python学习-时间序列
时间序列需要导入
from datetime import datetime #以毫秒形式存储日期和时间 from datetime import timedelta #表示两个datetime的时间差
1、日期类型和字符串互转
#日期转字符串 stamp = datetime(2011,1,3)#datetime.datetime(2011, 1, 3, 0, 0) str(stamp)#'2011-01-03 00:00:00' stamp.strftime('%Y-%m-%d') #'2011-01-03' #字符串转日期 value='2011-01-03' datetime.strptime(value,'%Y-%m-%d')#datetime.datetime(2011, 1, 3, 0, 0) datestrs = ["7/6/2011", "8/6/2011"] [datetime.strptime(x, "%m/%d/%Y") for x in datestrs] from dateutil.parser import parse #日期解析工具，可以解析大部分格式 parse('Jan 31 1997 10:45 PM') #datetime.datetime(1997, 1, 31, 22, 45) #pandas处理成组日期 datestrs = ["2011-07-06 12:00:00", "2011-08-06 00:00:00"] pd.to_datetime(datestrs) #处理缺失值空字符串或者None值 pd.to_datetime(datestrs+[None])
2、时间序列基础
| Alias | Offset type | Description |
|---|---|---|
D |
Day |
Calendar daily |
B |
BusinessDay |
Business daily |
H |
Hour |
Hourly |
T or min |
Minute |
Minutely |
S |
Second |
Secondly |
L or ms |
Milli |
Millisecond (1/1,000 of 1 second) |
U |
Micro |
Microsecond (1/1,000,000 of 1 second) |
M |
MonthEnd |
Last calendar day of month |
BM |
BusinessMonthEnd |
Last business day (weekday) of month |
MS |
MonthBegin |
First calendar day of month |
BMS |
BusinessMonthBegin |
First weekday of month |
W-MON, W-TUE, ... |
Week |
Weekly on given day of week (MON, TUE, WED, THU, FRI, SAT, or SUN) |
WOM-1MON, WOM-2MON, ... |
WeekOfMonth |
Generate weekly dates in the first, second, third, or fourth week of the month (e.g., WOM-3FRI for the third Friday of each month) |
Q-JAN, Q-FEB, ... |
QuarterEnd |
Quarterly dates anchored on last calendar day of each month, for year ending in indicated month (JAN, FEB, MAR, APR, MAY, JUN, JUL, AUG, SEP, OCT, NOV, or DEC) |
BQ-JAN, BQ-FEB, ... |
BusinessQuarterEnd |
Quarterly dates anchored on last weekday day of each month, for year ending in indicated month |
QS-JAN, QS-FEB, ... |
QuarterBegin |
Quarterly dates anchored on first calendar day of each month, for year ending in indicated month |
BQS-JAN, BQS-FEB, ... |
BusinessQuarterBegin |
Quarterly dates anchored on first weekday day of each month, for year ending in indicated month |
A-JAN, A-FEB, ... |
YearEnd |
Annual dates anchored on last calendar day of given month (JAN, FEB, MAR, APR, MAY, JUN, JUL, AUG, SEP, OCT, NOV, or DEC) |
BA-JAN, BA-FEB, ... |
BusinessYearEnd |
Annual dates anchored on last weekday of given month |
AS-JAN, AS-FEB, ... |
YearBegin |
Annual dates anchored on first day of given month |
BAS-JAN, BAS-FEB, ... |
BusinessYearBegin |
Annual dates anchored on first weekday of given month |
dates = [datetime(2011, 1, 2), datetime(2011, 1, 5), datetime(2011, 1, 7), datetime(2011, 1, 8), datetime(2011, 1, 10), datetime(2011, 1, 12)] ts = pd.Series(np.random.standard_normal(6), index=dates) stamp=ts.index[1]#获取索引 ts[stamp] #获取索引对应的值 #或者传入可以被解释为日期的字符串 ts['20110110'] ts['2011-01-10'] #较长的时间序列的切片获取 longer_ts = pd.Series(np.random.standard_normal(1000),index=pd.date_range("2000-01-01", periods=1000)) longer_ts['2001'] #通过传入年获取切片 longer_ts['2001-05']#通过传入年月获取切片 #通过日期进行切片的方式只对Series有效 ts[datetime(2011, 1, 8):] ts[datetime(2011, 1, 7):datetime(2011, 1, 10)] #也可用于不存在与该时间序列里的时间戳进行切片（即时间查询） ts.truncate(after="2011-01-09") #另一个切片方法 #以上对dataFrame同样适用 dates = pd.date_range("2000-01-01", periods=100, freq="W-WED") long_df = pd.DataFrame(np.random.standard_normal((100, 4)), index=dates,columns=["Colorado", "Texas","New York", "Ohio"]) long_df.loc["2001-05"]
