如何从CSV绘制Bokeh多行数据框

Question

I have a data frame read in from a .csv with the following format:

version, 2x8x8, 2x8x10, 2x8x12, ...
v1.0.0,  10.2,  9.2,    8.2,
v1.0.1,  11.3,  10.4,   10.2,
v1.0.2,  9.5,   9.3,    9.1,  
...

I want to plot this data frame as a multiline plot in bokeh where each column is its own line. The x axis being version num, and the y values being the content of the column excluding the header.

I have tried referencing the bokeh docs themselves but I cannot find the best way to extract the columns as "lists of lists" as bokeh expects.

# produces empty plot
f = figure(title='Example title')
versions = list(df['version'])
values = [list(df['2x8x8']), list(df['2x8x10']), ...]
f.multi_line(xs=versions, ys=values)

When I attempt to use the alternate ColumnDataSource approach, also specified in the bokeh docs , the plot takes all my y values and makes a new line for each.

# produces plot seen below
df = pd.read_csv(my.csv)
data_source = ColumnDataSource(df)
f = figure(title="Example")
f.line(x='version', y='2x8x8', source=data_source, line_width=2, legend='2x8x8')
f.line(x='version', y='2x8x10', source=data_source, line_width=2, legend='2x8x10')
f.xaxis.axis_label = 'version'

Any help is greatly appreciated!

Answer 1

I think this is what you want (tested on Bokeh v1.0.4):

import pandas as pd
import numpy as np
from bokeh.palettes import Spectral11
from bokeh.plotting import figure, show

toy_df = pd.DataFrame(data = {'version': ['v1.0.0', 'v1.0.1', 'v1.0.2', 'v1.0.3'],
                              '2x8x8': [10.2, 11.3, 9.5, 10.9],
                              '2x8x10': [9.2, 10.4, 9.3, 9.9],
                              '2x8x12': [8.2, 10.2, 9.1, 11.1]}, columns = ('version', '2x8x8' , '2x8x10', '2x8x12'))

numlines = len(toy_df.columns)
mypalette = Spectral11[0:numlines]

p = figure(width = 500, height = 300, x_range = toy_df['version'])
p.multi_line(xs = [toy_df['version'].values] * numlines,
             ys = [toy_df[name].values for name in toy_df],
             line_color = mypalette,
             line_width = 5)
show(p)

Result:

Answer 2

Another version including labels. This is a different approach using explicitly ColumnDataSource instead of pandas DataFrame .

Please note that if you want to use p.legend.click_policy = "hide" to toggle visibility or mute separate lines then you should rather use line glyph instead of multi_line . This code works for Bokeh v1.0.4

from bokeh.palettes import Spectral11
from bokeh.plotting import figure, show
from bokeh.models import Legend, ColumnDataSource

versions = ['v1.0.0', 'v1.0.1', 'v1.0.2', 'v1.0.3']
data = {'version': [versions] * 3,
        'values': [[10.2, 11.3, 9.5, 10.9],
                   [9.2, 10.4, 9.3, 9.9],
                   [8.2, 10.2, 9.1, 11.1]],
        'columns': ['2x8x8', '2x8x10', '2x8x12'],
        'color': Spectral11[0:3] }

source = ColumnDataSource(data)

p = figure(width = 500, height = 300, x_range = versions)
p.multi_line(xs = 'version',
             ys = 'values',
             color = 'color',
             legend = 'columns',
             line_width = 5,
             source = source)
show(p)

Result: