import adata
import numpy as np
import pandas as pd
# k_type: k线类型：1.日；2.周；3.月 默认：1 日k
df_stock_byday = pd.concat([adata.stock.market.get_market(stock_code='002241', k_type=1, start_date='2021-01-01').assign(stock_code='002241'),adata.stock.market.get_market(stock_code='000001', k_type=1, start_date='2021-01-01').assign(stock_code='000001')])
df_stock_byday[['open', 'close', 'volume', 'high', 'low']] = df_stock_byday[['open', 'close', 'volume', 'high', 'low']].astype(np.float64)import pytimetk as tk
df_stock_byday.glimpse()<class 'pandas.core.frame.DataFrame'>: 1728 rows of 13 columns
trade_time: object ['2021-01-04 00:00:00', '2021-01-05 0 ...
trade_date: object ['2021-01-04', '2021-01-05', '2021-01 ...
open: float64 [36.8, 35.94, 37.8, 37.29, 38.9, 38.6 ...
close: float64 [36.19, 37.93, 37.13, 38.45, 38.69, 4 ...
high: float64 [36.8, 38.35, 38.2, 38.95, 39.3, 41.1 ...
low: float64 [35.85, 35.64, 36.85, 37.05, 37.73, 3 ...
volume: float64 [1285439.0, 1675280.0, 953221.0, 1365 ...
amount: float64 [4707993856.0, 6236502016.0, 36116663 ...
change_pct: float64 [-1.58, 4.81, -2.11, 3.56, 0.62, 5.58 ...
change: float64 [-0.58, 1.74, -0.8, 1.32, 0.24, 2.16, ...
turnover_ratio: float64 [4.61, 6.01, 3.42, 4.89, 3.62, 6.23, ...
pre_close: float64 [36.769999999999996, 36.19, 37.93, 37 ...
stock_code: object ['002241', '002241', '002241', '00224 ...df_stock_byday['trade_date'] = pd.to_datetime(df_stock_byday['trade_date'], format='%Y-%m-%d')
df_stock_byday['trade_time'] = pd.to_datetime(df_stock_byday['trade_time'], infer_datetime_format=True)Pandas 提供了多种频率字符串（也称为偏移别名）来定义时间序列的频率。以下是 Pandas 中使用的一些常见频率字符串：
‘B’：工作日
‘D’：日历日
‘W’：每周
‘M’：月末
‘BM’：营业月末
‘MS’：月份开始
‘BMS’：营业月份开始
‘Q’：季度末
‘BQ’：业务季度结束
‘QS’：季度开始
‘BQS’：业务季度开始
‘A’ 或 ‘Y’：年末
“BA” 或 “BY”：业务年度结束
‘AS’ 或 ‘YS’：年份开始
‘BAS’ 或 ‘BYS’：营业年度开始
‘H’：每小时
‘T’ 或 ‘min’：每分钟
‘S’：其次
‘L’ 或 ‘ms’：毫秒
‘U’：微秒
‘N’：纳秒
您还可以通过组合基本频率来创建自定义频率，例如：
‘2D’：每 2 天
‘3W’：每 3 周
‘4H’：每 4 小时
‘1H30T’：每 1 小时 30 分钟
您可以将多个频率相加在一起。
‘1D1H’：1 天 1 小时
‘1H30T’：1 小时 30 分钟
date_range_two_days = pd.date_range(start='2023-01-01', end='2023-01-10', freq='1H30T')
date_range_two_daysDatetimeIndex(['2023-01-01 00:00:00', '2023-01-01 01:30:00',
'2023-01-01 03:00:00', '2023-01-01 04:30:00',
'2023-01-01 06:00:00', '2023-01-01 07:30:00',
'2023-01-01 09:00:00', '2023-01-01 10:30:00',
'2023-01-01 12:00:00', '2023-01-01 13:30:00',
...
