论文辅助笔记：TEMPO之feature.py

2024-05-03 02:58:02
开发
29

"""
Reference: Informer (2020).
"""

0 导入库


from typing import List
import numpy as np
import pandas as pd
from pandas.tseries import offsets
from pandas.tseries.frequencies import to_offset

1 TimeFeature

基类，没有实际功能，只定义了基本接口，包括 __call__ 和 __repr__ 方法

python笔记：类方法之 __call__ 和__repr__-CSDN博客

class TimeFeature:
    def __init__(self):
        pass

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        pass

    def __repr__(self):
        return self.__class__.__name__ + "()"

2 SecondOfMinute

提取时间戳对应分钟内的秒数，并标准化到 [-0.5, 0.5] 范围内

class SecondOfMinute(TimeFeature):
    """Minute of hour encoded as value between [-0.5, 0.5]"""

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        return index.second / 59.0 - 0.5

3 MinuteOfHour

提取时间戳对应小时内的分钟数，并标准化到 [-0.5, 0.5] 范围内

class MinuteOfHour(TimeFeature):
    """Minute of hour encoded as value between [-0.5, 0.5]"""

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        return index.minute / 59.0 - 0.5

4 HourOfDay

提取时间戳对应一天内的小时数，并标准化到 [-0.5, 0.5] 范围内

class HourOfDay(TimeFeature):
    """Hour of day encoded as value between [-0.5, 0.5]"""

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        return index.hour / 23.0 - 0.5

5 DayOfWeek

提取时间戳对应一周内的天数（星期几），并标准化到 [-0.5, 0.5] 范围内

class DayOfWeek(TimeFeature):
    """Hour of day encoded as value between [-0.5, 0.5]"""

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        return index.dayofweek / 6.0 - 0.5

6 DayOfMonth

提取时间戳对应一个月内的天数，并标准化到 [-0.5, 0.5] 范围内

class DayOfMonth(TimeFeature):
    """Day of month encoded as value between [-0.5, 0.5]"""

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        return (index.day - 1) / 30.0 - 0.5

7 DayOfYear

提取时间戳对应一年内的天数，并标准化到 [-0.5, 0.5] 范围内

class DayOfYear(TimeFeature):
    """Day of year encoded as value between [-0.5, 0.5]"""

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        return (index.dayofyear - 1) / 365.0 - 0.5

8 MonthOfYear

提取时间戳对应一年内的天数，并标准化到 [-0.5, 0.5] 范围内

class MonthOfYear(TimeFeature):
    """Month of year encoded as value between [-0.5, 0.5]"""

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        return (index.month - 1) / 11.0 - 0.5

9 WeekOfYear

提取时间戳对应一年的周数，并标准化到 [-0.5, 0.5] 范围内

class WeekOfYear(TimeFeature):
    """Week of year encoded as value between [-0.5, 0.5]"""

    def __call__(self, index: pd.DatetimeIndex) -> np.ndarray:
        return (index.week - 1) / 52.0 - 0.5

10time_features_from_frequency_str

根据传入的频率字符串 freq_str 返回相应的时间特征列表

def time_features_from_frequency_str(freq_str: str) -> List[TimeFeature]:

    features_by_offsets = {
        offsets.YearEnd: [],
        offsets.QuarterEnd: [MonthOfYear],
        offsets.MonthEnd: [MonthOfYear],
        offsets.Week: [DayOfMonth, WeekOfYear],
        offsets.Day: [DayOfWeek, DayOfMonth, DayOfYear],
        offsets.BusinessDay: [DayOfWeek, DayOfMonth, DayOfYear],
        offsets.Hour: [HourOfDay, DayOfWeek, DayOfMonth, DayOfYear],
        offsets.Minute: [
            MinuteOfHour,
            HourOfDay,
            DayOfWeek,
            DayOfMonth,
            DayOfYear,
        ],
        offsets.Second: [
            SecondOfMinute,
            MinuteOfHour,
            HourOfDay,
            DayOfWeek,
            DayOfMonth,
            DayOfYear,
        ],
    }
    '''
    这是一个映射 pandas 时间偏移类型到对应时间特征类的字典：
        键是 offsets 模块中的各种时间偏移类型（如 offsets.YearEnd、offsets.MonthEnd 等）。
        值是与这些偏移类型相关的时间特征类列表，如 MonthOfYear、WeekOfYear 等
    '''


    offset = to_offset(freq_str)
    '''
    将 字符串类型的freq_str 转换为一个 pandas 时间偏移对象 offset，以便在 features_by_offsets 字典中查找对应的特征类列表
    '''


    for offset_type, feature_classes in features_by_offsets.items():
        if isinstance(offset, offset_type):
            return [cls() for cls in feature_classes]
    '''
        遍历 features_by_offsets 字典，将 offset 与字典中的键类型进行比较

        如果匹配，则返回与之对应的特征类实例列表
    '''

    supported_freq_msg = f"""
    Unsupported frequency {freq_str}
    The following frequencies are supported:
        Y   - yearly
            alias: A
        M   - monthly
        W   - weekly
        D   - daily
        B   - business days
        H   - hourly
        T   - minutely
            alias: min
        S   - secondly
    """

    '''
    如果 offset 与任何 features_by_offsets 中的键都不匹配，则函数会抛出一个 RuntimeError 异常
    并提供一个错误消息，指出当前支持的频率列表，包括 Y, M, W, D, B, H, T, S 及其别名
    '''
    raise RuntimeError(supported_freq_msg)

