如何使用统计模型循环自递归递归过滤器以生成多个列

Question

您好，我正在尝试在整个数据帧中包含 statmodel 的自回归过滤器函数以生成新列。

但是我在运行脚本时遇到错误，如果我只在单列上计算它，就不会发生这种情况：

类型错误：无法将序列乘以类型为"float"的非 int

import statsmodels.api as sm
import statsmodels.tsa as tsa
import statsmodels.formula.api as smf
import pandas as pd

#generate dataset :

data = pd.DataFrame({'effect1': [], 'effect2': [], 'effect3': []})
for i in range(10):
data = data.append({'effect1': i, 'effect2': i*2, 'effect3': i*3}, ignore_index=True)

#check data type    
data = data.astype('int64')

#compute autoregressive recursive filter
carry_over = (.1,.2)
rolling_df = (tsa.filters.filtertools.recursive_filter(col, i)
.rename({col: '{0}_{1:d}'.format(col, i)       
for col in data.columns}, axis=1) 
for i in carry_over)                                
data = pd.concat((data, *rolling_df), axis=1)   

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-42-74a899bff628> in <module>
10                 for i in carry_over)                                
11 
---> 12 data = pd.concat((data, *rolling_df), axis=1)
<ipython-input-42-74a899bff628> in <genexpr>(.0)
8                 .rename({col: '{0}_{1:d}'.format(col, i)       
9                                for col in data.columns}, axis=1) 
---> 10                 for i in carry_over)                                
11 
12 data = pd.concat((data, *rolling_df), axis=1)
~AppDataLocalContinuumanaconda3envsregressorlibsite-packagesstatsmodelstsafiltersfiltertools.py in recursive_filter(x, ar_coeff, init)
208         zi = None
209 
--> 210     y = signal.lfilter([1.], np.r_[1, -ar_coeff], x, zi=zi)
211 
212     if init is not None:
~AppDataLocalContinuumanaconda3envsregressorlibsite-packagesscipysignalsignaltools.py in lfilter(b, a, x, axis, zi)
1395     else:
1396         if zi is None:
-> 1397             return sigtools._linear_filter(b, a, x, axis)
1398         else:
1399             return sigtools._linear_filter(b, a, x, axis, zi)
TypeError: can't multiply sequence by non-int of type 'float'

Answer 1

以下是如何计算rolling_df的修改：

# Optional: bind the function to a shorter name. You could also 
# import this as something terse, or just leave it as is
f = tsa.filters.filtertools.recursive_filter
rolling_df = [data.apply(f, ar_coeff=i)
.add_suffix(f'_{i}') 
for i in carry_over]
data = pd.concat((data, *rolling_df), axis=1)
data
effect1  effect2  effect3  effect1_0.1  effect2_0.1  effect3_0.1  effect1_0.2  effect2_0.2  effect3_0.2
0        0        0        0     0.000000     0.000000     0.000000     0.000000     0.000000     0.000000
1        1        2        3     1.000000     2.000000     3.000000     1.000000     2.000000     3.000000
2        2        4        6     2.100000     4.200000     6.300000     2.200000     4.400000     6.600000
3        3        6        9     3.210000     6.420000     9.630000     3.440000     6.880000    10.320000
4        4        8       12     4.321000     8.642000    12.963000     4.688000     9.376000    14.064000
5        5       10       15     5.432100    10.864200    16.296300     5.937600    11.875200    17.812800
6        6       12       18     6.543210    13.086420    19.629630     7.187520    14.375040    21.562560
7        7       14       21     7.654321    15.308642    22.962963     8.437504    16.875008    25.312512
8        8       16       24     8.765432    17.530864    26.296296     9.687501    19.375002    29.062502
9        9       18       27     9.876543    19.753086    29.629630    10.937500    21.875000    32.812500