'2023-01-09 10:30:00', '2023-01-09 12:00:00',
'2023-01-09 13:30:00', '2023-01-09 15:00:00',
'2023-01-09 16:30:00', '2023-01-09 18:00:00',
'2023-01-09 19:30:00', '2023-01-09 21:00:00',
'2023-01-09 22:30:00', '2023-01-10 00:00:00'],
dtype='datetime64[ns]', length=145, freq='90min')df_stock_byday = df_stock_byday.pad_by_time('trade_date', freq = 'B')
df_stock_byday.index.freq = 'B'
df_stock_byday.set_index(['trade_date'],inplace = True)
df_stock_byday.tail()| trade_time | open | close | high | low | volume | amount | change_pct | change | turnover_ratio | pre_close | stock_code | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| trade_date | ||||||||||||
| 2024-07-24 | 2024-07-24 | 22.17 | 21.76 | 22.49 | 21.58 | 804468.0 | 1.763040e+09 | -1.54 | -0.34 | 2.67 | 22.10 | 002241 |
| 2024-07-25 | 2024-07-25 | 10.12 | 10.09 | 10.15 | 10.00 | 856404.0 | 8.617504e+08 | -0.39 | -0.04 | 0.44 | 10.13 | 000001 |
| 2024-07-25 | 2024-07-25 | 21.96 | 21.50 | 22.05 | 21.06 | 846324.0 | 1.812192e+09 | -1.19 | -0.26 | 2.81 | 21.76 | 002241 |
| 2024-07-26 | 2024-07-26 | 21.54 | 21.62 | 21.93 | 21.24 | 729641.0 | 1.572982e+09 | 0.56 | 0.12 | 2.42 | 21.50 | 002241 |
| 2024-07-26 | 2024-07-26 | 10.10 | 10.03 | 10.12 | 9.97 | 929552.0 | 9.318795e+08 | -0.59 | -0.06 | 0.48 | 10.09 | 000001 |
长表模式
summary_stock_code_df = df_stock_byday \
.groupby("stock_code") \
.summarize_by_time(
date_column = 'trade_time',
value_column = 'close',
freq = "MS",
agg_func = ['mean', 'median', 'min', 'max'],
wide_format = False
)
summary_stock_code_df.head()| stock_code | trade_time | close_mean | close_median | close_min | close_max | |
|---|---|---|---|---|---|---|
| 0 | 000001 | 2021-01-01 | 19.837000 | 20.105 | 16.76 | 21.68 |
| 1 | 000001 | 2021-02-01 | 22.163333 | 22.440 | 19.97 | 23.54 |
| 2 | 000001 | 2021-03-01 | 20.079565 | 20.080 | 18.99 | 21.60 |
| 3 | 000001 | 2021-04-01 | 20.516667 | 20.270 | 18.85 | 22.18 |
| 4 | 000001 | 2021-05-01 | 22.591667 | 22.410 | 21.66 | 23.78 |
宽表模式
df_stock_byday \
.groupby("stock_code") \
.summarize_by_time(
date_column = 'trade_time',
value_column = 'close',
freq = "MS",
agg_func = 'mean',
wide_format = True
). \
head()| trade_time | close_000001 | close_002241 | |
|---|---|---|---|
| 0 | 2021-01-01 | 19.837000 | 39.413500 |
| 1 | 2021-02-01 | 22.163333 | 32.794000 |
| 2 | 2021-03-01 | 20.079565 | 28.499565 |
| 3 | 2021-04-01 | 20.516667 | 31.700952 |
| 4 | 2021-05-01 | 22.591667 | 36.726111 |
df_stock_byday_with_sig = df_stock_byday.augment_timeseries_signature(date_column = 'trade_time')
df_stock_byday_with_sig.head()| trade_time | open | close | high | low | volume | amount | change_pct | change | turnover_ratio | ... | trade_time_mday | trade_time_qday | trade_time_yday | trade_time_weekend | trade_time_hour | trade_time_minute | trade_time_second | trade_time_msecond | trade_time_nsecond | trade_time_am_pm | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| trade_date | |||||||||||||||||||||
| 2021-01-04 | 2021-01-04 | 36.80 | 36.19 | 36.80 | 35.85 | 1285439.0 | 4.707994e+09 | -1.58 | -0.58 | 4.61 | ... | 4.0 | 4.0 | 4.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
| 2021-01-04 | 2021-01-04 | 17.69 | 17.19 | 17.69 | 17.03 | 1554216.0 | 2.891682e+09 | -4.13 | -0.74 | 0.80 | ... | 4.0 | 4.0 | 4.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
| 2021-01-05 | 2021-01-05 | 35.94 | 37.93 | 38.35 | 35.64 | 1675280.0 | 6.236502e+09 | 4.81 | 1.74 | 6.01 | ... | 5.0 | 5.0 | 5.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
