Python seaborn：无法使我的图表看起来像 excel 图表

Question

我有一个如下表，它存储在名为“数据”的 pandas 数据框中。

Column1	Device1	event_rate %	% event dist	% non-event dist	% total dist
0	Android	3.08	27.3	32.96	32.75
1	Chrome OS	4.05	0.47	0.42	0.43
2	Chromium OS	9.95	0.23	0.08	0.09
3	Linux	2.27	0.05	0.09	0.09
4	Mac OS	6.43	4.39	2.45	2.52
5	Others	2.64	7.41	10.48	10.36
6	Windows	5.7	15.89	10.08	10.3
7	iOS	3.76	44.26	43.44	43.47

我正在尝试创建一个所需的 seaborn/matplot 图表，如下所示，它是在 excel 中创建的。

这是我的python代码：

feature = 'Device1'
fig, ax1 = plt.subplots(figsize=(10,6))
color = 'tab:blue'
title = 'Event rate by ' + feature
ax1.set_title(title, fontsize=14)
ax1.set_xlabel(feature, fontsize=14)
ax2 = sns.barplot(x=feature, y='% non-event dist', data = data, color=color)
ax2 = sns.barplot(x=feature, y='% event dist', data = data, color='orange')
plt.xticks(rotation=45)
ax1.set_ylabel('% Dist', fontsize=14, color=color)
ax1.tick_params(axis='y')
ax2 = ax1.twinx()
color = 'tab:red'
ax2.set_ylabel('Event Rate %', fontsize=14, color=color)
ax2 = sns.lineplot(x=feature, y='event_rate %', data = data, sort=False, color=color)
ax2.tick_params(axis='y', color=color)
handles1, labels1 = ax1.get_legend_handles_labels()
handles2, labels2 = ax2.get_legend_handles_labels()
handles = handles1 + handles2
labels = labels1 + labels2
plt.legend(handles,labels)
plt.show()

这是我得到的

问题：

1. 图例未显示。
2. 条形图相互重叠。
3. 有没有办法显示数据标签？

如何使我的 seaborn 绘图看起来与我的 excel 绘图相似？谢谢。

Answer 1

加载和塑造数据框

• 绘制数据最重要的部分是正确塑造绘图 API 的 DataFrame。
• 我认为使用.stack 将 DataFrame 从宽格式转换为长格式更容易
• .iloc[:, :-1] 选择所有行，但将 '% total dist' 排除在外。

import pandas as pd
import seaborn as sns

# create dataframe
data = {'Device1': ['Android', 'Chrome OS', 'Chromium OS', 'Linux', 'Mac OS', 'Others', 'Windows', 'iOS'],
        'event_rate %': [3.08, 4.05, 9.95, 2.27, 6.43, 2.64, 5.7, 3.76],
        '% event dist': [27.3, 0.47, 0.23, 0.05, 4.39, 7.41, 15.89, 44.26],
        '% non-event dist': [32.96, 0.42, 0.08, 0.09, 2.45, 10.48, 10.08, 43.44],
        '% total dist': [32.75, 0.43, 0.09, 0.09, 2.52, 10.36, 10.3, 43.47]}

df = pd.DataFrame(data)

# display(df.head())
       Device1  event_rate %  % event dist  % non-event dist  % total dist
0      Android          3.08         27.30             32.96         32.75
1    Chrome OS          4.05          0.47              0.42          0.43
2  Chromium OS          9.95          0.23              0.08          0.09
3        Linux          2.27          0.05              0.09          0.09
4       Mac OS          6.43          4.39              2.45          2.52

# convert from a wide to long format
dfl = df.iloc[:, :-1].set_index('Device1').stack().reset_index(name='Values').rename({'level_1': 'Type'}, axis=1)

# select the desired data
dist = dfl[dfl.Type.str.contains('dist')]
rate = dfl[dfl.Type.str.contains('rate')]

# display(dist.head())
       Device1              Type  Values
1      Android      % event dist   27.30
2      Android  % non-event dist   32.96
4    Chrome OS      % event dist    0.47
5    Chrome OS  % non-event dist    0.42
7  Chromium OS      % event dist    0.23

# display(rate.head())
        Device1          Type  Values
0       Android  event_rate %    3.08
3     Chrome OS  event_rate %    4.05
6   Chromium OS  event_rate %    9.95
9         Linux  event_rate %    2.27
12       Mac OS  event_rate %    6.43

