使用CoxTimeVaryingFitter遇类型错误：numpy.float64无exp属性

尝试在数据集上用CoxTimeVaryingFitter建模，但baseline_cumulative_hazard_出现类型相关问题。已尝试缩减特征排查，仍无法拟合数据集，请问问题源于数据还是模型？

代码 #

from lifelines import CoxTimeVaryingFitter
import autograd.numpy as np
ctv = CoxTimeVaryingFitter()

comp = 'comp_comp1' #start with comp1
event = 'failure_'+comp.split("_")[1]
cols = ['start', 'stop',
        'machineID', 
        'age', 
         event,
        'volt_24_ma','rotate_24_ma', 'vibration_24_ma', 'pressure_24_ma'
         ]

ctv.fit(df_X_train[cols].dropna(),
    id_col='machineID',
    event_col=event,
    start_col='start',
    stop_col='stop', 
    show_progress=True,
    fit_options={'step_size':0.25})
ctv.print_summary()
ctv.plot()

数据情况 #

• 数据类型：各字段类型符合生存分析基本要求
• 时序数据：为时间依赖型的生存数据结构

错误信息 #

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
AttributeError: 'numpy.float64' object has no attribute 'exp'

The above exception was the direct cause of the following exception:

TypeError                                 Traceback (most recent call last)
<command-1950361299690996> in <module>
     23     ]
     24 
---> 25 ctv.fit(df_X_train[cols].dropna(),
     26     id_col='machineID',
     27     event_col=event,

/databricks/python/lib/python3.8/site-packages/lifelines/fitters/cox_time_varying_fitter.py in fit(self, df, event_col, start_col, stop_col, weights_col, id_col, show_progress, robust, strata, initial_point, formula, fit_options)
    237         self.confidence_intervals_ = self._compute_confidence_intervals()
    238         self.baseline_cumulative_hazard_ = self._compute_cumulative_baseline_hazard(df, events, start, stop, weights)
---> 239         self.baseline_survival_ = self._compute_baseline_survival()
    240         self.event_observed = events
    241         self.start_stop_and_events = pd.DataFrame({"event": events, "start": start, "stop": stop})

/databricks/python/lib/python3.8/site-packages/lifelines/fitters/cox_time_varying_fitter.py in _compute_baseline_survival(self)
    815 
    816     def _compute_baseline_survival(self):
---> 817         survival_df = np.exp(-self.baseline_cumulative_hazard_)
    818         survival_df.columns = ["baseline survival"]
    819         return survival_df

/databricks/python/lib/python3.8/site-packages/pandas/core/generic.py in __array_ufunc__(self, ufunc, method, *inputs, **kwargs)
   1934         self, ufunc: Callable, method: str, *inputs: Any, **kwargs: Any
   1935     ):
-> 1936         return arraylike.array_ufunc(self, ufunc, method, *inputs, **kwargs)
   1937 
   1938     # ideally we would define this to avoid the getattr checks, but

/databricks/python/lib/python3.8/site-packages/pandas/core/arraylike.py in array_ufunc(self, ufunc, method, *inputs, **kwargs)
    364             # take this path if there are no kwargs
    365             mgr = inputs[0]._mgr
-> 366             result = mgr.apply(getattr(ufunc, method))
    367         else:
    368             # otherwise specific ufunc methods (eg np.<ufunc>.accumulate(..))

/databricks/python/lib/python3.8/site-packages/pandas/core/internals/managers.py in apply(self, f, align_keys, ignore_failures, **kwargs)
    423             try:
    424                 if callable(f):
-> 425                     applied = b.apply(f, **kwargs)
    426                 else:
    427                     applied = getattr(b, f)(**kwargs)

/databricks/python/lib/python3.8/site-packages/pandas/core/internals/blocks.py in apply(self, func, **kwargs)
    376         """
    377         with np.errstate(all="ignore"):
-> 378             result = func(self.values, **kwargs)
    379 
    380         return self._split_op_result(result)

TypeError: loop of ufunc does not support argument 0 of type numpy.float64 which has no callable exp method

解答 #

这个问题源于autograd和numpy的兼容性冲突，不是数据问题。

你导入的是autograd.numpy而非标准numpy，但lifelines内部在计算基线生存函数时，期望用标准numpy的exp方法处理pandas数据结构。autograd的numpy实现和pandas的数组操作逻辑不兼容，导致np.exp无法正确识别pandas中的numpy.float64类型。

解决步骤：

1. 替换导入语句，改用标准numpy：

   import numpy as np
   # 移除 import autograd.numpy as np
   

2. 如果你的代码中确实需要autograd的功能，可单独导入并使用别名区分：

   import numpy as np
   import autograd.numpy as anp
   

3. 重新运行拟合代码，问题即可解决。

内容的提问来源于stack exchange，提问作者Cameron