| 2021-01-05 | 2021-01-05 | 16.99 | 16.76 | 17.07 | 16.39 | 1821352.0 | 3.284607e+09 | -2.50 | -0.43 | 0.94 | ... | 5.0 | 5.0 | 5.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
| 2021-01-06 | 2021-01-06 | 37.80 | 37.13 | 38.20 | 36.85 | 953221.0 | 3.611666e+09 | -2.11 | -0.80 | 3.42 | ... | 6.0 | 6.0 | 6.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
5 rows × 41 columns
df_stock_byday \
.groupby('stock_code') \
.augment_lags(date_column = 'trade_time',value_column = 'close',lags = (1, 7)) \
.head()| trade_time | open | close | high | low | volume | amount | change_pct | change | turnover_ratio | pre_close | stock_code | close_lag_1 | close_lag_2 | close_lag_3 | close_lag_4 | close_lag_5 | close_lag_6 | close_lag_7 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| trade_date | |||||||||||||||||||
| 2021-01-04 | 2021-01-04 | 36.80 | 36.19 | 36.80 | 35.85 | 1285439.0 | 4.707994e+09 | -1.58 | -0.58 | 4.61 | 36.77 | 002241 | NaN | NaN | NaN | NaN | NaN | NaN | NaN |
| 2021-01-04 | 2021-01-04 | 17.69 | 17.19 | 17.69 | 17.03 | 1554216.0 | 2.891682e+09 | -4.13 | -0.74 | 0.80 | 17.93 | 000001 | NaN | NaN | NaN | NaN | NaN | NaN | NaN |
| 2021-01-05 | 2021-01-05 | 35.94 | 37.93 | 38.35 | 35.64 | 1675280.0 | 6.236502e+09 | 4.81 | 1.74 | 6.01 | 36.19 | 002241 | 36.19 | NaN | NaN | NaN | NaN | NaN | NaN |
| 2021-01-05 | 2021-01-05 | 16.99 | 16.76 | 17.07 | 16.39 | 1821352.0 | 3.284607e+09 | -2.50 | -0.43 | 0.94 | 17.19 | 000001 | 17.19 | NaN | NaN | NaN | NaN | NaN | NaN |
| 2021-01-06 | 2021-01-06 | 37.80 | 37.13 | 38.20 | 36.85 | 953221.0 | 3.611666e+09 | -2.11 | -0.80 | 3.42 | 37.93 | 002241 | 37.93 | 36.19 | NaN | NaN | NaN | NaN | NaN |
df_stock_byday \
.groupby('stock_code') \
.augment_rolling(
date_column = 'trade_time',
value_column = 'close',
window = [2,7],
window_func = ['mean', ('std', lambda x: x.std())]
)| trade_time | open | close | high | low | volume | amount | change_pct | change | turnover_ratio | pre_close | stock_code | close_rolling_mean_win_2 | close_rolling_std_win_2 | close_rolling_mean_win_7 | close_rolling_std_win_7 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| trade_date | ||||||||||||||||
| 2021-01-04 | 2021-01-04 | 17.69 | 17.19 | 17.69 | 17.03 | 1554216.0 | 2.891682e+09 | -4.13 | -0.74 | 0.80 | 17.93 | 000001 | NaN | NaN | NaN | NaN |
| 2021-01-04 | 2021-01-04 | 36.80 | 36.19 | 36.80 | 35.85 | 1285439.0 | 4.707994e+09 | -1.58 | -0.58 | 4.61 | 36.77 | 002241 | NaN | NaN | NaN | NaN |
| 2021-01-05 | 2021-01-05 | 16.99 | 16.76 | 17.07 | 16.39 | 1821352.0 | 3.284607e+09 | -2.50 | -0.43 | 0.94 | 17.19 | 000001 | 16.975 | 0.215 | 16.975000 | 0.215000 |
| 2021-01-05 | 2021-01-05 | 35.94 | 37.93 | 38.35 | 35.64 | 1675280.0 | 6.236502e+09 | 4.81 | 1.74 | 6.01 | 36.19 | 002241 | 37.060 | 0.870 | 37.060000 | 0.870000 |
| 2021-01-06 | 2021-01-06 | 16.67 | 18.15 | 18.15 | 16.59 | 1934945.0 | 3.648522e+09 | 8.29 | 1.39 | 1.00 | 16.76 | 000001 | 17.455 | 0.695 | 17.366667 | 0.581053 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 2024-07-24 | 2024-07-24 | 10.18 | 10.13 | 10.24 | 10.12 | 724518.0 | 7.363328e+08 | -0.49 | -0.05 | 0.37 | 10.18 | 000001 | 10.155 | 0.025 | 10.288571 | 0.103016 |
| 2024-07-25 | 2024-07-25 | 21.96 | 21.50 | 22.05 | 21.06 | 846324.0 | 1.812192e+09 | -1.19 | -0.26 | 2.81 | 21.76 | 002241 | 21.630 | 0.130 | 22.440000 | 0.597351 |
| 2024-07-25 | 2024-07-25 | 10.12 | 10.09 | 10.15 | 10.00 | 856404.0 | 8.617504e+08 | -0.39 | -0.04 | 0.44 | 10.13 | 000001 | 10.110 | 0.020 | 10.258571 | 0.123800 |
