Pandas和Matplotlib - fill_between()vs datetime64

Question

有一个Pandas DataFrame:

<class 'pandas.core.frame.DataFrame'>
Int64Index: 300 entries, 5220 to 5519
Data columns (total 3 columns):
Date             300 non-null datetime64[ns]
A                300 non-null float64
B                300 non-null float64
dtypes: datetime64[ns](1), float64(2)
memory usage: 30.5 KB

我想绘制A和B系列与日期的关系.

plt.plot_date(data['Date'], data['A'], '-')
plt.plot_date(data['Date'], data['B'], '-')

然后我想在A和B系列之间的区域上应用fill_between():

plt.fill_between(data['Date'], data['A'], data['B'],
                where=data['A'] >= data['B'],
                facecolor='green', alpha=0.2, interpolate=True)

哪个输出:

TypeError: ufunc 'isfinite' not supported for the input types, and the inputs
could not be safely coerced to any supported types according to the casting 
rule ''safe''

matplotlib是否在fill_between()函数中接受pandas datetime64对象？我应该将其转换为不同的日期类型吗？

Answer 1

Pandas注册了一个转换器,matplotlib.units.registry其中将许多日期时间类型(例如pandas DatetimeIndex和datetime64dtype的numpy数组)转换为matplotlib datenums,但它不处理Series带有dtype的Pandas datetime64.

In [67]: import pandas.tseries.converter as converter

In [68]: c = converter.DatetimeConverter()

In [69]: type(c.convert(df['Date'].values, None, None))
Out[69]: numpy.ndarray              # converted (good)

In [70]: type(c.convert(df['Date'], None, None))
Out[70]: pandas.core.series.Series  # left unchanged

fill_between 检查并使用转换器处理数据(如果存在).

因此,作为一种解决方法,您可以将日期转换为datetime64's 的NumPy数组:

d = data['Date'].values
plt.fill_between(d, data['A'], data['B'],
                where=data['A'] >= data['B'],
                facecolor='green', alpha=0.2, interpolate=True)

例如,

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

N = 300
dates = pd.date_range('2000-1-1', periods=N, freq='D')
x = np.linspace(0, 2*np.pi, N)
data = pd.DataFrame({'A': np.sin(x), 'B': np.cos(x),
               'Date': dates})
plt.plot_date(data['Date'], data['A'], '-')
plt.plot_date(data['Date'], data['B'], '-')

d = data['Date'].values
plt.fill_between(d, data['A'], data['B'],
                where=data['A'] >= data['B'],
                facecolor='green', alpha=0.2, interpolate=True)
plt.xticks(rotation=25)
plt.show()

Answer 2

WillZ指出,Pandas 0.21破坏了unutbu的解决方案.但是,将日期时间转换为日期会对数据分析产生显着的负面影响.此解决方案目前有效并保持日期时间:

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

N = 300
dates = pd.date_range('2000-1-1', periods=N, freq='ms')
x = np.linspace(0, 2*np.pi, N)
data = pd.DataFrame({'A': np.sin(x), 'B': np.cos(x),
           'Date': dates})
d = data['Date'].dt.to_pydatetime()
plt.plot_date(d, data['A'], '-')
plt.plot_date(d, data['B'], '-')


plt.fill_between(d, data['A'], data['B'],
            where=data['A'] >= data['B'],
            facecolor='green', alpha=0.2, interpolate=True)
plt.xticks(rotation=25)
plt.show()

编辑:根据jedi的评论,我开始确定以下三个选项中最快的方法:

• method1 =原始答案
• method2 = jedi的评论+原始答案
• method3 = jedi的评论

method2略快,但更加一致,因此我编辑了上面的答案以反映最佳方法.

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


N = 300
dates = pd.date_range('2000-1-1', periods=N, freq='ms')
x = np.linspace(0, 2*np.pi, N)
data = pd.DataFrame({'A': np.sin(x), 'B': np.cos(x),
           'Date': dates})
time_data = pd.DataFrame(columns=['1', '2', '3', '4', '5', '6', '7', '8', '9', '10'])
method1 = []
method2 = []
method3 = []
for i in range(0, 10):
    start = time.clock()
    for i in range(0, 500):
        d = [pd.Timestamp(x).to_pydatetime() for x in data['Date']]
        #d = data['Date'].dt.to_pydatetime()
        plt.plot_date(d, data['A'], '-')
        plt.plot_date(d, data['B'], '-')


        plt.fill_between(d, data['A'], data['B'],
            where=data['A'] >= data['B'],
            facecolor='green', alpha=0.2, interpolate=True)
        plt.xticks(rotation=25)
        plt.gcf().clear()
    method1.append(time.clock() - start)

for i  in range(0, 10):
    start = time.clock()
    for i in range(0, 500):
        #d = [pd.Timestamp(x).to_pydatetime() for x in data['Date']]
        d = data['Date'].dt.to_pydatetime()
        plt.plot_date(d, data['A'], '-')
        plt.plot_date(d, data['B'], '-')


        plt.fill_between(d, data['A'], data['B'],
            where=data['A'] >= data['B'],
            facecolor='green', alpha=0.2, interpolate=True)
        plt.xticks(rotation=25)
        plt.gcf().clear()
    method2.append(time.clock() - start)

for i in range(0, 10):
    start = time.clock()
    for i in range(0, 500):
        #d = [pd.Timestamp(x).to_pydatetime() for x in data['Date']]
        #d = data['Date'].dt.to_pydatetime()
        plt.plot_date(data['Date'].dt.to_pydatetime(), data['A'], '-')
        plt.plot_date(data['Date'].dt.to_pydatetime(), data['B'], '-')


        plt.fill_between(data['Date'].dt.to_pydatetime(), data['A'], data['B'],
            where=data['A'] >= data['B'],
            facecolor='green', alpha=0.2, interpolate=True)
        plt.xticks(rotation=25)
        plt.gcf().clear()
    method3.append(time.clock() - start)

time_data.loc['method1'] = method1
time_data.loc['method2'] = method2
time_data.loc['method3'] = method3
print(time_data)
plt.errorbar(time_data.index, time_data.mean(axis=1), yerr=time_data.std(axis=1))