Pandas 如何利用时间序列

时间序列数据主要用于处理随时间变化的数据。处理这些数据在时间序列数据的数据分析中起着非常重要的作用。Pandas是Python中一个流行的数据处理和分析库，提供了强大的功能来处理时间序列数据。在本文中，我们将通过示例和解释来了解如何有效地利用Pandas中的时间序列。

利用时间序列数据的方式

在下面的方法中，我们将使用从Kaggle获取的Electric_ptoduction时间序列数据集。您可以从这里下载数据集。

导入和操作时间序列数据

在Pandas中处理时间序列数据时，我们首先需要导入必要的库并将数据加载到DataFrame中。Pandas提供了从不同来源（包括CSV文件、数据库和Web API）读取时间序列数据的各种方法。在数据加载后，Pandas提供了强大的工具来操作、清理和预处理时间序列数据。

import pandas as pd

# Load time series data from a CSV file
data = pd.read_csv('Electric_Production.csv')

# Display the first few rows of the DataFrame
print(data.head())

# Set the 'timestamp' column as the index
data['DATE'] = pd.to_datetime(data['DATE'])
data.set_index('DATE', inplace=True)

# Resample the data to a daily frequency
daily_data = data.resample('D').mean()

输出

DATE  IPG2211A2N
0  1/1/1985     72.5052
1  2/1/1985     70.6720
2  3/1/1985     62.4502
3  4/1/1985     57.4714
4  5/1/1985     55.3151

索引和切片时间序列数据

Pandas包含多种索引和切片方法，用于从时间序列数据中提取特定的时间段或观测值。Pandas中的DateTimeIndex可以基于时间进行直观的索引和选择。

import pandas as pd

# 从CSV文件加载时间序列数据
data = pd.read_csv('Electric_Production.csv')

# 将 'timestamp' 列设置为索引
data['DATE'] = pd.to_datetime(data['DATE'])
data.set_index('DATE', inplace=True)

# 将数据重新采样到每天的频率
daily_data = data.resample('D').mean()

# 选择特定日期范围的数据
subset_1 = data['2017-01-01':'2017-10-30']
print(subset_1)

# 选择特定月份的数据
subset_2 = data[data.index.month == 3]
print(subset_2)

# 选择特定年份的数据
subset_3 = data[data.index.year == 2016]
print(subset_3)

输出

IPG2211A2N
DATE
2017-01-01    114.8505
2017-02-01     99.4901
2017-03-01    101.0396
2017-04-01     88.3530
2017-05-01     92.0805
2017-06-01    102.1532
2017-07-01    112.1538
2017-08-01    108.9312
2017-09-01     98.6154
2017-10-01     93.6137
            IPG2211A2N
DATE
1985-03-01     62.4502
1986-03-01     62.2221
1987-03-01     65.6100
1988-03-01     70.2928
1989-03-01     73.3523
1990-03-01     73.1964
1991-03-01     73.3650
1992-03-01     74.5275
1993-03-01     79.4747
1994-03-01     79.2456
1995-03-01     81.2661
1996-03-01     86.9356
1997-03-01     83.0125
1998-03-01     86.5549
1999-03-01     90.7381
2000-03-01     88.0927
2001-03-01     92.8283
2002-03-01     93.2556
2003-03-01     94.5532
2004-03-01     95.4029
2005-03-01     98.9565
2006-03-01     98.4017
2007-03-01     99.1925
2008-03-01    100.4386
2009-03-01     97.8529
2010-03-01     98.2672
2011-03-01     99.1028
2012-03-01     93.5772
2013-03-01    102.9948
2014-03-01    104.7631
2015-03-01    104.4706
2016-03-01     95.3548
2017-03-01    101.0396
            IPG2211A2N
DATE
2016-01-01    117.0837
2016-02-01    106.6688
2016-03-01     95.3548
2016-04-01     89.3254
2016-05-01     90.7369
2016-06-01    104.0375
2016-07-01    114.5397
2016-08-01    115.5159
2016-09-01    102.7637
2016-10-01     91.4867
2016-11-01     92.8900
2016-12-01    112.7694

处理缺失数据

时间序列数据经常包含缺失值，这可能会妨碍分析和建模。Pandas提供了几种处理缺失数据的方法，例如插值、向前填充或向后填充。这些方法有助于确保时间序列的连续性。

import pandas as pd

# Load time series data from a CSV file
data = pd.read_csv('Electric_Production.csv')

# Display the first few rows of the DataFrame
# print(data.head())

# Set the 'timestamp' column as the index
data['DATE'] = pd.to_datetime(data['DATE'])
data.set_index('DATE', inplace=True)

# Resample the data to a daily frequency
daily_data = data.resample('D').mean()

## Interpolate missing values
data['value'] = data['value'].interpolate()
print(data.head())

# Forward-fill missing values
data['value'] = data['value'].ffill()
print(data.head())

# Backward-fill missing values
data['value'] = data['value'].bfill()
print(data.head())

输出

value
DATE
1985-01-01  72.5052
1985-02-01  70.6720
1985-03-01  64.0717
1985-04-01  57.4714
1985-05-01  55.3151
              value
DATE
1985-01-01  72.5052
1985-02-01  70.6720
1985-03-01  64.0717
1985-04-01  57.4714
1985-05-01  55.3151
              value
DATE
1985-01-01  72.5052
1985-02-01  70.6720
1985-03-01  64.0717
1985-04-01  57.4714
1985-05-01  55.3151

重新采样和频率转换

重新采样涉及改变时间序列数据的频率。Pandas提供了上采样（增加频率）和下采样（减少频率）时间序列数据的方法。这允许在不同的时间间隔上对数据进行聚合或插值处理。

import pandas as pd

# Load time series data from a CSV file
data = pd.read_csv('Electric_Production.csv')

# Display the first few rows of the DataFrame
# print(data.head())

# Set the 'timestamp' column as the index
data['DATE'] = pd.to_datetime(data['DATE'])
data.set_index('DATE', inplace=True)

# Resample the data to a daily frequency
daily_data = data.resample('D').mean()
print(daily_data.head())

# Resample the data to a weekly frequency, taking the mean value
weekly_data = data.resample('W').mean()
print(weekly_data.head())

# Resample the data to a monthly frequency, taking the sum value
monthly_data = data.resample('M').sum()
print(weekly_data.head())

输出

value
DATE
1985-01-01  72.5052
1985-01-02      NaN
1985-01-03      NaN
1985-01-04      NaN
1985-01-05      NaN
              value
DATE
1985-01-06  72.5052
1985-01-13      NaN
1985-01-20      NaN
1985-01-27      NaN
1985-02-03  70.6720
              value
DATE
1985-01-06  72.5052
1985-01-13      NaN
1985-01-20      NaN
1985-01-27      NaN
1985-02-03  70.6720

绘制和可视化时间序列数据

Pandas与Matplotlib集成，后者是一个流行的数据可视化库，使得创建洞察力的图形和可视化时间序列数据变得容易。可视化可以帮助理解数据中的趋势、模式和异常。

import pandas as pd
import matplotlib.pyplot as plt

# Load time series data from a CSV file
data = pd.read_csv('Electric_Production.csv')

# Display the first few rows of the DataFrame
# print(data.head())

# Set the 'timestamp' column as the index
data['DATE'] = pd.to_datetime(data['DATE'])
data.set_index('DATE', inplace=True)

# Plot the time series data
data.plot()
plt.title('Time Series Data')
plt.xlabel('Date')
plt.ylabel('Value')
plt.show()