How to reverse the legends of stacked barplot in pandas

Question

I have a dataset with a few records about some crop production by year. So I am visualizing the top produced crop by each year in a stacked bar chart. Dataset I have used can be found in kaggle PMFBY Coverage.csv.

Here is my code.

# Top Crop by year
plt.figure(figsize=(12, 6))

df_crg_[df_crg_.year==2018].groupby('cropName').size().nlargest(5).plot(kind='barh', color='red', label='2018')
df_crg_[df_crg_.year==2019].groupby('cropName').size().nlargest(5).plot(kind='barh', color='green', label='2019')
df_crg_[df_crg_.year==2020].groupby('cropName').size().nlargest(5).plot(kind='barh', color='blue', label='2020')
df_crg_[df_crg_.year==2021].groupby('cropName').size().nlargest(5).plot(kind='barh', color='maroon', label='2021')

plt.legend(loc="upper right")
plt.xlabel('Total Production Time')
plt.title('Top Crop by year')
plt.show()

And this was the output

Now if you look at the graph you would notice the stacked bar chart legends are revered, it is showing 2021 status first instead of 2018. So I want to reverse this order of representation.

I found one solution for this question but I don't know how to apply it, as it is assigning plotting commands to one variable but in my case, there are four plotting commands.

---

Only this answer would do, but if know and can answer any other method of extracting top produced crop by year then that would be great. If you notice here I am manually going through each year then extracting that year's top crop. I tried doing it with groupby but I wasn't able to get the answer.

Thanks

Answer 1

First off, the same 5 crops need to be selected each year. Otherwise, you can't have a fixed ordering on the y-axis.

The easiest way to get a plot with the overall 5 most-frequent crops, is seaborn's sns.countplot and limiting to the 5 largest. Note that seaborn is strongly objected to stacked bar plots, so you'll get "dodged" bars (which are easier to compare, year by year, and crop by crop):

import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd
import numpy as np

df = pd.read_csv('PMFBY coverage.csv')

sns.set_style('white')
order = df.groupby('cropName').size().sort_values(ascending=False)[:5].index
plt.figure(figsize=(12, 5))
ax = sns.countplot(data=df, y='cropName', order=order, hue='year')
for bars in ax.containers:
    ax.bar_label(bars, fmt='%.0f', label_type='edge', padding=2)
sns.despine()
plt.tight_layout()
plt.show()

With pandas, you can get stacked bars, but you need a bit more manipulation:

import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd
import numpy as np

df = pd.read_csv('PMFBY coverage.csv')
sns.set_style('white')
order = df.groupby('cropName').size().sort_values(ascending=False)[:5].index
df_5_largest = df[df['cropName'].isin(order)]
df_5_largest_year_count = df_5_largest.groupby(['cropName', 'year']).size().unstack('year').reindex(order)
ax = df_5_largest_year_count.plot.barh(stacked=True, figsize=(12, 5))
ax.invert_yaxis()
for bars in ax.containers:
    ax.bar_label(bars, fmt='%.0f', label_type='center', color='white', fontsize=16)
sns.despine()
plt.tight_layout()
plt.show()

Now, compare this with how the bars would look like if you'd consider the 5 largest crops of each individual year. Notice how the crops and their order is different each year. How would you combine such information to a single plot?

sns.set_style('white')
fig, axs = plt.subplots(2, 2, figsize=(14, 8))

df[df.year == 2018].groupby('cropName').size().nlargest(5).plot(kind='barh', color='C0', title='2018', ax=axs[0, 0])
df[df.year == 2019].groupby('cropName').size().nlargest(5).plot(kind='barh', color='C1', title='2019', ax=axs[0, 1])
df[df.year == 2020].groupby('cropName').size().nlargest(5).plot(kind='barh', color='C2', title='2020', ax=axs[1, 0])
df[df.year == 2021].groupby('cropName').size().nlargest(5).plot(kind='barh', color='C3', title='2021', ax=axs[1, 1])
for ax in axs.flat:
    ax.bar_label(ax.containers[0], fmt='%.0f', label_type='edge', padding=2)
    ax.margins(x=0.1)
sns.despine()
plt.tight_layout()
plt.show()