熊猫 - 如何分组和绘制每周每天的每个小时

Question

I need help figuring out how to plot sub-plots for easy comparison from my dataframe shown:

  Date                   A        B         C              
2017-03-22 15:00:00     obj1    value_a    other_1
2017-03-22 14:00:00     obj2    value_ns   other_5
2017-03-21 15:00:00     obj3    value_kdsa other_23
2014-05-08 17:00:00     obj2    value_as   other_4
2010-07-01 20:00:00     obj1    value_as   other_0

I am trying to graph the occurrences of each hour for each respective day of the week. So count the number of occurrences for each day of the week and hour and plot them on subplots like the ones shown below.

If this question sounds confusing please let me know if you have any questions. Thanks.

Answer 1

You can accomplish this with multiple groupby . Since we know there are 7 days in a week, we can specify that number of panels. If you groupby(df.Date.dt.dayofweek) , you can use the group index as the index for your subplot axes:

Sample Data

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

n = 10000
np.random.seed(123)
df = pd.DataFrame({'Date': pd.date_range('2010-01-01', freq='1.09min', periods=n),
                   'A': np.random.randint(1,10,n),
                   'B': np.random.normal(0,1,n)})

Code:

fig, ax = plt.subplots(ncols=7, figsize=(30,5))
plt.subplots_adjust(wspace=0.05)  #Remove some whitespace between subplots

for idx, gp in df.groupby(df.Date.dt.dayofweek):
    ax[idx].set_title(gp.Date.dt.day_name().iloc[0])  #Set title to the weekday

    (gp.groupby(gp.Date.dt.hour).size().rename_axis('Tweet Hour').to_frame('')
        .reindex(np.arange(0,24,1)).fillna(0)
        .plot(kind='bar', ax=ax[idx], rot=0, ec='k', legend=False))

    # Ticks and labels on leftmost only
    if idx == 0:
        _ = ax[idx].set_ylabel('Counts', fontsize=11)

    _ = ax[idx].tick_params(axis='both', which='major', labelsize=7,
                            labelleft=(idx == 0), left=(idx == 0))

# Consistent bounds between subplots. 
lb, ub = list(zip(*[axis.get_ylim() for axis in ax]))
for axis in ax:
    axis.set_ylim(min(lb), max(ub)) 

plt.show()

---

If you'd like to make the aspect ratio less extreme, then consider plotting a 4x2 grid. It's a very similar plot as above, once we flatten the axis array. There's some integer and remainder division to figure out which axes need the labels.

fig, ax = plt.subplots(nrows=2, ncols=4, figsize=(20,10))
fig.delaxes(ax[1,3])  #7 days in a week, remove 8th panel
ax = ax.flatten()  #Far easier to work with a flattened array

lsize=8
plt.subplots_adjust(wspace=0.05, hspace=0.15)  #Remove some whitespace between subplots

for idx, gp in df.groupby(df.Date.dt.dayofweek):
    ax[idx].set_title(gp.Date.dt.day_name().iloc[0])  #Set title to the weekday

    (gp.groupby(gp.Date.dt.hour).size().rename_axis([None]).to_frame()
        .reindex(np.arange(0,24,1)).fillna(0)
        .plot(kind='bar', ax=ax[idx], rot=0, ec='k', legend=False))

    # Titles on correct panels
    if idx%4 == 0:
        _ = ax[idx].set_ylabel('Counts', fontsize=11)
    if (idx//4 == 1) | (idx%4 == 3):
        _ = ax[idx].set_xlabel('Tweet Hour', fontsize=11) 

    # Ticks on correct panels
    _ = ax[idx].tick_params(axis='both', which='major', labelsize=lsize,
                            labelbottom=(idx//4 == 1) | (idx%4 == 3), 
                            bottom=(idx//4 == 1) | (idx%4 == 3),
                            labelleft=(idx%4 == 0), 
                            left=(idx%4 == 0))

# Consistent bounds between subplots. 
lb, ub = list(zip(*[axis.get_ylim() for axis in ax]))
for axis in ax:
    axis.set_ylim(min(lb), max(ub)) 

plt.show()

Answer 2

What about using seaborn ? sns.FacetGrid was made for this:

import pandas as pd
import seaborn as sns

# make some data
date = pd.date_range('today', periods=100, freq='2.5H')

# put in dataframe
df = pd.DataFrame({
    'date' : date
})

# create day_of_week and hour columns
df['dow'] = df.date.dt.day_name()
df['hour'] = df.date.dt.hour

# create facet grid
g = sns.FacetGrid(data=df.groupby([
    'dow',
    'hour'
]).hour.count().to_frame(name='day_hour_count').reset_index(), col='dow', col_order=[
    'Sunday',
    'Monday',
    'Tuesday',
    'Wednesday',
    'Thursday',
    'Friday',
    'Saturday'
], col_wrap=3)

# map barplot to each subplot
g.map(sns.barplot, 'hour', 'day_hour_count');