1

我正在尝试将数据阴影图放在我正在处理的散景服务器仪表板中。它主要工作，除了当我尝试放大绘图的 y 轴时，它会“重置”。我设法制作了重现我遇到的问题的脚本。我的相关环境是：

bokeh = 0.12.11
datashader = 0.6.3
holoviews = 1.9.1-x-gc1d66c015

这是我使用的代码bokeh serve test.py。我也在远程机器上运行它，并使用散景服务器文档中描述的 SSH 隧道。几乎相同的代码在 jupyter notebook 中运行良好，在同一台机器上运行。所以我怀疑这是散景服务器中缺少的问题或配置。

from bokeh.io import curdoc
import holoviews as hv
import os
import pandas as pd
import numpy as np
import datashader as ds
from bokeh.models import DatetimeTickFormatter
from bokeh.palettes import Category20
from holoviews.operation.datashader import datashade, dynspread
import holoviews.plotting.bokeh  # noqa (Activate backend)

def read_data():
    t = pd.date_range(start='2017-01-01', freq='1T', periods=1e6)
    x = np.arange(t.size)
    y1 = np.cos(x / 100)
    y2 = 10 + np.sin(x / 100)

    classes = list(range(1, 15))
    class_list = []
    for c in classes:
        class_list += [c] * 100
        cluster = class_list * (t.size // 100 + 1)
    df = pd.DataFrame(data=dict(cluster=cluster[:t.size],
                                T1C1=y1, T1C2=y2),
                      index=t)
    df.index.name = 'time'
    return df

df = read_data()

# Convert index to integer so datashader will work correctly.
df.index = df.index.values.astype(np.int64) / 1e6
class_names = [str(c) for c in sorted(df['cluster'].unique()) if int(c) >= 0]

def apply_formatter(plot, element):
         plot.handles['xaxis'].formatter = DatetimeTickFormatter()

def get_points_data(feature):
    d = {}
    df.index.name = 'time'
    for cluster_num in class_names:
        subset = df.loc[df['cluster'] == int(cluster_num), feature].dropna()
        d[cluster_num] = hv.Points(data=subset.reset_index(), kdims=['time', feature])
    return hv.NdOverlay(d, kdims='k')

spec = {'plot': {'RGB': {'width': 800, 'height': 300,
                              'finalize_hooks': [apply_formatter],
                              # 'tools': 'pan,box_zoom,wheel_zoom,save,hover'
                              }
                      },
             'norm': {'RGB': {'framewise': True}}}

n_clusters = df['cluster'].nunique()
if n_clusters <= 20:
    colors = Category20[n_clusters]
else:
    num_repeats = n_clusters // 20 + 1
    colors = Category20[20] * num_repeats
    colors = colors[:n_clusters]

renderer = hv.Store.renderers['bokeh'].instance(mode='server', holomap='server')
DataSelector = hv.streams.Stream.define('feature', feature='T1C1')
data_stream = DataSelector(feature='T1C1')
dmap_points = hv.DynamicMap(get_points_data, streams=[data_stream])
rangexy = hv.streams.RangeXY(transient=True)
plotsize = hv.streams.PlotSize()
datashaded_points = dynspread(datashade(dmap_points, color_key=colors,
                                             streams=[rangexy, plotsize],
                                             aggregator=ds.count_cat('k'),
                                             min_alpha=0.1
                                             ), threshold=0.5, max_px=10).opts(spec)


renderer = hv.Store.renderers['bokeh'].instance(mode='server', holomap='server')
doc = curdoc()
plot = renderer.get_plot(datashaded_points, doc=doc)

doc.add_root(plot.state)
4

0 回答 0