我创建了一个开源 Python 包,名为时间分解 https://github.com/jstephenj14/timedisagg它基于 R tempdisagg 包。该包实现了基本的 Chow-Lin 和 Litterman 方法。它还允许像 R 包一样进行基本平均、总和、第一个和最后一个转换选择。
给定 R 中的以下函数调用来分解sales.a
作为一个函数exports.q
:
model <- td(sales.a ~ 0 + exports.q,method="chow-lin-maxlog",conversion="sum")
可以使用 timedisagg 进行类似的调用,如下所示:
from timedisagg.td import TempDisagg
td_obj = TempDisagg(conversion="sum", method="chow-lin-maxlog")
final_disaggregated_output = td_obj(expected_dataset)
哪里的expected_dataset
是一个 pandas 数据框,格式如下:
index grain X y
0 1972 1 1432.63900 NaN
1 1972 2 1456.89100 NaN
2 1972 3 1342.56200 NaN
3 1972 4 1539.39400 NaN
4 1973 1 1535.75400 NaN
5 1973 2 1578.45800 NaN
6 1973 3 1574.72400 NaN
7 1973 4 1652.17100 NaN
8 1974 1 2047.83400 NaN
9 1974 2 2117.97100 NaN
10 1974 3 1925.92600 NaN
11 1974 4 1798.19000 NaN
12 1975 1 1818.81700 136.702329
13 1975 2 1808.22500 136.702329
14 1975 3 1649.20600 136.702329
15 1975 4 1799.66500 136.702329
16 1976 1 1985.75300 151.056074
17 1976 2 2064.66300 151.056074
18 1976 3 1856.38700 151.056074
19 1976 4 1919.08700 151.056074
.. ... ... ... ...
152 2010 1 19915.79514 988.309676
153 2010 2 19482.48000 988.309676
154 2010 3 18484.64900 988.309676
155 2010 4 18026.46869 988.309676
156 2011 1 19687.52100 NaN
157 2011 2 18913.06608 NaN
这里 X 是exports.q
y 是sales.a
.
输出final_disaggregated_output
将出现如下所示y_hat
是分类销售额:
index grain X y y_hat
0 1972 1 1432.639 NaN 21.656879
1 1972 2 1456.891 NaN 22.219737
2 1972 3 1342.562 NaN 20.855413
3 1972 4 1539.394 NaN 23.937916
4 1973 1 1535.754 NaN 24.229008
编辑 - 如果有人需要帮助将他们的数据处理到我的包中,请随时在git https://github.com/jstephenj14/timedisagg对于包裹。