Sort bar chart by list values in matplotlib

Question

I am encountering an issue regarding the sorting my features by their value. I would like to see my image with bars getting shorter based on high they are on the y axis. Unfortunately, my barplot looks like this, with the features being sorted alphabetically:

Right now I am running the following code:

unsorted_list = [(importance, feature) for feature, importance in 
                  zip(features, importances)]
sorted_list = sorted(unsorted_list)

features_sorted = []
importance_sorted = []

for i in sorted_list:
    features_sorted += [i[1]]
    importance_sorted += [i[0]]

plt.title("Feature importance", fontsize=15)
plt.xlabel("Importance", fontsize=13)

plt.barh(features_sorted,importance_sorted, color="green", edgecolor='green')

# plt.savefig('importance_barh.png', dpi=100)

Here is the data going through there:

unsorted_list =  
 [('HR', 0.28804817462980353),
 ('BR', 0.04062328177704225),
 ('Posture', 0.09011618483921582),
 ('Activity', 0.0017821837085763366),
 ('PeakAccel', 0.002649111136700579),
 ('HRV', 0.13598729040097057),
 ('ROGState', 0.014534726412631642),
 ('ROGTime', 0.22986192060475388),
 ('VerticalMin', 0.016099772399198357),
 ('VerticalPeak', 0.012697214182994502),
 ('LateralMin', 0.029479112475744584),
 ('LateralPeak', 0.022745210003295983),
 ('SagittalMin', 0.08653071485979484),
 ('SagittalPeak', 0.028845102569277088)]

sorted_list = 
[(0.0017821837085763366, 'Activity'),
 (0.002649111136700579, 'PeakAccel'),
 (0.012697214182994502, 'VerticalPeak'),
 (0.014534726412631642, 'ROGState'),
 (0.016099772399198357, 'VerticalMin'),
 (0.022745210003295983, 'LateralPeak'),
 (0.028845102569277088, 'SagittalPeak'),
 (0.029479112475744584, 'LateralMin'),
 (0.04062328177704225, 'BR'),
 (0.08653071485979484, 'SagittalMin'),
 (0.09011618483921582, 'Posture'),
 (0.13598729040097057, 'HRV'),
 (0.22986192060475388, 'ROGTime'),
 (0.28804817462980353, 'HR')]

I recently upgraded to matplotlib 3.0.2

Thanks in advance for helping me out!

Sheldore · Accepted Answer · 2019-01-29T16:42:32.710

6

EDIT (based on the comments)

Your code works fine on matplotlib 2.2.2 and the issue seems to be with your list naming convention and some confusion among them. It will work as expected on 3.0.2. Nevertheless, you might be interested in knowing the workaround

features_sorted = []
importance_sorted = []

for i in sorted_list:
    features_sorted += [i[1]]
    importance_sorted += [i[0]]

plt.title("Feature importance", fontsize=15)
plt.xlabel("Importance", fontsize=13)

plt.barh(range(len(importance_sorted)), importance_sorted, color="green", edgecolor='green')
plt.yticks(range(len(importance_sorted)), features_sorted);

Alternative suggested by @tmdavison

plt.barh(range(len(importance_sorted)), importance_sorted, color="green", 
     edgecolor='green', tick_label=features_sorted)

edited Jan 29 '19 at 16:42

answered Jan 29 '19 at 15:32

Sheldore

35,129
6
43
58

2

I think you can also use the `tick_label` option to `barh` instead of `plt.yticks` – tmdavison Jan 29 '19 at 15:35
@tmdavison: Thanks. I added your nice suggestion – Sheldore Jan 29 '19 at 15:41
What is the issue with matplotlib 3.0.2? – ImportanceOfBeingErnest Jan 29 '19 at 15:54
I don’t know but I believe the OP encountered some problem because the posted code works as such and producesthe desired plot in 2.2.2 but produces unsorted plot in 3.0.2. Can you try with 3.0.2 and reproduce the OP’s problem? Perhaps some bug? – Sheldore Jan 29 '19 at 15:57
No, there is no problem in any of the versions mentionned here. If anything, OP got mixed up with the various lists of the same name, or possibly they use a matplotlib version prior to 2.2. – ImportanceOfBeingErnest Jan 29 '19 at 16:01
Ok it seems 3.0.2 sorts the yaxis alphabetically in this case which is not the case for 2.2.2 – Sheldore Jan 29 '19 at 16:02
No, 3.0.2 does not sort the axis. – ImportanceOfBeingErnest Jan 29 '19 at 16:05
Thanks a lot for adressing my issue, it works great! This may actually be something for the devs of matplotlib to get rid of. – Willem van der Spek Jan 29 '19 at 16:06
I am a dev of matplotlib. What exactly is the problem? Your code runs fine. No need to use the workaround suggested in this answer. – ImportanceOfBeingErnest Jan 29 '19 at 16:09
Willem: make sure you are using 3.0.2 and using the correct lists. – Sheldore Jan 29 '19 at 16:18
@ImportanceOfBeingErnest I just rebooted my laptop and it seemed to have resolved the issue indeed. False alarm sorry – Willem van der Spek Jan 29 '19 at 16:20

score 0 · Answer 2 · answered Jan 29 '19 at 16:26

To avoid confusion from the other answer here, note that the code in the question runs fine and gives the desired output for any version of matplotlib >= 2.2.

import matplotlib
print(matplotlib.__version__)
import matplotlib.pyplot as plt


sorted_list = [(0.0017821837085763366, 'Activity'),
 (0.002649111136700579, 'PeakAccel'),
 (0.012697214182994502, 'VerticalPeak'),
 (0.014534726412631642, 'ROGState'),
 (0.016099772399198357, 'VerticalMin'),
 (0.022745210003295983, 'LateralPeak'),
 (0.028845102569277088, 'SagittalPeak'),
 (0.029479112475744584, 'LateralMin'),
 (0.04062328177704225, 'BR'),
 (0.08653071485979484, 'SagittalMin'),
 (0.09011618483921582, 'Posture'),
 (0.13598729040097057, 'HRV'),
 (0.22986192060475388, 'ROGTime'),
 (0.28804817462980353, 'HR')]

features_sorted = []
importance_sorted = []

for i in sorted_list:
    features_sorted += [i[1]]
    importance_sorted += [i[0]]

plt.title("Feature importance", fontsize=15)
plt.xlabel("Importance", fontsize=13)

plt.barh(features_sorted, importance_sorted, color="green", edgecolor='green')
plt.show()

The issue OP reports about is most probably caused by naming distinct lists by the same name and not restarting the kernel in between or similar non-reproducible things.

Sort bar chart by list values in matplotlib

2 Answers2

Linked

Related