3.6/site-packages/streamz/core.py", line 226, in _emit
    r = downstream.update(x, who=self)
  File "/tmp/test/lib/python3.6/site-packages/streamz/core.py", line 490, in update
    return self._emit(result)
  File "/tmp/test/lib/python3.6/site-packages/streamz/core.py", line 226, in _emit
    r = downstream.update(x, who=self)
  File "/tmp/test/lib/python3.6/site-packages/streamz/core.py", line 443, in update
    result = self.func(x, *self.args, **self.kwargs)
  File "/tmp/test/lib/python3.6/site-packages/holoviews/streams.py", line 389, in send
    self.event(data=data)
  File "/tmp/test/lib/python3.6/site-packages/holoviews/streams.py", line 328, in event
    self.trigger([self])
  File "/tmp/test/lib/python3.6/site-packages/holoviews/streams.py", line 154, in trigger
    subscriber(**dict(union))
  File "/tmp/test/lib/python3.6/site-packages/holoviews/plotting/plot.py", line 513, in refresh
    self.update(stream_key)
  File "/tmp/test/lib/python3.6/site-packages/holoviews/plotting/plot.py", line 492, in update
    item = self.__getitem__(key)
  File "/tmp/test/lib/python3.6/site-packages/holoviews/plotting/plot.py", line 230, in __getitem__
    self.update_frame(frame)
  File "/tmp/test/lib/python3.6/site-packages/holoviews/plotting/bokeh/element.py", line 802, in update_frame
    self._update_glyphs(element, ranges)
  File "/tmp/test/lib/python3.6/site-packages/holoviews/plotting/bokeh/element.py", line 751, in _update_glyphs
    self._update_datasource(source, data)
  File "/tmp/test/lib/python3.6/site-packages/holoviews/plotting/bokeh/plot.py", line 194, in _update_datasource
    source.stream(data, stream.length)
  File "/tmp/test/lib/python3.6/site-packages/bokeh/models/sources.py", line 367, in stream
    self._stream(new_data, rollover)
  File "/tmp/test/lib/python3.6/site-packages/bokeh/models/sources.py", line 471, in _stream
    self.data._stream(self.document, self, new_data, rollover, setter)
  File "/tmp/test/lib/python3.6/site-packages/bokeh/core/property/containers.py", line 383, in _stream
    data = np.append(self[k], new_data[k])
  File "/tmp/test/lib/python3.6/site-packages/numpy/lib/function_base.py", line 5166, in append
    return concatenate((arr, values), axis=axis)
TypeError: invalid type promotion

Note: For some reason, each 1/10 times I execute the cell, I don't get the traceback, but the stream still stops, indicating that the error did occur. If you don't get a traceback while reproducing this, just try again.

At some point, the input pandas DataFrame is converted into a dictionary, changing the type of the index generated by streamz.dataframe.Random to float (rather than a pandas timestamp).
Since it expects the same type as the static example provided in the beginning of the stream, this TypeError is thrown.

Workaround: This happens with bokeh==0.12.14 (latest), but if I downgrade to bokeh==0.12.10, it works as expected.

My setup:

dferreira:/tmp$ virtualenv -p ~/.pyenv/versions/3.6.4/bin/python test
dferreira:/tmp$ cd test
dferreira:/tmp/test$ source bin/activate
(test) dferreira:/tmp/test$ pip install jupyter bokeh streamz matplotlib pandas git+git://github.com/ioam/holoviews.git
(test) dferreira:/tmp/test$ jupyter notebook
# test example, get error
(test) dferreira:/tmp/test$ pip install -U bokeh==0.12.10
(test) dferreira:/tmp/test$ jupyter notebook
# test again, everything works

[philippjfr]

This is fixed by #2383, which has already been applied in the last public release of holoviews (version 1.9.5). I've also just submitted a fix to bokeh to fix the underlying issue: bokeh/bokeh#7609

[dcferreira]

Oops, should have searched better. Thanks for the fix :)