带有重复索引的时间序列
dates = pd.DatetimeIndex(["2000-01-01", "2000-01-02", "2000-01-02","2000-01-02", "2000-01-03"]) dup_ts = pd.Series(np.arange(5), index=dates) dup_ts.index.is_unique grouped = dup_ts.groupby(level=0) grouped.mean() grouped.count()
日期的频率、范围、移动
resampler =ts.resample('D') #按日频率产生样本 #日期范围 index = pd.date_range("2012-04-01", "2012-06-01") pd.date_range(start="2012-04-01", periods=20) pd.date_range(end="2012-06-01", periods=20) #频率为每个月的月底 pd.date_range("2000-01-01", "2000-12-01", freq="BM") pd.date_range("2012-05-02 12:56:31", periods=5) #保留时间戳信息 pd.date_range("2012-05-02 12:56:31", periods=5, normalize=True)#不保留时间戳信息 #时间偏移量 from pandas.tseries.offsets import Hour, Minute pd.date_range("2000-01-01", "2000-01-03 23:59", freq="4H") #WOM-3FRI 每月第三个星期五 monthly_dates = pd.date_range("2012-01-01", "2012-09-01", freq="WOM-3FRI") #shift 沿着时间轴将数据前移或者后移 ts = pd.Series(np.random.standard_normal(4), index=pd.date_range("2000-01-01", periods=4, freq="M")) ts.shift(2) ts.shift(-2) ts.shift(2,freq='M')#偏移2月 ts.shift(3, freq="D")#偏移3天 ts.shift(1, freq="90T")#偏移1个半小时 from pandas.tseries.offsets import Day, MonthEnd now = datetime(2011, 11, 17) now + 3 * Day() #加3天 #Timestamp('2011-11-20 00:00:00') now + MonthEnd() # #日期滚动到月底 Timestamp('2011-11-30 00:00:00') now+MonthEnd(2) Timestamp('2011-12-31 00:00:00') #显示的将日期先前或者向后滚动 offset = MonthEnd() offset.rollforward(now) offset.rollback(now) #利用rollforward 、rollback巧妙分组 ts = pd.Series(np.random.standard_normal(20),index=pd.date_range("2000-01-15", periods=20, freq="4D")) ts.groupby(MonthEnd().rollforward).mean() ts.resample("M").mean()
时区和本地化
import pytz pytz.common_timezones[-5:] #获取时区名称列表 tz = pytz.timezone("Asia/Shanghai") #获取时区对象 #时间序列的时区操作 ts = pd.Series(np.random.standard_normal(len(dates)), index=dates) pd.date_range("2012-03-09 09:30", periods=10, tz="UTC")#生成日期范围时加时区 ts_utc = ts.tz_localize("UTC")#通过tz_localize转换为本地化时区 ts_utc.index ts_utc.tz_convert("America/New_York")#本地化到特定时区后可以用tz_convert 转换到别的时区 ts.index.tz_localize("Asia/Shanghai") #Timestamp时区操作 stamp = pd.Timestamp("2011-03-12 04:00") stamp_utc = stamp.tz_localize("utc") stamp_utc.tz_convert("Asia/Shanghai") stamp_Shanghai = pd.Timestamp("2011-03-12 04:00", tz="Asia/Shanghai")#生成时间时加时区 # Timestamp对象在内部保存了一个UTC时间戳（1970年1月1日算起的秒数），这个值在时区转换时不发生变化 stamp_utc.value stamp_utc.tz_convert("America/New_York").value
#通过数组创建时期
data = pd.read_csv("examples/macrodata.csv")
index = pd.PeriodIndex(year=data["year"], quarter=data["quarter"],freq="Q-DEC")
data.index=index
Period和算术运算