| 2024-07-26 | 2024-07-26 | 10.10 | 10.03 | 10.12 | 9.97 | 929552.0 | 9.318795e+08 | -0.59 | -0.06 | 0.48 | 10.09 | 000001 | 10.060 | 0.030 | 10.205714 | 0.130915 |
| 2024-07-26 | 2024-07-26 | 21.54 | 21.62 | 21.93 | 21.24 | 729641.0 | 1.572982e+09 | 0.56 | 0.12 | 2.42 | 21.50 | 002241 | 21.560 | 0.060 | 22.231429 | 0.592342 |
1728 rows × 16 columns
基于数据框
df_stock_byday \
.groupby('stock_code') \
.future_frame('trade_time', length_out = 365) \
.augment_timeseries_signature('trade_time') \
.query("close.isna()") \
.tail() | trade_time | open | close | high | low | volume | amount | change_pct | change | turnover_ratio | ... | trade_time_mday | trade_time_qday | trade_time_yday | trade_time_weekend | trade_time_hour | trade_time_minute | trade_time_second | trade_time_msecond | trade_time_nsecond | trade_time_am_pm | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2519 | 2025-07-22 | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | ... | 22.0 | 22.0 | 203.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
| 2520 | 2025-07-23 | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | ... | 23.0 | 23.0 | 204.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
| 2521 | 2025-07-24 | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | ... | 24.0 | 24.0 | 205.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
| 2522 | 2025-07-25 | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | ... | 25.0 | 25.0 | 206.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
| 2523 | 2025-07-26 | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | NaN | ... | 26.0 | 26.0 | 207.0 | 0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | am |
5 rows × 41 columns
基于Series
pd.Series(pd.date_range("2023", "2024", freq = "D")) \
.make_future_timeseries(12) \
.get_timeseries_signature()| idx | idx_index_num | idx_year | idx_year_iso | idx_yearstart | idx_yearend | idx_leapyear | idx_half | idx_quarter | idx_quarteryear | ... | idx_mday | idx_qday | idx_yday | idx_weekend | idx_hour | idx_minute | idx_second | idx_msecond | idx_nsecond | idx_am_pm | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 2024-01-02 | 1704153600 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 2 | 2 | 2 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 1 | 2024-01-03 | 1704240000 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 3 | 3 | 3 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 2 | 2024-01-04 | 1704326400 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 4 | 4 | 4 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 3 | 2024-01-05 | 1704412800 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 5 | 5 | 5 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 4 | 2024-01-06 | 1704499200 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 6 | 6 | 6 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 5 | 2024-01-07 | 1704585600 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 7 | 7 | 7 | 1 | 0 | 0 | 0 | 0 | 0 | am |
| 6 | 2024-01-08 | 1704672000 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 8 | 8 | 8 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 7 | 2024-01-09 | 1704758400 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 9 | 9 | 9 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 8 | 2024-01-10 | 1704844800 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 10 | 10 | 10 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 9 | 2024-01-11 | 1704931200 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 11 | 11 | 11 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 10 | 2024-01-12 | 1705017600 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 12 | 12 | 12 | 0 | 0 | 0 | 0 | 0 | 0 | am |
