[BUG?] Legend goes crazy if plotting more than 24 Vertical Lines?

I’m pasting here full code for reproducibility. Please jump to bottom section Problem Description and Questions first, if you wish.

# Versions

bokeh==0.12.13 holoviews==1.9.2 pandas==0.22.0

Imports

import numpy as np
import pandas as pd
import datetime
import random
import holoviews as hv
hv.extension('bokeh')

Generation and plotting functions

#generate dummy df
def generate_df(size):
    d = {
        'dates':
        pd.date_range('1980-01-01', periods=size, freq='1T'),
        'signal1':
        np.random.normal(0, 0.3, size=size).cumsum() + 50,
        'signal2':
        np.random.normal(0, 0.3, size=size).cumsum() + 50,
        'signal3':
        np.random.normal(0, 0.3, size=size).cumsum() + 50
    }

    df = pd.DataFrame(d)

    return df.melt(id_vars='dates').sort_values('dates')

#generate dummy dates
def generate_random_date(begin, end):
    
    diff = end - begin
    
    
    return begin + datetime.timedelta(seconds=random.randint(0, int(diff.total_seconds())))

def generate_random_dates(begin, end, n):
    
    return [generate_random_date(begin, end) for i in range(0,n)]

#plot a timeseries
def plot_timeseries(df, width=800, height=400, title='plot', legend_position='top_right'):

        plot_opts=dict()
        plot_opts['show_grid'] = True
        plot_opts['width'] = width
        plot_opts['height'] = height
        plot_opts['legend_position'] = legend_position
        plot_opts['legend_limit'] = 1000


        curve_1 = hv.Curve( df,
                     kdims=['dates'],
                     vdims=['value', 'variable'],
                     label=title).groupby('variable')


        return curve_1.overlay().opts(plot=plot_opts)

#Overlay a list of vertical lines
def vertical_lines(vlines_list):

        plot_opts=dict()
        plot_opts['show_legend'] = False
        plot_opts['show_title'] = False
        
        style_opts = dict()
        style_opts['line_color'] = 'brown'
        style_opts['linewidth'] = '4'
        
        
        vlines = []
        for line in vlines_list:
            vlines.append(hv.VLine(line)(plot=plot_opts,
                                         style=style_opts))

        return hv.Overlay(vlines)

Generate df and dates and plot them

small_df = generate_df(10000)
small_df.head()

<div>

	dates	variable	value
0	1980-01-01 00:00:00	signal1	49.800754
20000	1980-01-01 00:00:00	signal3	50.292620
10000	1980-01-01 00:00:00	signal2	50.425544
1	1980-01-01 00:01:00	signal1	49.896840
20001	1980-01-01 00:01:00	signal3	50.497810

</div>

With 24 lines

vlines_list = generate_random_dates(small_df.dates.min(), small_df.dates.max(), 24)


vertical_lines(vlines_list) * plot_timeseries(small_df, title='24 lines')

With 25 lines

vlines_list = generate_random_dates(small_df.dates.min(), small_df.dates.max(), 25)


vertical_lines(vlines_list) * plot_timeseries(small_df, title='25 lines')

Problem Description and Questions

I am overlaying a Curves Overlay (function plot_timeseries) with many vertical lines (function vertical_lines).

If I generate and plot 24 vertical lines the legend appears correctly (see legend of figure section With 24 lines). If I generate and plot 25 vertical lines the legend appears incorrectly (see legend of figure section With 25 lines).

Questions:

1. Is this a bug?
2. Or am I doing something wrong? For instance, is there a more efficient way of plotting several vertical lines besides calling VLine repeatedly? Such as, sending a list as argument.

[philippjfr]

Legends are limited to 25 entries by default because things often getting messy otherwise. Since VLine does not actually have a legend entry it shouldn’t count toward this limit but seemingly it does, so I’ll leave this open. For now you raise the limit with:

overlay.opts(plot=dict(legend_limit=50))

or do so globally:

hv.opts({'Overlay': {'plot': dict(legend_limit=50)}})

[philippjfr]

Probably not, deserves a little note in the VLine/HLine reference notebooks for bokeh (and maybe matplotlib too) as I’m not sure axvline/axhline participate in legends there either.