python plotly:痕迹数不固定

python plotly: unfixed number of traces

我的代码从 .xlsx 文件中读取数据,并使用 plotly 绘制气泡图。 Bubble Diagram 当我知道需要绘制多少条迹线时,这项任务就很容易了。但是,由于行数是可变的,所以当迹数不固定时,我感到很困惑。


       1991  1992  1993  1994  1995  1996  1997
US       10    14    16    18    20    42    64
JAPAN   100    30    70    85    30    42    64
CN       50    22    30    65    70    66    60

这是我未完成的代码:

# Version 2 could read data from .xlsx file.
import plotly as py
import plotly.graph_objs as go
import openpyxl

wb = openpyxl.load_workbook(('grape output.xlsx'))     
sheet = wb['Sheet1']       
row_max = sheet.max_row
col_max = sheet.max_column
l=[]

for row_n in range(row_max-1):
    l.append([])
    for col_n in range(col_max-1):
        l[row_n].append(sheet.cell(row=row_n+2, column=col_n+2).value)

trace0 = go.Scatter(
    x=[1991, 1992, 1993, 1994, 1995, 1996, 1997],
    y=['US', 'US', 'US', 'US', 'US', 'US', 'US'],
    mode='markers+text',
    marker=dict(
        color='rgb(150,204,90)',
        size= l[0],
        showscale = False,
        ),
    text=list(map(str, l[0])),     
    textposition='middle center',   
)

trace1 = go.Scatter(
    x=[1991, 1992, 1993, 1994, 1995, 1996, 1997],
    y=['JAPAN', 'JAPAN', 'JAPAN', 'JAPAN', 'JAPAN', 'JAPAN', 'JAPAN'],
    mode='markers+text',

    marker=dict(
        color='rgb(255, 130, 71)',
        size=l[1],
        showscale=False,
    ),
    text=list(map(str,l[1])),
    textposition='middle center',
)

trace2 = go.Scatter(
    x=[1991, 1992, 1993, 1994, 1995, 1996, 1997],
    y=['CN', 'CN', 'CN', 'CN', 'CN', 'CN', 'CN'],
    mode='markers+text',

    marker=dict(
        color='rgb(255, 193, 37)',
        size=l[2],
        showscale=False,
    ),
    text=list(map(str,l[2])),
    textposition='middle center',
)

layout = go.Layout(plot_bgcolor='rgb(10, 10, 10)',  
                   paper_bgcolor='rgb(20, 55, 100)',  
                   font={               
                       'size': 15,
                       'family': 'sans-serif',
                       'color': 'rgb(255, 255, 255)'  
                   },
                   width=1000,
                   height=500,
                   xaxis=dict(title='Output of grapes per year in US, JAPAN and CN', ),  
                   showlegend=False,
                   margin=dict(l=100, r=100, t=100, b=100),
                   hovermode = False,       
                   )

data = [trace0, trace1, trace2]
fig = go.Figure(data=data, layout=layout)


py.offline.init_notebook_mode()
py.offline.plot(fig, filename='basic-scatter.html')

你能教我怎么画吗?谢谢

我应该指出,如果您将原始数据作为文本或可以更轻松地复制和粘贴的内容附加,您的代码将更具可重现性。但是,无论如何,我仍然可以回答您的问题并为您指明正确的方向。

你应该做的是使用循环,并从查看行 data = [trace0, trace1, trace2] 开始。正如您所注意到的,如果您有 100 个国家而不是 3 个国家,则此方法不会扩大规模。

相反,您可以使用列表推导将 data 创建为列表,并更新每个轨迹发生变化的部分。 trace0trace1trace2 除了国家/地区、值和颜色外没有太大区别。为了向您展示我的意思,我使用 DataFrame 重新创建了您的数据,然后创建了包含您的国家和颜色的单独列表。

# Version 2 could read data from .xlsx file.
import plotly as py
import plotly.graph_objs as go
import openpyxl

# wb = openpyxl.load_workbook(('grape output.xlsx'))     
# sheet = wb['Sheet1']       
# row_max = sheet.max_row
# col_max = sheet.max_column
# l=[]

# for row_n in range(row_max-1):
#     l.append([])
#     for col_n in range(col_max-1):
#         l[row_n].append(sheet.cell(row=row_n+2, column=col_n+2).value)

import pandas as pd

df = pd.DataFrame({1991:[10,100,50], 1992:[14,30,22], 1993:[16,70,30], 1994:[18,85,65], 1995:[20,30,70], 1996:[42,42,66], 1997:[64,64,60]})
df.index = ['US','JAPAN','CN']
colors = ['rgb(150,204,90)','rgb(255, 130, 71)','rgb(255, 193, 37)']

data = [go.Scatter(
    x=df.columns,
    y=[country]*len(df.columns),
    mode='markers+text',
    marker=dict(
        color=colors[num],
        size= df.loc[country],
        showscale = False,
        ),
    text=list(map(str, df.loc[country])),     
    textposition='middle center',   
    )
    for num, country in enumerate(df.index)
]

layout = go.Layout(plot_bgcolor='rgb(10, 10, 10)',  
                   paper_bgcolor='rgb(20, 55, 100)',  
                   font={               
                       'size': 15,
                       'family': 'sans-serif',
                       'color': 'rgb(255, 255, 255)'  
                   },
                   width=1000,
                   height=500,
                   xaxis=dict(title='Output of grapes per year in US, JAPAN and CN', ),  
                   showlegend=False,
                   margin=dict(l=100, r=100, t=100, b=100),
                   hovermode = False,       
                   )

