2

I am plotting a histogram with given 'ages':

import matplotlib.pyplot as plt
import numpy as np

ages=[1, 1, 1, 2, 2, 2, 2, 2, 2, 30, 30, 30, 150, 152]

plt.hist(ages, edgecolor='black')

plt.title("Scores vs. Frequency")
plt.xlabel("Score")
plt.ylabel("Frequency")

plt.tight_layout()
plt.show()

I am trying to draw a histogram that looks like this, but I can't get right using this code. enter image description here

Where is the problem? I mean is it even possible?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
cpx
  • 263
  • 3
  • 9

1 Answers1

1

I don't think that a "histogram" such as the one you show in your question is appropriate because the horizontal axis doesn't display ages on a true numerical scale. As it stands it is more like a bar chart with arbitrary labels.

Maybe you want to use $\log_{10}(Age)$ on the horizontal scale. In R, you can do this as shown below. [Argument br suggests the approximate number of bins.] Maybe Python has a way to label the horizontal scale with values for Age instead of log(Age).

ages=c(1, 1, 1, 2, 2, 2, 2, 2, 2, 30, 30, 30, 150, 152)
hdr="Histogram of 14 Ages (1 through 152) on Log Scale"
hist(log10(ages), br=8, col="skyblue2", main=hdr)

enter image description here

Note: If you google something like histogram python log scale you will get a lot of examples for plotting histograms of logged data with various kinds of software, including Python (some with logged values on the vertical scale, which you can ignore).

Addendum per Comment:

Here are three datasets (each of size $n=100)$ of types that might arise in a real statistical application. The histograms at left are of the actual data; some of them might be useful without transforming the data. The histograms at right are for log10 transformations of the same three datasets. Some of them might be more useful.

Distributions are lognormal data (sometimes used to model financial data and earthquake magnitudes), gamma data (waiting times), and Weibull data (reliability).

enter image description here

R code for six histograms above.

par(mfrow=c(3, 2))
set.seed(1776)
w = rlnorm(100, 100, 20)
 hist(w, prob=T, col="wheat")
  hist(log10(w), prob=T, col="skyblue2")
x = rgamma(100, 2, .01)
 hist(x, prob=T, col="wheat")
  hist(log10(x), prob=T, col="skyblue2")
y = rweibull(100, 1, 100)
 hist(y, prob=T, col="wheat")
  hist(log10(y), prob=T, col="skyblue2")
par(mfrow=c(1,1))
BruceET
  • 47,896
  • 2
  • 28
  • 76
  • Thanks. I tried this with another example `[19, 19, 19, 20, 20, 20, 20, 20, 20, 21, 21, 21, 145, 147]` it seems here the `log10` method only gives 2 bars. Any reason why it doesn't seem to work well? – cpx Nov 13 '20 at 12:57
  • Your sample has too many relatively small values along with a couple of large ones. It is not difficult to _contrive_ a problematic sample that makes an ugly histogram. // In practice one wants to make histograms for data sampled from naturally occurring populations. // See Addendum to my answer for some examples. – BruceET Nov 13 '20 at 18:05