p = pd.Period("2022", freq="A-DEC")#表示2022年1月1日到2022年12月31日之间的整个时间段 p+5 periods = pd.period_range("2000-01-01", "2000-06-30", freq="M") #PeriodIndex(['2000-01', '2000-02', '2000-03', '2000-04', '2000-05', '2000-06'], dtype='period[M]') pd.Series(np.random.standard_normal(6), index=periods) #构造函数直接使用字符串 values = ["2001Q3", "2002Q2", "2003Q1"] index = pd.PeriodIndex(values, freq="Q-DEC") #时期的频率转换 p = pd.Period("2011", freq="A-DEC") p p.asfreq("M", how="start") p.asfreq("M", how="end") p.asfreq("M") p = pd.Period("2011", freq="A-JUN") p.asfreq("M", how="start") p.asfreq("M", how="end") p = pd.Period("Aug-2011", "M") #Period('2011-08', 'M') p.asfreq("A-JUN") #Period('2012', 'A-JUN') 2011-08属于2012年这个时期 #Series频率转换也类似 periods = pd.period_range("2006", "2009", freq="A-DEC") ts = pd.Series(np.random.standard_normal(len(periods)), index=periods) ts.asfreq("M", how="start") ts.asfreq("B", how="end") #按季度计算时期频率 p = pd.Period("2012Q4", freq="Q-JAN") p.asfreq("D", how="start") p.asfreq("D", how="end") #该季度倒数第二个工作日下午4点的时间戳 p4pm = (p.asfreq("B", how="end") - 1).asfreq("T", how="start") + 16 * 60 #季度型范围 periods = pd.period_range("2011Q3", "2012Q4", freq="Q-JAN") ts = pd.Series(np.arange(len(periods)), index=periods) #Timestamp和period互转 dates = pd.date_range("2000-01-01", periods=3, freq="M") ts = pd.Series(np.random.standard_normal(3), index=dates) pts = ts.to_period() dates = pd.date_range("2000-01-29", periods=6) ts2 = pd.Series(np.random.standard_normal(6), index=dates) ts2.to_period("M") pts = ts2.to_period() pts.to_timestamp(how="end")
3、采样
#重采样resample及频率转换 dates = pd.date_range("2000-01-01", periods=100) ts = pd.Series(np.random.standard_normal(len(dates)), index=dates) ts = pd.Series(np.random.standard_normal(len(dates)), index=dates) ts.resample("M").mean() ts.resample("M", kind="period").mean() #降采样 dates = pd.date_range("2000-01-01", periods=12, freq="T") ts = pd.Series(np.arange(len(dates)), index=dates) ts.resample("5min").sum() ts.resample("5min", closed="right").sum() ts.resample("5min", closed="right", label="right").sum() #OHLC重采样 ts = pd.Series(np.random.permutation(np.arange(len(dates))), index=dates) ts.resample("5min").ohlc() #升采样 frame = pd.DataFrame(np.random.standard_normal((2, 4)), index=pd.date_range("2000-01-01", periods=2,freq="W-WED"), columns=["Colorado", "Texas", "New York", "Ohio"]) df_daily = frame.resample("D").asfreq() frame.resample("D").ffill() frame.resample("D").ffill(limit=2) frame.resample("W-THU").ffill() #通过时期进行重采样 frame = pd.DataFrame(np.random.standard_normal((24, 4)), index=pd.period_range("1-2000", "12-2001",freq="M"), columns=["Colorado", "Texas", "New York", "Ohio"]) annual_frame = frame.resample("A-DEC").mean() annual_frame.resample("Q-DEC").ffill() annual_frame.resample("Q-DEC", convention="end").asfreq() annual_frame.resample("Q-MAR").ffill()
3、时间序列绘图
close_px_all = pd.read_csv("examples/stock_px.csv", parse_dates=True, index_col=0) close_px = close_px_all[["AAPL", "MSFT", "XOM"]] close_px["AAPL"].plot() #每日价格图表 close_px["AAPL"].rolling(250).mean().plot() #250日均值图表 #移动窗口函数 自动排除缺失值 plt.figure() std250 = close_px["AAPL"].pct_change().rolling(250, min_periods=10).std() std250.plot() expanding_mean = std250.expanding().mean() close_px.rolling(60).mean().plot(logy=True) close_px.rolling("20D").mean()
