使用xlogit的MultinomialLogit报错：y值不一致问题求助

我正在学习多项Logit模型，使用xlogit库运行代码时遭遇ValueError，提示**"inconsistent 'y' values. Make sure the data has one choice per sample"**。

已将宽格式数据转为长格式，y（DEPVAR_1）取值为{1,2,3,4}对应四个备选方案，y与alt维度均为(1420,)且无缺失值，尝试reshape时又报错**"y must be an array of one dimension in long format"**，现寻求解决该y值不一致问题的建议。

运行代码 #

# Long format
from xlogit.utils import wide_to_long
ATUS_data_LA_2020_Long = wide_to_long(ATUS_data_LA_2020_wide, id_col='custom_id', alt_name='alt', sep='_',
                  alt_list=['Non-Shopping', 'In-store Shopping', 'Online Shopping', 'Both'], empty_val=0, alt_is_prefix=True)
# List of variables 
index_var_names = ['MSASIZE','FAMINCOME','AGE_1','SEX','EDUC','EMPSTAT','DIFFMOB_1','FSTRUC']

# Reshape
y = ATUS_data_LA_2020_Long['DEPVAR_1'].values.ravel() # reshape 1 dimension
alt = ATUS_data_LA_2020_Long['alt'].values.ravel() # reshape 1 dimension

# Model
from xlogit import MultinomialLogit
model = MultinomialLogit()
model.fit(X=ATUS_data_LA_2020_Long[index_var_names],
          y=y,
          varnames=index_var_names,
          ids=ATUS_data_LA_2020_Long['custom_id'],
          alts=alt,
          fit_intercept=True,
          weights=np.asarray(ATUS_data_LA_2020_Long['WEIGHT']))
model.summary()

报错信息 #

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[14], line 40
     38 from xlogit import MultinomialLogit
     39 model = MultinomialLogit()
---> 40 model.fit(X=ATUS_data_LA_2020_Long[index_var_names],
     41           y=y,
     42           varnames=index_var_names,
     43           ids=ATUS_data_LA_2020_Long['custom_id'],
     44           alts=alt,
     45           fit_intercept=True,
     46           weights=np.asarray(ATUS_data_LA_2020_Long['WEIGHT']))
     47 model.summary()

File ~\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\xlogit\multinomial_logit.py:139, in MultinomialLogit.fit(self, X, y, varnames, alts, ids, isvars, weights, avail, base_alt, fit_intercept, init_coeff, maxiter, random_state, tol_opts, verbose, robust, num_hess, scale_factor)
    134 self._validate_inputs(X, y, alts, varnames, isvars, ids, weights)
    136 self._pre_fit(alts, varnames, isvars, base_alt, fit_intercept, maxiter)
    138 betas, X, y, weights, avail, Xnames, scale = \
--> 139     self._setup_input_data(X, y, varnames, alts, ids, 
    140                            isvars=isvars, weights=weights, avail=avail,
    141                            init_coeff=init_coeff,
    142                            random_state=random_state, verbose=verbose,
    143                            predict_mode=False, scale_factor=scale_factor)
    145 tol = {'ftol': 1e-10}
    146 if tol_opts is not None:

File ~\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\xlogit\multinomial_logit.py:276, in MultinomialLogit._setup_input_data(self, X, y, varnames, alts, ids, isvars, weights, avail, base_alt, fit_intercept, init_coeff, random_state, verbose, predict_mode, scale_factor)
    271 def _setup_input_data(self, X, y, varnames, alts, ids, isvars=None,
    272         weights=None, avail=None, base_alt=None, fit_intercept=False,
    273         init_coeff=None, random_state=None, verbose=1, predict_mode=False,
    274         scale_factor=None):
    275     self._check_long_format_consistency(ids, alts)
--> 276     y = self._format_choice_var(y, alts) if not predict_mode else None
    277     X, Xnames = self._setup_design_matrix(X)
    278     N, J, K = X.shape

File ~\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.10_qbz5n2kfra8p0\LocalCache\local-packages\Python310\site-packages\xlogit\_choice_model.py:193, in ChoiceModel._format_choice_var(self, y, alts)
    191     return y1h
    192 else:
--> 193     raise ValueError("inconsistent 'y' values. Make sure the "
    194                      "data has one choice per sample")

ValueError: inconsistent 'y' values. Make sure the data has one choice per sample

• 修正y变量的标记逻辑：xlogit要求长格式数据中，每个样本（custom_id）对应的多条备选方案行里，只有选中的那个方案对应的y值为1，其余为0，而非直接用1-4表示选中的方案编号。可按以下代码调整：

  # 给备选方案映射对应编号，匹配DEPVAR_1的1-4取值
  alt_mapping = {
      'Non-Shopping': 1,
      'In-store Shopping': 2,
      'Online Shopping': 3,
      'Both': 4
  }
  ATUS_data_LA_2020_Long['alt_code'] = ATUS_data_LA_2020_Long['alt'].map(alt_mapping)
  # 生成正确的选择标记：选中的方案y=1，其余为0
  ATUS_data_LA_2020_Long['y'] = (ATUS_data_LA_2020_Long['alt_code'] == ATUS_data_LA_2020_Long['DEPVAR_1']).astype(int)
  # 提取y值
  y = ATUS_data_LA_2020_Long['y'].values
  

• 验证样本-备选方案结构：确保每个custom_id对应的行数等于备选方案数量（这里是4行），避免样本行数异常导致的结构错误：

  sample_counts = ATUS_data_LA_2020_Long['custom_id'].value_counts()
  # 输出行数不等于4的样本
  print(sample_counts[sample_counts != 4])
  

  若存在异常样本，检查wide_to_long转换参数，比如alt_list是否完整、sep是否匹配宽格式列名的分隔规则。

• 简化y的维度处理：xlogit本身支持一维数组的y输入，无需额外ravel()操作，核心问题是y的取值逻辑错误，而非维度问题。

内容的提问来源于stack exchange，提问作者Catalina V