Added IEX trading example

[jlstevens]

This example visualizes trades made on the IEX exchange, building up to the following dashboard:

You can zoom in to reveal the metadata of individual trades:

There are three notebooks:

1. IEX_to_CSV.ipynb which explains how the CSV files were generated from the pcap files offered by the exchange. This notebook is for reference and should not be tested in CI (the pcap files are very large).
2. IEX_trading.ipynb which introduces the data and basic concepts needed to build a trade explorer.
3. IEX_stocks.ipynb which generalizes IEX_trading.ipynb to build a useful dashboard.

On S3 there is IEX_2019-10-25.csv in addition to IEX_2019-10-21.csv: I am planning to use the former in yet another notebook aimed at comparing the two datasets. At this point, I think the material here is already complete enough to present without that extra example.

[jlstevens]

Some other notes:

• nodejs is listed as a dependency to build panel's Bokeh models. This shouldn't be required anymore (in the new panel dev release). I'll test that I can remove this now.
• This holoviews PR will need to be merged and a dev release of holoviews used until those changes are released.

[jlstevens]

@jbednar @philippjfr Some other annoyances that ought to be fixed:

1. I have to loosen hv.config.image_rtol or datashader would break when zooming in (datetime related)
2. I want to be using rasterize instead of spikes_aggregate but that would require holoviews to pass the parameters down (TODO).
3. The formatting of volume on the colorbar is pretty ugly scientific notation when zoomed out.

I may expand this list if I remember anything else!

[jbednar]

Nice!

[philippjfr]

You've switched to summing the trade volumes so maybe this should also state the fraction of the overall volume?

[philippjfr]

For that matter maybe select the top stocks by volume as well.

[jlstevens]

Not a bad suggestion even though the number of trades is what tests datashader performance, not the volume. The top stocks by trade counts are what will push datashader the hardest here and any ten stocks will do (all I need to do is select between some stocks). It might also be less cognitive overhead if I don't switch the names/symbols of the stocks inspected. @jbednar What do you think?

[jbednar]

Here you're distinguishing number of trades from number of stocks traded? I guess I don't have much opinion; I'll leave it up to you as author of the example.

[jlstevens]

@jbednar @jsignell Ready to merge if we are ok depending on my holoviews branch for now (I'll try to get that merged soon). Will need updating once that code makes its way into a holoviews release.

[jsignell]

There should be a badge here that indicates that this project is dev.