嘿，我正在对几个月内迭代的一些横截面数据进行多个OLS回归。我在第四个月遇到了一个ValueError：零大小的数组到没有标识的最大缩减操作。但我不知道为什么。数据没有NaN，我已经用dropna测试了这一点，因为它是在另一个问题链接中建议的：
df_month.dropna(how='all')
每个月的数据为零是由于数据的标准化，这在前三次迭代中不会造成任何麻烦，只有在第四次迭代中。同样奇怪的是，如果我在第四行执行OLS回归之前停止循环，然后在另一个单元格中手动再次运行回归，它就会正常工作。这个问题可能是由于存储造成的吗？
这是我的代码一个数据复制错误：
import statsmodels.formula.api as smf #ols (minor letters)
import pandas as pd
import numpy as np
df = pd.read_csv('df_all.csv', index_col='Instrument', sep=',', decimal='.')
df.drop(columns='Unnamed: 0', inplace=True)
#Creates an array with all the years to iterate through single year
years = df_all['Date'].dt.year.unique()
df_store = pd.DataFrame(index=[], columns=['year', 'month', 'R2_adj'])
for year in years:
df_year = df_all[df_all['Date'].dt.year == year]
df_year_t1 = df_all[df_all['Date'].dt.year == year+1]
Jan_date = df_year['Date'][0]
year_start = df_year[(df_year['Date'] == Jan_date) & (df_year['HQ'] == 'United States of America')
& (df_year['ESG'] > 0)]
year_start_firms = year_start.index.unique()
df_year_firms = df_year[['Date', 'eTR', 'MC', 'ESG']].loc[year_start_firms]
df_year_t1 = df_year_t1[['Date', 'eTR', 'MC', 'ESG']]
print(year)
print(" ")
#Normalizes the ESG Scores to the interval (0; 1) and substitues it in; 0.5 is the mean
df = df_year_firms[['ESG']]
Normalized_ESG_year = (df - df.min()) / (df.max() - df.min())
df_year_firms_norm = df_year_firms
df_year_firms_norm[['ESG_norm']] = Normalized_ESG_year.values
df_year_firms_norm = df_year_firms_norm.drop('ESG', axis=1)
df_year_firms_norm = df_year_firms_norm.ffill(axis=0)
df_year_firms_norm.loc[:, 'Month'] = df_year_firms_norm['Date'].dt.month.values.reshape(len(df_year_firms_norm), 1)
df_year_t1.loc[:, 'Month'] = df_year_t1['Date'].dt.month.values.reshape(len(df_year_t1), 1)
data = pd.merge(df_year_firms_norm, df_year_t1.iloc[:, [0,1,-1]], how='inner', on=['Instrument', 'Month'], suffixes=('_t', '_t1'))
#Resets index
data.reset_index(inplace=True)
#Monthwise iteration within the year loop
for month in range(1, 13, 1):
df_month = data[data['Month'] == month]
#ols by statsmodels.formula.api = smf (ols with minor letters)
#DataFrame input
#smf not sm (statsmodels.formula.api instead of statsmodels.api)
ESG_ols = smf.ols(formula = 'eTR_t1 ~ ESG_norm', data=df_month).fit(cov_type='HAC',cov_kwds={'maxlags':0})
ESG_ols_tstat = ESG_ols.tvalues
ESG_ols_coeff = ESG_ols.params
results_df = pd.DataFrame({
'ESG_ols_coeff':ESG_ols_coeff, 'ESG_ols_tstat':ESG_ols_tstat},
index = ['Intercept', 'ESG'])
#Produces the table
df_output = summary_col(
results=[ESG_ols], stars=True, float_format='%0.4f',
model_names=['ESG_ols'],
info_dict={'N':lambda x: "{0:d}".format(int(x.nobs))},
regressor_order = ['Intercept', 'ESG_norm'])
helper = pd.DataFrame(data=[[year, month, ESG_ols.rsquared_adj]],
columns=['year', 'month', 'R2_adj'])
df_store = df_store.append(helper)
print(month)
break
df_store
我也很高兴在效率方面有任何建议！
根据要求，以下是完整的Traceback:
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
~AppDataLocalTemp/ipykernel_8408/1348884998.py in <module>
91 # break
92 #smf not sm (statsmodels.formula.api instead of statsmodels.api)
---> 93 ESG_ols = smf.ols(formula = 'eTR_t1 ~ ESG_norm', data=df_month).fit(cov_type='HAC',cov_kwds={'maxlags':1})
94 #!Assumption: maxlags=0 should be reasonable since we do not have any TS analysis, right?
