复杂散景图中CheckboxButtonGroup与图例之间的交互

Question

我有一个复杂的多元数据集，其结构与此类似：

import pandas as pd
import numpy as np
import datetime as dt
from itertools import cycle, islice

N = 24
start_date = dt.date(2016,1,1)
nbdays = int(365 / N)

df = pd.DataFrame({'Date': [start_date + dt.timedelta(days=i*nbdays) for i in range(1,N+1)], 
                   'Rating':    [(100/N)*i for i in range(1,N+1)], 
                   'Plot':      list(islice(cycle(range(1, 9)), 0, N)), 
                   'Treatment': list(islice(cycle(range(1, 7)), 0, N)), 
                   'Trial':     list(islice(cycle(range(1, 4)), 0, N)), 
                   'Name':      list(islice(cycle("ABCDEF"), 0, N)), 
                   'Target':    list(islice(cycle("JKLMNOP"), 0, N)), 
                   'Part':      list(islice(cycle("WXYZ"), 0, N)) 
                   })

我想要：

• 绘制Date与Rating ，按Treatment着色
• 具有交互式图例，因此单击治疗可切换治疗的可见性
• 在图的侧面具有用于其他参数（ Plot ， Trial ， Name ， Target ， Part ）的按钮，因此单击按钮可切换相应点的可见性
• 悬停在一个点上时显示所有参数

这是我的代码（变量df数据集）：

from bokeh.plotting import figure
from bokeh.models import ColumnDataSource
from bokeh.palettes import Set1 
from bokeh.models import (CDSView, BooleanFilter, Legend,
                          DatetimeTickFormatter, Range1d,
                          HoverTool)
from bokeh.models.widgets import CheckboxButtonGroup, Div
from bokeh.layouts import widgetbox, layout
from bokeh.io import curdoc

columns = ['Treatment', 'Plot', 'Trial', 'Name', 'Target', 'Part']
categories = [sorted(df[column].unique()) for column in columns]
all_columns = ['Date', 'Rating'] + columns

treatment_colormap = dict(zip(categories[0], Set1[6])) 

# Create Input controls
divs = [Div(text=column+':') for column in columns[1:]]
controls   = [CheckboxButtonGroup(labels=list(map(str, category)), active=list(range(len(category)))) for category in categories[1:]]

# Create Column Data Source that will be used by the plot
source = ColumnDataSource(data=dict((column, []) for column in all_columns))


def select():
    actives = [control.active for control in controls]
    actives_names = [[category[a] for a in active] for (active, category) in zip(actives, categories[1:])]

    presence = [df[column].isin(active_names) for (column, active_names) in zip(columns[1:], actives_names)]
    result = df[np.logical_and.reduce(presence)] # https://stackoverflow.com/a/49027984/50065
    return result

def update():
    sdf = select()
    source.data = dict((column, sdf[column]) for column in all_columns)

    glyphs = []
    selected_treatments = sorted(sdf['Treatment'].unique())
    for treatment in selected_treatments:
        booleans = [value == treatment for value in source.data['Treatment']]
        view = CDSView(source=source, filters=[BooleanFilter(booleans)])
        color = treatment_colormap[treatment]
        glyphs.append(p.circle(x='Date', y='Rating', source=source, view=view, line_color=color, fill_color=color))

    legend = Legend(items=[
        ("treatment {}".format(treatment), [glyph]) for treatment, glyph
        in zip(selected_treatments, glyphs)
        ])

    p.add_layout(legend, 'below')
    p.legend.click_policy='hide'
    p.legend.location = 'bottom_center'
    p.legend.orientation = 'horizontal'

for control in controls:
    control.on_change('active', lambda attr, old, new: update())

def datetime_in_miliseconds(date):
    date = dt.datetime.strptime(date, '%d/%m/%Y')
    epoch = dt.datetime.utcfromtimestamp(0)
    return (date - epoch).total_seconds() * 1000.0

hover = HoverTool(tooltips=[('Date', '@Date{%d/%m/%Y}')] + [(column, '@'+column) 
                            for column in all_columns[1:]], formatters={
                             'Date': 'datetime', # use 'datetime' formatter for 'Date' field
                             })


p = figure(x_axis_type="datetime", tools=[hover])
p.title.text = 'Date vs Rating'
p.xaxis.axis_label = 'Date'
p.xaxis.formatter = DatetimeTickFormatter(days = ['%d/%m/%y'])
start = datetime_in_miliseconds('01/01/2016')
end = datetime_in_miliseconds('31/12/2016')
p.x_range=Range1d(start, end)
p.yaxis.axis_label = 'Rating'
p.ygrid.band_fill_color="olive"
p.ygrid.band_fill_alpha = 0.1
p.y_range=Range1d(0,100)

sizing_mode = 'scale_width'
inputs = widgetbox(*sum(zip(divs, controls), tuple()), sizing_mode=sizing_mode)

l = layout([[p, inputs]], sizing_mode=sizing_mode)
update()  # initial load of the data
curdoc().add_root(l)

