Pandas可视化时间序列

Question

I have time series class data. First column contains join time . Second column contains leave time for various students.Third column is Class ID . So there is possibility that student left the class in 10 min and again joined it after some time. His time is again recorded for both activities. I want to visualize data in order to see at what time maximum students attended class.

data['Join Time Hour'] = data['Join Time'].dt.hour        
data['Join Time Date'] = data['Join Time'].dt.date
data['Leave Time Hour'] = data['Leave Time'].dt.hour        
data['Leave Time Date'] = data['Leave Time'].dt.date

My approach:

1.

# Added one dummy
data['Dummy Column'] = 1     

2.

data1 = (
    pd.pivot_table(data, 
                   values='Dummy Column', 
                   index='Join Time Date', 
                   columns='Join Time Hour', 
                   aggfunc='sum')  
)

3.

sns.heatmap(data1, cmap="Blues")
plt.show()

Otput:

This output gives me heatmap based on sum of dummy variable for given hour of class. Its not considering leave time in visualization.

I want

1. visualize (in heatmap or other visualization) exactly which moments in videos are most-watched, and
2. where student tend to drop off last time

Thanks!!

Answer 1

I think a sankey diagram can resolve your problem. Below is my test code.

import pandas as pd
import numpy as np
from itertools import product
import seaborn as sns

from plotly.offline import init_notebook_mode, iplot

init_notebook_mode(connected=True)

# generate test hours between 8:00 and 12:00
times = pd.date_range("202005010800", '202005011200', freq="H")

# generate test 1000 students id 
students = list(np.arange(1000))

# generate test class
classes = ["class A", "class B"]

# make nodes including classes, join times and leave times
nodes = []
join_time = [time.strftime("join time: %Y%m%d %H") for time in times]
leave_time = [time.strftime("leave time: %Y%m%d %H") for time in times]
nodes.extend(classes)
nodes.extend(join_time)
nodes.extend(leave_time)

# every node has a color and an id
df_nodes = pd.DataFrame(nodes, columns=['node'])
df_nodes['color'] = list(sns.palettes.xkcd_rgb.values())[:len(df_nodes)]
df_nodes['node id'] = df_nodes.index

# nodes dict used in links 
nodes_id_dict = dict(zip(df_nodes['node'], df_nodes['node id']))
nodes_color_dict = dict(zip(df_nodes['node'], df_nodes['color']))

# make records
records = product(times, times, students, classes)

# filter records whose leave time is later than join time
records = [record for record in records if record[1] > record[0]]

df_records = pd.DataFrame(
    records, columns=['join time', 'leave time', 'student', 'class']
)

#  pick 10000 records randomly
df_records = df_records.sample(10000)

# format time to use nodes dict 
df_records['join time'] = df_records['join time'].\
dt.strftime("join time: %Y%m%d %H")
df_records['leave time'] = df_records['leave time'].\
dt.strftime("leave time: %Y%m%d %H")

# the first link from class to join time
class_join_time = df_records.groupby(['class', 'join time']) \
['student'].count().reset_index()
class_join_time.columns = ['source', 'target', 'value']

# the second link from join time to leave time
join_leave_time = df_records.groupby(['join time', 'leave time'])\
['student'].count().reset_index()
join_leave_time.columns = ['source', 'target', 'value']

# merge the two links
df_links = pd.concat([class_join_time, join_leave_time])

# use nodes dict to get node id and link color
# you can generate colors customly
df_links['source id'] = df_links['source'].replace(nodes_id_dict)
df_links['target id'] = df_links['target'].replace(nodes_id_dict)
df_links['link color'] = df_links['target'].replace(nodes_color_dict)

# configure the data_trace
data_trace = dict(
    type='sankey',
    domain = dict(
      x =  [0,1],
      y =  [0,1]
    ),
    orientation = "h", # horizontal
    valueformat = ".0f", 
    node = dict(
       pad = 10,
       line = dict(
         color = "rgba(0,0,0,0.5)",
         width = 0.1
      ),
         label =  df_nodes['node'],
         color = df_nodes['color']
    ),
    link = dict(
         source = df_links['source id'],
         target = df_links['target id'],
         value = df_links['value'],
         color = df_links['link color'],
         line = dict(
             color = "rgba(0,0,0,0.5)",
             width = 0.1
        ),
    )
)
# cofigure the layout
layout = dict(
    title = "Sankey Diagram Test",
    height = 640,
    width = 900,
    font = dict(
        size=12
    )
)

# plot
fig = dict(data=[data_trace], layout=layout)
iplot(fig, validate=False)

The sankey diagram is shown below:

Answer 2

I tried visualization with a heat map. The data is created appropriately.

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

join_t = pd.date_range('2020-04-01 09:00:00', '2020-04-30 20:00:00', freq='BH')
leave_t = pd.date_range('2020-04-01 09:00:00', '2020-04-30 20:00:00', freq='BH')
persons = np.random.randint(1,40,(176,))

df = pd.DataFrame({'Join Time':pd.to_datetime(join_t),
                  'Leave Time':pd.to_datetime(leave_t),
                  'Persons':persons})

df['Leave Time'] = df['Leave Time'].shift(-1, fill_value=df.iloc[-1]['Leave Time'])
df['Join Time Hour'] = df['Join Time'].dt.hour    
df['Join Time Date'] = df['Join Time'].dt.date
df['Leave Time Hour'] = df['Leave Time'].dt.hour       
df['Leave Time Date'] = df['Leave Time'].dt.date

df.loc[:,['Join Time Date','Join Time Hour','Persons']]


fig = plt.figure(figsize=(8,6),dpi=144)
ax = fig.add_subplot(111)

data = df.pivot(index='Join Time Date', columns='Join Time Hour', values='Persons')
ax = sns.heatmap(data, ax=ax, annot=True, cmap="YlGnBu")

plt.show()