CODEWITHSUNDEEP
pythonmatplotlib
OverLordGoldDragon
OverLordGoldDragon

Reputation: 19776

How to exclude long peaks in histograms?

Sometimes I get histograms that look like below:

enter image description here

I see the peaks loud and clear, but nigh for much else; is there a way to drop the "bin outliers" from a histogram so that the rest of the distribution can be seen better?

Upvotes: 0

Views: 330

Answers (1)

OverLordGoldDragon
OverLordGoldDragon

Reputation: 19776

This can be accomplished by simply setting ylim; however, this rids of the peaks information. To retain, we can include it via annotations, as follows:

  1. Fetching histogram heights, N, and positions, bins
  2. Selecting a ymax; e.g. 2nd or 3rd max N
  3. Packing (position, height) into a string, and annotating

All combined and an example below; I used your exact data for comparison, since you are me.

import numpy as np
import matplotlib.pyplot as plt
np.random.seed(0)

data = np.random.randn(100, 100) ** 3

data[:50] = 0
hist_visible(data, peaks_to_clip=3, bins=500, annot_kw={})
data[:95] = 0
hist_visible(data, peaks_to_clip=3, bins=500, annot_kw={})

enter image description here

Function:

def hist_visible(data, peaks_to_clip=1, bins=200, annot_kw=None):
    def _annotate(peaks_info, annot_kw):
        def _process_annot_kw(annot_kw):
            defaults = dict(weight='bold', fontsize=13, color='r',
                            xy=(.85, .85), xycoords='axes fraction')
            if not annot_kw:
                annot_kw = defaults.copy()
            else:
                annot_kw = annot_kw.copy()  # ensure external dict unaffected
                # if `defaults` key not in `annot_kw`, add it & its value
                for k, v in defaults.items():
                    if k not in annot_kw:
                        annot_kw[k] = v
            return annot_kw

        def _make_annotation(peaks_info):
            txt = ''
            for entry in peaks_info:
                txt += "({:.2f}, {})\n".format(entry[0], int(entry[1]))
            return txt.rstrip('\n')

        annot_kw = _process_annot_kw(annot_kw)
        txt = _make_annotation(peaks_info)
        plt.annotate(txt, **annot_kw)

    N, bins, _ = plt.hist(np.asarray(data).ravel(), bins=bins)

    Ns = np.sort(N)
    lower_max = Ns[-(peaks_to_clip + 1)]

    peaks_info = []
    for peak_idx in range(1, peaks_to_clip + 1):
        patch_idx = np.where(N == Ns[-peak_idx])[0][0]
        peaks_info.append([bins[patch_idx], N[patch_idx]])
    plt.ylim(0, lower_max)

    if annot_kw is not None:
        _annotate(peaks_info, annot_kw)
    plt.show()

Upvotes: 1

Related Questions