| 11 | 2024-01-13 | 1705104000 | 2024 | 2024 | 0 | 0 | 1 | 1 | 1 | 2024Q1 | ... | 13 | 13 | 13 | 0 | 0 | 0 | 0 | 0 | 0 | am |
12 rows × 30 columns
绘制折线图观察每日收盘价趋势
df_stock_byday['year'] = pd.to_datetime(df_stock_byday['trade_time']).dt.year
df_stock_byday. \
reset_index(). \
dropna(). \
groupby("stock_code"). \
plot_timeseries(date_column = 'trade_time',
facet_ncol = 2,
color_column = 'year',
facet_scales = "free",
value_column = 'close')针对股票时序数据可以绘制k线图
from datetime import datetime
import vectorbt as vbt
df_stock_byday = adata.stock.market.get_market(stock_code='002241', k_type=1, start_date='2021-01-01').assign(stock_code='002241')
df_stock_byday[['open', 'close', 'volume', 'high', 'low']] = df_stock_byday[['open', 'close', 'volume', 'high', 'low']].astype(np.float64)
df_stock_byday['trade_date'] = pd.to_datetime(df_stock_byday['trade_date'], format='%Y-%m-%d')
df_stock_byday['trade_time'] = pd.to_datetime(df_stock_byday['trade_time'], infer_datetime_format=True)
plot_Candlestick = df_stock_byday.vbt.ohlcv.plot(plot_type='Candlestick')
plot_Candlestick.update_layout(height=None,width=None)
plot_Candlestick.show()df_stock_byday = df_stock_byday.loc[df_stock_byday.stock_code == '002241']
# df_stock_byday = df_stock_byday.pad_by_time('trade_time', freq = 'B') # 按工作日重采样，缺失的日期的值用NA填补
# df_stock_byday.set_index(['trade_time'],inplace = True)
# df_stock_byday.index.freq = 'B'
anomalize_df = tk.anomalize(
data = df_stock_byday,
date_column = 'trade_time',
value_column = 'close',
period = 7,
iqr_alpha = 0.05, # using the default
clean_alpha = 0.75, # using the default
clean = "min_max"
)
anomalize_df.glimpse()<class 'pandas.core.frame.DataFrame'>: 864 rows of 12 columns
trade_time: datetime64[ns] [Timestamp('2021-01-04 00:00:00'), ...
observed: float64 [36.19, 37.93, 37.13, 38.45, 38.69 ...
seasonal: float64 [4.797307111756358, -2.89055462491 ...
seasadj: float64 [31.39269288824364, 40.82055462491 ...
trend: float64 [41.13074580232775, 40.88098513275 ...
remainder: float64 [-9.738052914084111, -0.0604305078 ...
anomaly: object ['Yes', 'No', 'No', 'No', 'No', 'N ...
anomaly_score: float64 [10.492364565686895, 0.81474215944 ...
anomaly_direction: int32 [-1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, ...
recomposed_l1: float64 [42.16966810551952, 34.23204569928 ...
recomposed_l2: float64 [51.19506102585427, 43.25743861961 ...
observed_clean: float64 [43.297842220561364, 37.93, 37.13, ...绘制异常值
# Plot anomalies
tk.plot_anomalies(
data = anomalize_df,
date_column = 'trade_time',
engine = 'plotly',
title = '异常值'
)清理异常值后
tk.plot_anomalies_cleaned(
data = anomalize_df,
date_column = 'trade_time',
engine = 'plotly',
title = '清理异常值后'
)时间序列分解是一种将时间序列分解为多个组成部分的统计方法，每个组成部分代表模式的基本类别之一。这些组成部分通常包括：
趋势 (T)：长期上升或下降的趋势
季节性 (S)：在一年或更短的时间内重复出现的周期性波动
周期性 (C)：比季节性更长的周期性波动
不规则性 (I)：随机的、不可预测的变化
tk.plot_anomalies_decomp(
data = anomalize_df,
date_column = 'trade_time',
engine = 'plotly',
title = '时间序列分解'
)| 单变量时间序列模型 | 多变量时间序列模型 |
|---|---|
| 只使用一个变量 | 使用多个变量 |
| 无法使用外部数据 | 可以使用外部数据 |
| 仅基于过去和现在之间的关系 | 基于过去和现在之间的关系，以及变量之间的关系 |
| 预测未来某个时间点该变量的值 | 预测未来某个时间点一个或多个变量的值 |
import pandas as pd
import numpy as np
import pytimetk as tk
from sklearn.ensemble import RandomForestRegressor
df_stock_byday.glimpse()