当您运行bokeh serve --show main.py （ bokeh版本0.12.10）时，它看起来像这样：

什么有效：

• 单击图例可切换治疗的可见性

什么不起作用：

• 悬浮标签中显示的信息不正确（前六个点在其悬浮标签中具有相同的信息，接下来的六个点也具有相同的悬浮标签，依此类推）。
• 单击右侧按钮弄乱了绘图：轴标签消失了，在绘图上方而不是下方显示了第二个图例）

我该如何解决最后两点？

Answer 1

这里有一些想法：

• 使用CustomJSFilter封装一个javascript函数来进行数据过滤。
• 仅使用一次p.circle()调用即可绘制所有圆圈。
• 使用factor_cmap将处理列映射到颜色。
• 使用tags属性在Python中保存数据并在javascript中读取数据。

由于只有一个GlyphRenderer ，因此可见性切换不适用于其图例。

要解决此问题，请创建一个虚拟的ColumnDataSource并p.circle()多次调用p.circle()以创建一个虚拟GlyphRenderer列表。 为这些虚拟GlyphRenderer创建图例，并将其可见属性更改与调用CustomJS source.change.emit()的CustomJS链接以重绘该图。

由于所有过滤器计算均由javascript执行，因此您可以创建一个可以与用户输入进行交互的静态html文件。

这是笔记本：

http://nbviewer.jupyter.org/gist/ruoyu0088/01ddf28ed041508304843f49a794d66a

from bokeh.plotting import figure
from bokeh.models import ColumnDataSource, CustomJS, CDSView, CustomJSFilter, HoverTool
from bokeh.models.widgets import CheckboxButtonGroup
from bokeh.io import show, output_notebook
from bokeh.palettes import Set1 
from bokeh.transform import factor_cmap
from bokeh.layouts import widgetbox, layout

import pandas as pd
import numpy as np
import datetime as dt
from itertools import cycle, islice

output_notebook()

N = 24
start_date = dt.date(2016,1,1)
nbdays = int(365 / N)

df = pd.DataFrame({'Date': [start_date + dt.timedelta(days=i*nbdays) for i in range(1,N+1)], 
                   'Rating':    [(100/N)*i for i in range(1,N+1)], 
                   'Plot':      list(islice(cycle(range(1, 9)), 0, N)), 
                   'Treatment': list(islice(cycle(range(1, 7)), 0, N)), 
                   'Trial':     list(islice(cycle(range(1, 4)), 0, N)), 
                   'Name':      list(islice(cycle("ABCDEF"), 0, N)), 
                   'Target':    list(islice(cycle("JKLMNOP"), 0, N)), 
                   'Part':      list(islice(cycle("WXYZ"), 0, N)) 
                   })

columns = 'Plot', 'Trial', 'Name', 'Target', 'Part'
unique_items = [df[col].unique() for col in columns]

df["Treatment"] = df["Treatment"].astype(str)

source = ColumnDataSource(data=df)
dummy_source = ColumnDataSource(data={"x":[], "y":[]})

hover = HoverTool(tooltips=[('Date', '@Date{%d/%m/%Y}')] + [(column, '@'+column) 
                            for column in columns], formatters={
                             'Date': 'datetime', # use 'datetime' formatter for 'Date' field
                             })

p = figure(x_axis_type="datetime", tools=[hover])
color = factor_cmap("Treatment", Set1[9], df.Treatment.unique())

for i, label in enumerate(df.Treatment.unique()):
    dummy_circle = p.circle(x="x", y="y", source=dummy_source, legend="Treatment {}".format(label), color=Set1[9][i])
    dummy_circle.tags = [label]

p.legend.location = "bottom_right"
p.legend.click_policy = "hide"

def source_change(source=source):
    source.change.emit()

callback_source_change = CustomJS.from_py_func(source_change)

for item in p.legend[0].items:
    item.renderers[0].js_on_change("visible", callback_source_change)

controls = [CheckboxButtonGroup(labels=items.astype(str).tolist(), active=list(range(len(items)))) for items in unique_items]
widgets = widgetbox(*controls)

for name, control in zip(columns, controls):
    control.tags = [name]

def func_filter(source=source, legend=p.legend[0], widgets=widgets):
    window.widgets = widgets
    visible_treatments = [item.renderers[0].tags for item in legend.items if item.renderers[0].visible]
    date = source.data['Date']
    treatments = source.data['Treatment']
    res = []

    selectors = {}
    for widget in widgets.children:
        col = widget.tags[0]
        selectors[col] = dict([(widget.labels[i], i) for i in widget.active])

    for i in range(len(date)):
        flag = treatments[i] in visible_treatments
        for key, val in selectors.items():
            if source.data[key][i] not in val:
                flag = False
                break
        res.append(flag)
    return res

view = CDSView(source=source, filters=[CustomJSFilter.from_py_func(func_filter)])
p.circle(x='Date', y='Rating', source=source, view=view, line_color=color, fill_color=color)  

for control in controls:
    control.js_on_change("active", callback_source_change)

show(layout([[p, widgets]]))