seasonal_decompose 引发错误：TypeError：给定的 periodIndex。检查 `freq` 属性而不是使用 infer_freq

Question

我正在尝试对常用的航空公司乘客数据集运行基本的seasonal_decompose，该数据集以这些行开头：

Month
1949-02    4.770685
1949-03    4.882802
1949-04    4.859812
1949-05    4.795791
1949-06    4.905275
1949-07    4.997212
1949-08    4.997212
1949-09    4.912655
1949-10    4.779123
1949-11    4.644391
1949-12    4.770685
1950-01    4.744932
1950-02    4.836282
1950-03    4.948760
1950-04    4.905275
1950-05    4.828314
1950-06    5.003946
1950-07    5.135798
1950-08    5.135798
Freq: M, Name: Passengers, dtype: float64

我的索引类型是：

pandas.tseries.period.PeriodIndex

我尝试运行一些非常简单的代码：

from statsmodels.tsa.seasonal import seasonal_decompose
log_passengers.interpolate(inplace = True)
decomposition = seasonal_decompose(log_passengers)

这是错误的完整输出：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-113-bf122d457673> in <module>()
      1 from statsmodels.tsa.seasonal import seasonal_decompose
      2 log_passengers.interpolate(inplace = True)
----> 3 decomposition = seasonal_decompose(log_passengers)

/Users/ann/anaconda/lib/python3.5/site-packages/statsmodels/tsa/seasonal.py in seasonal_decompose(x, model, filt, freq)
     56     statsmodels.tsa.filters.convolution_filter
     57     """
---> 58     _pandas_wrapper, pfreq = _maybe_get_pandas_wrapper_freq(x)
     59     x = np.asanyarray(x).squeeze()
     60     nobs = len(x)

/Users/ann/anaconda/lib/python3.5/site-packages/statsmodels/tsa/filters/_utils.py in _maybe_get_pandas_wrapper_freq(X, trim)
     44         index = X.index
     45         func = _get_pandas_wrapper(X, trim)
---> 46         freq = index.inferred_freq
     47         return func, freq
     48     else:

pandas/src/properties.pyx in pandas.lib.cache_readonly.__get__ (pandas/lib.c:44097)()

/Users/ann/anaconda/lib/python3.5/site-packages/pandas/tseries/base.py in inferred_freq(self)
    233         """
    234         try:
--> 235             return frequencies.infer_freq(self)
    236         except ValueError:
    237             return None

/Users/ann/anaconda/lib/python3.5/site-packages/pandas/tseries/frequencies.py in infer_freq(index, warn)
    854 
    855     if com.is_period_arraylike(index):
--> 856         raise TypeError("PeriodIndex given. Check the `freq` attribute "
    857                         "instead of using infer_freq.")
    858     elif isinstance(index, pd.TimedeltaIndex):

TypeError: PeriodIndex given. Check the `freq` attribute instead of using infer_freq.

这是我尝试过的：

• 使用decomposition = seasonal_decompose(log_passengers, infer_freq = True)会产生错误： TypeError: seasonal_decompose() got an unexpected keyword argument 'infer_freq'
• 使用decomposition = seasonal_decompose(log_passengers, freq = 'M')会导致错误：TypeError: PeriodIndex given. Check thefreqattribute instead of using infer_freq.
• 我还验证了周期索引索引中的每个周期索引与代码列表具有相同的频率：set([x.freq for x in log_passengers.index])这确实产生了一组仅一个频率：{<MonthEnd>}

我在各种 Github 问题上看到了一些关于此问题的讨论（https://github.com/pydata/pandas/issues/6771），但所讨论的内容似乎都没有帮助。关于如何解决此问题或我在这个简单的 seasona_decompose 中做错了什么有什么建议吗？

Answer 1

seasonal_decompose 不接受 periodIndex，解决方法是使用 to_timestamp 方法将索引转换为 DatetimeIndex：

from statsmodels.tsa.seasonal import seasonal_decompose
log_passengers.interpolate(inplace = True)
log_passengers.index=log_passengers.index.to_timestamp()
decomposition = seasonal_decompose(log_passengers)