Normalizing to bin height with matplotlib

Question

I have a set of histograms, each one using a single column of a pandas dataframe and the matplotlib.pyplot.hist function. However, each set of data is a different length, so I want to normalize each histogram; using the built in density option does not make sense for my data, so I want to divide each bin height by the maximum bin height.

Overall I want to know how to 1- extract the bin heights from the histogram made by plt.hist 2- divide all the bin heights by the maximum (got confused by datatypes here, I think Im trying to divide two tuples?) 3- plot a new histogram with the normalized bin heights.

Ideally I want to do this in an order where I can tweak my choice of bin number in the original plot and then re-run to update both the original and normalized plot.

I tried naming what the plt.hist function returns and then dividing by the max, but the only version of this that did not throw an error gave me a plot that made no sense (I think I divided the values Im binning instead of the bin heights, I also don't really understand what n, bins, and patches are)

(n, bins, patches) = plt.hist(df['values'], bins=50)

plt.hist(df['values']/max(n), bins 50) 

Answer 1

plt.hist() has 3 return values:

• the counts of each bar (this is what is shown by default)
• the edges between the bars (there are 51 edges for 50 bars)
• the graphical elements that form the bars (rectangular patches)

To use the return values again, you need to create a bar plot, not a histogram. A new histogram would bin the 50 counts again into new counts.

import matplotlib.pyplot as plt
import numpy as np

plt.figure()
values = np.random.randn(10000).cumsum()
counts, bin_edges, _bars = plt.hist(values, bins=50)
plt.xlabel('Values')
plt.ylabel('Counts')
plt.show()

plt.figure()
plt.bar(bin_edges[:-1], counts / counts.max(), width=np.diff(bin_edges), align='edge')
plt.xlabel('Values')
plt.ylabel('Percentage vs highest bar')
plt.show()

Drawing the original histogram can be skipped by calling np.histogram() instead. It has the same return values, except for the graphical elements. Here is how a standalone code example could look like, with the y-axis formatted as percentages:

import matplotlib.pyplot as plt
from matplotlib.ticker import PercentFormatter
import numpy as np

values = np.random.randn(10000).cumsum()
counts, bin_edges = np.histogram(values, bins=50)
plt.bar(bin_edges[:-1], counts / counts.max(), width=np.diff(bin_edges), align='edge')
plt.xlabel('Values')
plt.ylabel('Percentage vs highest bar')
plt.gca().yaxis.set_major_formatter(PercentFormatter(1))
plt.show()