绘图和注释

• 我已将图例放在它们各自的轴旁边
  • 有关放置图例的其他选项，请参阅 How to put the legend out of the plot。
• 引用此 SO Question 来创建组合图例。
• 调整bbox_to_anchor=(0.8, -0.25) 中的值以移动图例。

# create the figure and primary axes
fig, ax = plt.subplots(figsize=(11, 7))

# plot and format the bars
sns.barplot(data=dist, x='Device1', y='Values', hue='Type')
ax.set_ylabel('% Dist')
ax.set_xticklabels(ax.get_xticklabels(), rotation=90)
l1 = ax.legend(bbox_to_anchor=(-0.24, 1), loc='upper left')

# create the secondary axes
ax2 = ax.twinx()

# plot and format the line
sns.lineplot(data=rate, x='Device1', y='Values', ax=ax2, color='grey', label='event rate %', marker='o')
ax2.set_ylabel('% Event Rate')
l2 = ax2.legend(bbox_to_anchor=(1.04, 1), loc='upper left')

# combined legend by extracting the components from legend l1 and l2
plt.legend(l1.get_patches() + l2.get_lines(), 
           [text.get_text() for text in l1.get_texts() + l2.get_texts()], 
           bbox_to_anchor=(0.8, -0.25), ncol=3)

# remove l1 from the plot
l1.remove()

# annotate the line
for _, x, _, y in rate.itertuples():
    ax2.text(x, y, y)

组合图例

单独的图例

• 如果您想要单独的图例，请删除 plt.legend(...) 和 l1.remove()

Answer 2

当使用sns.barplot() 时，您可以使用hue 参数来制作未堆叠的条形图，但您需要一个额外的变量来定义要取消堆叠的变量（我使用pd.melt() 来处理这个问题）。

图例可以使用fig.legend()。

要显示数据标签，请使用plt.annotate()

feature = 'Device1'
fig, ax1 = plt.subplots(figsize=(10,6))
color = 'tab:blue'
title = 'Event rate by ' + feature
ax1.set_title(title, fontsize=14)
ax1.set_xlabel(feature, fontsize=14)
# ax2 = sns.barplot(x=feature, y='% non-event dist', data = data, color=color)
# ax2 = sns.barplot(x=feature, y='% event dist', data = data, color='orange')
ax1 = sns.barplot(x=feature, y='value', hue='variable', data=data.melt(['Device1'], ['% event dist', '% non-event dist']))
plt.xticks(rotation=45)
ax1.set_ylabel('% Dist', fontsize=14, color=color)
ax1.tick_params(axis='y')
ax2 = ax1.twinx()
color = 'tab:red'
ax2.set_ylabel('Event Rate %', fontsize=14, color=color)
# ax2 = sns.lineplot(x=feature, y='event_rate %', data = data, sort=False, color=color)
ax2 = sns.lineplot(x=feature, y='event_rate %', data = data, sort=False, color=color, label='event_rate %')
ax2.tick_params(axis='y', color=color)
# handles1, labels1 = ax1.get_legend_handles_labels()
# handles2, labels2 = ax2.get_legend_handles_labels()
# handles = handles1[:-1] + handles2
# plt.legend(handles1, labels)
ax1.legend([])
ax2.legend([])
fig.legend(bbox_to_anchor=(0.75, -0.07), ncol=3)

# annotations
for x, y in enumerate(data['event_rate %']):
    plt.annotate(text=y, xy=(x-0.2,y+0.2))

## Use this for annotating if you want the annotations to go above and below the plot based on line direction
# prev_y = data['event_rate %'][0]
# for x, y in enumerate(data['event_rate %']):
#     plt.annotate(text=y, xy=(x-0.2,y+(0.2 if y >= prev_y else -0.2)))
#     prev_y = y

# set ylim to higher limit in order to accomodate the annotations    
ax2.set_ylim((0, 11))

plt.show()

如果您想要一个更好看的网格，请确保 yticks 中显示的值具有相同的长度。

ax1.set_ylim(0,50) 
# ax1.get_yticks() -> [ 0. 10. 20. 30. 40. 50.] # length = 6
ax2.set_ylim((0, 11)) 
# ax2.get_yticks() -> [ 0.  2.  4.  6.  8. 10. 12.] # length = 7
yticks = np.linspace(0,11,len(ax1.get_yticks()))
ax2.set_yticks(yticks) 
# ax2.get_yticks() -> [ 0.   2.2  4.4  6.6  8.8 11. ] # length = 6

或者关闭第二个轴网格，这会使网格看起来类似于excel图表。

ax2.grid(False)