dset = tk.load_dataset('walmart_sales_weekly', parse_dates = ['Date'])
dset = dset.drop(columns=[
'id', # This column can be removed as it is equivalent to 'Dept'
'Store', # This column has only one possible value
'Type', # This column has only one possible value
'Size', # This column has only one possible value
'MarkDown1', 'MarkDown2', 'MarkDown3', 'MarkDown4', 'MarkDown5',
'IsHoliday', 'Temperature', 'Fuel_Price', 'CPI',
'Unemployment'])
dset.head()
sales_df = dset
sales_df_with_futureframe = sales_df \
.groupby('Dept') \
.future_frame(
date_column = 'Date',
length_out = 5
)
sales_df_dates = sales_df_with_futureframe.augment_timeseries_signature(date_column = 'Date')
sales_df_dates.head(10)
df_with_lags = sales_df_dates \
.groupby('Dept') \
.augment_lags(
date_column = 'Date',
value_column = 'Weekly_Sales',
lags = [5,6,7,8,9]
)
lag_columns = [col for col in df_with_lags.columns if 'lag' in col]
df_with_rolling = df_with_lags \
.groupby('Dept') \
.augment_rolling(
date_column = 'Date',
value_column = lag_columns,
window = 4,
window_func = 'mean',
threads = 1 # Change to -1 to use all available cores
)
df_with_rolling[df_with_rolling.Dept ==1].head(10)
df_with_lags.head(5)
all_lag_columns = [col for col in df_with_rolling.columns if 'lag' in col]
df_no_nas = df_with_rolling \
.dropna(subset=all_lag_columns, inplace=False)
df_no_nas.head()
future = df_no_nas[df_no_nas.Weekly_Sales.isnull()]
train = df_no_nas[df_no_nas.Weekly_Sales.notnull()]
train_columns = [
'Dept'
, 'Date_year'
, 'Date_month'
, 'Date_yweek'
, 'Date_mweek'
, 'Weekly_Sales_lag_5'
, 'Weekly_Sales_lag_6'
, 'Weekly_Sales_lag_7'
, 'Weekly_Sales_lag_8'
, 'Weekly_Sales_lag_5_rolling_mean_win_4'
, 'Weekly_Sales_lag_6_rolling_mean_win_4'
, 'Weekly_Sales_lag_7_rolling_mean_win_4'
, 'Weekly_Sales_lag_8_rolling_mean_win_4'
]
X = train[train_columns]
y = train[['Weekly_Sales']]
model = RandomForestRegressor(random_state=123)
model = model.fit(X, y)
predicted_values = model.predict(future[train_columns])
future['y_pred'] = predicted_values
future.head(10)
train['type'] = 'actuals'
future['type'] = 'prediction'
full_df = pd.concat([train, future])
full_df.head(10)
full_df['Weekly_Sales'] = np.where(full_df.type =='actuals', full_df.Weekly_Sales, full_df.y_pred)
full_df \
.groupby('Dept') \
.plot_timeseries(
date_column = 'Date',
value_column = 'Weekly_Sales',
color_column = 'type',
smooth = False,
smooth_alpha = 0,
facet_ncol = 2,
facet_scales = "free",
y_intercept_color = tk.palette_timetk()['steel_blue'],
width = 800,
height = 600,
engine = 'plotly'
)<class 'pandas.core.frame.DataFrame'>: 864 rows of 13 columns
trade_time: datetime64[ns] [Timestamp('2021-01-04 00:00:00'), Ti ...
trade_date: datetime64[ns] [Timestamp('2021-01-04 00:00:00'), Ti ...
open: float64 [36.8, 35.94, 37.8, 37.29, 38.9, 38.6 ...
close: float64 [36.19, 37.93, 37.13, 38.45, 38.69, 4 ...
high: float64 [36.8, 38.35, 38.2, 38.95, 39.3, 41.1 ...
low: float64 [35.85, 35.64, 36.85, 37.05, 37.73, 3 ...
volume: float64 [1285439.0, 1675280.0, 953221.0, 1365 ...
amount: float64 [4707993856.0, 6236502016.0, 36116663 ...
change_pct: float64 [-1.58, 4.81, -2.11, 3.56, 0.62, 5.58 ...
change: float64 [-0.58, 1.74, -0.8, 1.32, 0.24, 2.16, ...
turnover_ratio: float64 [4.61, 6.01, 3.42, 4.89, 3.62, 6.23, ...
pre_close: float64 [36.769999999999996, 36.19, 37.93, 37 ...
stock_code: object ['002241', '002241', '002241', '00224 ...# !pip install git+https://github.com/business-science/pymodeltime.git
#
# !pip install autogluon
#
#
# !pip install h2o