11 time_features

从一个包含日期信息的 dates 数据框中提取日期特征，并根据参数 timeenc 和 freq 对这些特征进行编码

def time_features(dates, timeenc=1, freq="h"):
    
    if timeenc == 0:
        dates["month"] = dates.date.apply(lambda row: row.month, 1)
        dates["day"] = dates.date.apply(lambda row: row.day, 1)
        dates["weekday"] = dates.date.apply(lambda row: row.weekday(), 1)
        dates["hour"] = dates.date.apply(lambda row: row.hour, 1)
        dates["minute"] = dates.date.apply(lambda row: row.minute, 1)
        dates["minute"] = dates.minute.map(lambda x: x // 15)
        '''
        首先为 dates 数据框添加一些列，如 month, day, weekday, hour, minute，以此记录日期信息的相应特征

        特别地，minute 列会使用 .map(lambda x: x // 15) 将分钟数标准化到一个 0 到 3 之间的值，对应 15 分钟的时间段
        '''
        freq_map = {
            "y": [],
            "m": ["month"],
            "w": ["month"],
            "d": ["month", "day", "weekday"],
            "b": ["month", "day", "weekday"],
            "h": ["month", "day", "weekday", "hour"],
            "t": ["month", "day", "weekday", "hour", "minute"],
        }
        '''
        freq_map 是一个字典，将 freq 字符串映射到特定的特征列表，然后返回这些特征对应的列的值。
        '''
        return dates[freq_map[freq.lower()]].values
    if timeenc == 1:
        dates = pd.to_datetime(dates.date.values)
        #会将 dates 转换为 pd.DatetimeIndex 对象
        return np.vstack(
            [feat(dates) for feat in time_features_from_frequency_str(freq)]
        ).transpose(1, 0)
        '''
        调用 time_features_from_frequency_str 根据 freq 提取时间特征类列表。
            结果是一个二维数组，每一列对应一种时间特征

            通过 np.vstack 将不同时间特征垂直堆叠，并转置，使得每一行代表日期，每一列代表特征

     
               
        '''

11.1 举例：

假设有一个包含日期信息的数据框 dates，并且我们希望提取与这些日期相关的时间特征：

import pandas as pd

# 构造一个数据框，包含日期信息
data = {'date': ["2024-05-01 12:00:00", 
                 "2024-05-02 08:30:00", 
                 "2024-05-03 14:45:00"]}
dates = pd.DataFrame(data)
dates['date'] = pd.to_datetime(dates['date'])  
# 将日期字符串转换为 pandas 的 datetime 对象

调用 time_features 函数提取时间特征。比如说我们希望提取以小时为粒度的时间特征，并且 timeenc 为 1：

time_features(dates, timeenc=1, freq="H")
'''
array([[ 0.02173913, -0.16666667, -0.5       , -0.16849315],
       [-0.15217391,  0.        , -0.46666667, -0.16575342],
       [ 0.10869565,  0.16666667, -0.43333333, -0.1630137 ]])

'''

输出将是一个二维数组，其中每一列对应一种时间特征，每一行对应 dates 中的一行

HourOfDay: 对应于每个日期的小时特征，范围 [-0.5, 0.5]
DayOfWeek: 对应于每个日期的星期几，范围 [-0.5, 0.5]
DayOfMonth: 对应于每个日期的一个月中的第几天，范围 [-0.5, 0.5]
DayOfYear: 对应于每个日期在一年中的第几天，范围 [-0.5, 0.5]

11.2 return np.vstack(
[feat(dates) for feat in time_features_from_frequency_str(freq)]
).transpose(1, 0)的解读

time_features_from_frequency_str('H')
#[HourOfDay(), DayOfWeek(), DayOfMonth(), DayOfYear()]

for feat in time_features_from_frequency_str('H'):
    print(feat(dates1))
'''
Index([0.021739130434782594, -0.15217391304347827, 0.10869565217391308], dtype='float64')
Index([-0.16666666666666669, 0.0, 0.16666666666666663], dtype='float64')
Index([-0.5, -0.4666666666666667, -0.43333333333333335], dtype='float64')
Index([-0.16849315068493148, -0.16575342465753423, -0.16301369863013698], dtype='float64')
'''


[feat(dates1) for feat in time_features_from_frequency_str('H')]
'''
[Index([0.021739130434782594, -0.15217391304347827, 0.10869565217391308], dtype='float64'),
 Index([-0.16666666666666669, 0.0, 0.16666666666666663], dtype='float64'),
 Index([-0.5, -0.4666666666666667, -0.43333333333333335], dtype='float64'),
 Index([-0.16849315068493148, -0.16575342465753423, -0.16301369863013698], dtype='float64')]
'''


np.vstack(
            [feat(dates1) for feat in time_features_from_frequency_str('H')]
        )
'''
array([[ 0.02173913, -0.15217391,  0.10869565],
       [-0.16666667,  0.        ,  0.16666667],
       [-0.5       , -0.46666667, -0.43333333],
       [-0.16849315, -0.16575342, -0.1630137 ]])
'''

原文地址:https://blog.csdn.net/qq_40206371/article/details/138394699 本文来自互联网用户投稿，该文观点仅代表作者本人，不代表本站立场。本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如若转载，请注明出处：https://www.suanlizi.com/kf/1786107897474846720.html 如若内容造成侵权/违法违规/事实不符，请联系《酸梨子》网邮箱：1419361763@qq.com进行投诉反馈，一经查实，立即删除！

阅读全部