# data = [trace0, trace1, trace2]
fig = go.Figure(data=data, layout=layout)
fig.show()

# py.offline.init_notebook_mode()
# py.offline.plot(fig, filename='basic-scatter.html')

如果我随后将一个测试国家/地区添加到值为 1991-1997 的 DataFrame,我不需要更改其余代码,气泡图将相应更新。

# I added a test country with data
df = pd.DataFrame({1991:[10,100,50,10], 1992:[14,30,22,20], 1993:[16,70,30,30], 1994:[18,85,65,40], 1995:[20,30,70,50], 1996:[42,42,66,60], 1997:[64,64,60,70]})
df.index = ['US','JAPAN','CN','TEST']
colors = ['rgb(150,204,90)','rgb(255, 130, 71)','rgb(255, 193, 37)','rgb(100, 100, 100)']

Derek O. 的回答很完美,但我认为有一种更灵活的方法可以使用 plotly.express 来做到这一点,特别是如果您不想定义颜色。

想法是正确转换数据。

数据

import pandas as pd
df = pd.DataFrame({1991:[10,100,50], 1992:[14,30,22], 1993:[16,70,30], 1994:[18,85,65], 1995:[20,30,70], 1996:[42,42,66], 1997:[64,64,60]})
df.index = ['US','JAPAN','CN']
df = df.T.unstack()\
      .reset_index()\
      .rename(columns={"level_0": "country",
                       "level_1": "year",
                       0: "n"})
print(df)
   country  year    n
0       US  1991   10
1       US  1992   14
2       US  1993   16
3       US  1994   18
4       US  1995   20
5       US  1996   42
6       US  1997   64
7    JAPAN  1991  100
8    JAPAN  1992   30
9    JAPAN  1993   70
10   JAPAN  1994   85
11   JAPAN  1995   30
12   JAPAN  1996   42
13   JAPAN  1997   64
14      CN  1991   50
15      CN  1992   22
16      CN  1993   30
17      CN  1994   65
18      CN  1995   70
19      CN  1996   66
20      CN  1997   60

使用plotly.express

现在您的数据是长格式的,您可以使用 plotly.express 如下

import plotly.express as px
fig = px.scatter(df,
                 x="year",
                 y="country",
                 size="n",
                 color="country",
                 text="n",
                 size_max=50 # you need this otherwise the bubble are too small
                )

fig.update_layout(plot_bgcolor='rgb(10, 10, 10)',  
                  paper_bgcolor='rgb(20, 55, 100)',  
                  font={'size': 15,
                        'family': 'sans-serif',
                        'color': 'rgb(255, 255, 255)'
                       },
                  width=1000,
                  height=500,
                  xaxis=dict(title='Output of grapes per year in selected countries', ),  
                  showlegend=False,
                  margin=dict(l=100, r=100, t=100, b=100),
                  hovermode = False,)
# Uncomment this if you don't wont country as yaxis title
# fig.layout.yaxis.title.text = None
fig.show()

代码已更新到版本 2,可以从 .xlsx 文件读取数据并绘制气泡图。 名为'grape output.xlsx'的原始数据与之前的数据相比增加了新的项目:

             1991  1992  1993  1994  1995  1996  1997  1998  1999
         US    10    14    16    18    20    42    64   100    50
      JAPAN   100    30    70    85    30    42    64    98    24
         CN    50    22    30    65    70    66    60    45    45
      INDIA    90    88    35    50    90    60    40    66    76
         UK    40    50    70    50    25    30    22    40    60

代码如下:

# Version 2 
import plotly as py
import plotly.graph_objs as go
import openpyxl
import pandas as pd


wb = openpyxl.load_workbook('grape output.xlsx')
sheet = wb['Sheet1']
row_max = sheet.max_row
col_max = sheet.max_column
first_row_list = []
first_col_list = []
for col_n in range(2, col_max+1):
    first_row_list.append(sheet.cell(row=1, column=col_n).value)
for row_n in range(2,row_max+1):
    first_col_list.append(sheet.cell(row=row_n, column=1).value)

data_all = pd.read_excel('grape output.xlsx')
data = data_all.loc[:,first_row_list]

df = pd.DataFrame(data)
df.index = first_col_list
colors = ['rgb(150,204,90)','rgb(255, 130, 71)','rgb(255, 193, 37)','rgb(180,240,190)','rgb(255, 10, 1)',
          'rgb(25, 19, 3)','rgb(100, 100, 100)','rgb(45,24,200)','rgb(33, 58, 108)','rgb(35, 208, 232)']

data = [go.Scatter(
    x=df.columns,
    y=[country]*len(df.columns),
    mode='markers+text',
    marker=dict(
        color=colors[num],
        size= df.loc[country],
        showscale = False,
        ),
    text=list(map(str, df.loc[country])),
    textposition='middle center',
    )
    for num, country in enumerate(reversed(df.index))
]

layout = go.Layout(plot_bgcolor='rgb(10, 10, 10)',
                   paper_bgcolor='rgb(20, 55, 100)',
                   font={
                       'size': 15,
                       'family': 'sans-serif',
                       'color': 'rgb(255, 255, 255)'
                   },
                   width=1000,
                   height=800,
                   xaxis=dict(title='Output of grapes per year in US, JAPAN and CN'),
                   showlegend=False,
                   margin=dict(l=100, r=100, t=100, b=100),
                   hovermode = False,
                   )

fig = go.Figure(data=data, layout=layout)
py.offline.plot(fig, filename='basic-scatter.html')

现在的结果是这样的: 还有一些小问题:

  1. 如何去除1990和2000这两个数字以及1990和2000的白色垂直线?
  2. 1991、1993、1995、1997、1999如何用白线画出横坐标显示这些年份?

请更正代码 Versinon 2 以改进它。谢谢!