95 ESG_ols_tstat = ESG_ols.tvalues
~anaconda3libsite-packagesstatsmodelsbasemodel.py in from_formula(cls, formula, data, subset, drop_cols, *args, **kwargs)
193 'formula': formula, # attach formula for unpckling
194 'design_info': design_info})
--> 195 mod = cls(endog, exog, *args, **kwargs)
196 mod.formula = formula
197
~anaconda3libsite-packagesstatsmodelsregressionlinear_model.py in __init__(self, endog, exog, missing, hasconst, **kwargs)
870 def __init__(self, endog, exog=None, missing='none', hasconst=None,
871 **kwargs):
--> 872 super(OLS, self).__init__(endog, exog, missing=missing,
873 hasconst=hasconst, **kwargs)
874 if "weights" in self._init_keys:
~anaconda3libsite-packagesstatsmodelsregressionlinear_model.py in __init__(self, endog, exog, weights, missing, hasconst, **kwargs)
701 else:
702 weights = weights.squeeze()
--> 703 super(WLS, self).__init__(endog, exog, missing=missing,
704 weights=weights, hasconst=hasconst, **kwargs)
705 nobs = self.exog.shape[0]
~anaconda3libsite-packagesstatsmodelsregressionlinear_model.py in __init__(self, endog, exog, **kwargs)
188 """
189 def __init__(self, endog, exog, **kwargs):
--> 190 super(RegressionModel, self).__init__(endog, exog, **kwargs)
191 self._data_attr.extend(['pinv_wexog', 'weights'])
192
~anaconda3libsite-packagesstatsmodelsbasemodel.py in __init__(self, endog, exog, **kwargs)
235
236 def __init__(self, endog, exog=None, **kwargs):
--> 237 super(LikelihoodModel, self).__init__(endog, exog, **kwargs)
238 self.initialize()
239
~anaconda3libsite-packagesstatsmodelsbasemodel.py in __init__(self, endog, exog, **kwargs)
75 missing = kwargs.pop('missing', 'none')
76 hasconst = kwargs.pop('hasconst', None)
---> 77 self.data = self._handle_data(endog, exog, missing, hasconst,
78 **kwargs)
79 self.k_constant = self.data.k_constant
~anaconda3libsite-packagesstatsmodelsbasemodel.py in _handle_data(self, endog, exog, missing, hasconst, **kwargs)
99
100 def _handle_data(self, endog, exog, missing, hasconst, **kwargs):
--> 101 data = handle_data(endog, exog, missing, hasconst, **kwargs)
102 # kwargs arrays could have changed, easier to just attach here
103 for key in kwargs:
~anaconda3libsite-packagesstatsmodelsbasedata.py in handle_data(endog, exog, missing, hasconst, **kwargs)
670
671 klass = handle_data_class_factory(endog, exog)
--> 672 return klass(endog, exog=exog, missing=missing, hasconst=hasconst,
673 **kwargs)
~anaconda3libsite-packagesstatsmodelsbasedata.py in __init__(self, endog, exog, missing, hasconst, **kwargs)
85 self.const_idx = None
86 self.k_constant = 0
---> 87 self._handle_constant(hasconst)
88 self._check_integrity()
89 self._cache = {}
~anaconda3libsite-packagesstatsmodelsbasedata.py in _handle_constant(self, hasconst)
129 # detect where the constant is
130 check_implicit = False
--> 131 exog_max = np.max(self.exog, axis=0)
132 if not np.isfinite(exog_max).all():
133 raise MissingDataError('exog contains inf or nans')
<__array_function__ internals> in amax(*args, **kwargs)
~anaconda3libsite-packagesnumpycorefromnumeric.py in amax(a, axis, out, keepdims, initial, where)
2731 5
2732 """
-> 2733 return _wrapreduction(a, np.maximum, 'max', axis, None, out,
2734 keepdims=keepdims, initial=initial, where=where)
2735
~anaconda3libsite-packagesnumpycorefromnumeric.py in _wrapreduction(obj, ufunc, method, axis, dtype, out, **kwargs)
85 return reduction(axis=axis, out=out, **passkwargs)
86
---> 87 return ufunc.reduce(obj, axis, dtype, out, **passkwargs)
88
89
ValueError: zero-size array to reduction operation maximum which has no identity
我不愿意使用这种方法，但它很有效，即使在本应缺乏数据的月份，我仍然会得到结果。我在这里找到了克服空数组的ValueError的想法
try: #Catches the upcoming Value Error
results = smf.ols(formula = 'eTR_t1 ~ ESG_norm', data=df_month).fit(cov_type='HAC',cov_kwds={'maxlags':0})
except ValueError:
pass
感谢@Josef的建议和帮助。