Using the index of a data frame as a (default) x coordinate when providing a data source

[Krastanov]

For some data frames the index has quite useful semantic meaning. Particularly so for datetime indices. It would be useful if the api has an option to use the index as a coordinate.

Example given here: https://stackoverflow.com/questions/37904231/plot-a-pandas-dataframe-using-the-dataframe-index-for-x-coordinate-in-bokeh/37905026#37905026

[birdsarah]

I believe this is solved, can you try the latest dev build

[Krastanov]

I tried to pip install from github (pip install --upgrade git+...), and I also tried to clone and then do pip install -e . from the bokeh folder, but both of those hang up. Is this an unrelated bug? Or is this type of pip dev installation unsupported? In short: I am sorry, I was unable to set up the dev environment and it seems it will take me some time (I will try later in the week).

Could you point me to the reference for the dev branch (if it is uploaded somewhere)? You mentioned that this feature is already implemented, but you did not say what the proper arguments for the function call are.

[bryevdv]

Pip install from github will not work because pip is not capable of building the coffeescript and typescript components. There are simple instructions for installing pre-built dev builds in the docs:

http://bokeh.pydata.org/en/latest/docs/installation.html#developer-builds

[Krastanov]

Thanks for the pointer.

It is still not working in the dev release:

ds = ColumnDataSource(df) # contains a datatime index and the column `avg`
p = figure(x_axis_type='datetime')
p.square(source=ds, y='avg')
output_notebook()
show(p)

The code above produces a plot where all data source entries are in the correct y position, but all of them are at x=0. How can I specify that the x coordinate should be the data source index?

[bryevdv]

does x="index" work? I can't remember, ping @fpliger

[Krastanov]

x='index' gives this error message in jupyter:

Javascript error adding output!
Error: attempted to retrieve property array for nonexistent field 'index'
See your browser Javascript console for more details.

[canavandl]

Hi @Krastanov

The issue is that you have to specify which column should be the "x" column. If you don't specify the "x" value, the default behavior in bokeh.plotting is to try to find a column called "x" in your ColumnDataSource (which doesn't exist).

One tricky thing here is that you're using a named index ('timestamp') in pandas. That name is carried over when you create a ColumnDataSource, so that your source probably looks like:

ds = ColumnDataSource(df)
print(ds.data) # this lets you inspect the data dict
# the ts_n values would be the actual timestamps from the df
> {'timestamp': [ts_1, ts_2, ts_3, ts_4, ts_5], 'avg': [0.9, 0.8, 0.7, 0.8, 0.9]}

It would work if you use:

p.line(source=ds, x='timestamp', y='avg')

[Krastanov]

I see, I did not know that indices (as opposed to normal columns) can be named. Thanks.