Plotly: How to plot a cumulative “steps” histogram

问题:

I am trying to plot a cumulative histogram using Plotly in python, but make it look like "steps", i.e. bars with no color and only the top line is displayed. Something like this:

Basically, I'm trying to reproduce the behavior of the following matplotlib code:

import matplotlib.pyplot as plt
plt.hist(x, cumulative=True, histtype='step')

So far, the best I've been able to do is:

import plotly.graph_objs as go
from plotly.offline import iplot
h = go.Histogram(x=x,
                         cumulative=dict(enabled=True),
                         marker=dict(color="rgba(0,0,0,0)",
                                     line=dict(color="red", width=1)))
iplot([h])

Which results in something like:

So what's the trick?

回答1:

If you're willing to handle the binning and accumulation before you plot the data, you can use a go.Scatter object with the shape property of the line set to 'hvh'.

Plot:

Code: Setup for a Jupyter Notebook

#imports
import plotly.plotly as py
import plotly.graph_objs as go
from plotly.offline import download_plotlyjs, init_notebook_mode, plot, iplot

import numpy as np
import pandas as pd

# qtconsole for debugging
#%qtconsole -- style vim

# Notebook settings
init_notebook_mode(connected=True)

# Some sample data
x = np.random.normal(50, 5, 500)
binned = np.histogram(x, bins=25, density=True)
plot_y = np.cumsum(binned[0])

# Line
trace1 = go.Scatter(
    x=binned[1],
    y=plot_y,
    mode='lines',
    name="X",
    hoverinfo='all',
    line=dict(color = 'rgb(1255, 0, 0)', shape='hvh'
    )
)

data = [trace1]

# Layout
layout = dict(title = 'Binned data from normal distribution',
    legend=dict(
        y=0.5,
        traceorder='reversed',
        font=dict(
            size=16
        )
    )
)

# Make figure
fig = dict(data=data, layout=layout)

# Plot
iplot(fig, filename='line-shapes')

I hope this is something you can use!

Don't hesitate to let me know if not.

Some details:

The data sample is made using np.random.normal(). x is a sampled normal distribution with mean = 50, sigma = 5 and 500 observations. x is then put in 50 bins using np.histogram() which returns two arrays. These are used as data source for the plot.

Possible alternative approaches:

I also tried using your snippet with some random sample data and include shape='hvh' in your line=dict(color="red", width=1). That did not seem to work though. I also considered modifying the layout of your go.Histogram() so that only the top line of the bars were plotted, but I don't think it's possible.

回答2:

The accepted solution works but may be limiting as the bins are all of equal width. One approach is to use matplotlib to calculate stats, then plot with plotly:

# sample data
# I am not using a normal distribution on purpose so that the effect of varying bin widths is apparent.
x = np.random.rand(100)

# use matplotlib to get "n" and "bins"
# n_bins will affect the resolution of the cumilative histogram but not dictate the bin widths.
n_bins = 100
n, bins, patches = plt.hist(x, n_bins, density=True, histtype='step', cumulative=-1)

# use plotly (v3) to plot
data = []
trace = go.Scatter(
    x=bins,
    y=n,
    mode='lines',
    name= "test",
    line=dict(
        shape='hvh'
    )
)

data.append(trace)
fig = go.Figure(data=data)
iplot(fig)

The result should look something like this: