Question

嗨，我正在使用keras中的自定义损失函数进行编译，但是会引发错误

如果Y_test和Y_pred是二进制分类变量比我必须为Y_test和Y_pred创建哑变量。这里需要将张量对象转换为Numpy数组。

以下是自定义功能

def keras_custom_loss_function(Y_true, Y_pred):
    #convert Y_test and Y_pred into Numpy arrays
    Y_pred  = np.array(Y_pred)
    Y_test  = np.array(Y_test)

    #create Dummy variable for Y_test and Y_pred
    en_Y_test = pd.get_dummies(Y_test)
    en_Y_pred = pd.get_dummies(Y_pred)
     ...
     # DO further operations

＃在编译过程中，它抛出以下错误

model.compile(loss=keras_custom_loss_function, optimizer=opt, metrics=['accuracy'])

TypeError：无法遍历具有未知第一维的张量。

请帮助如何从通过编译函数生成的张量对象（Y_test，Y_pred）创建numpy数组

提前感谢

修改：1 我们正在尝试计算损失函数中的以下逻辑以下是示例假设Y_test和Y_pred值来自keras编译函数，

Y_true = pd.Series([2,1,2,2,1,2,2,2,2,1,2,2,2,2,2,2,2,2,2,1]
Y_pred = pd.Series([2,2,2,1,1,2,2,2,1,1,2,2,2,2,2,2,2,2,2,2]

比具有参数Y_test，Y_Pred的自定义损失函数

def keras_custom_loss_function(Y_true, Y_pred):
        #convert Y_test and Y_pred into Numpy arrays
        Y_pred  = np.array(Y_pred)
        Y_test  = np.array(Y_test)

        #create Dummy variable for Y_test and Y_pred
        en_Y_test = pd.get_dummies(Y_test)
        en_Y_pred = pd.get_dummies(Y_pred)
         # DO further operations

        #Change the column names
        en_Y_true.columns = ["D1","D2"]
        en_Y_pred.columns = ["Y1","Y2"]

        #Create dataframe and append 
        Custm_df = pd.concat([en_Y_true, en_Y_pred], axis=1)

        #convert datatypes 
        Custm_df['D1'] = Custm_df['D1'].astype('int32')
        Custm_df['D2'] = Custm_df['D2'].astype('int32')
        Custm_df['Y1'] = Custm_df['Y1'].astype('int32')
        Custm_df['Y2'] = Custm_df['Y2'].astype('int32')

        Custm_df['D1-Y1']  = Custm_df['D1'] -Custm_df['Y1']
        Custm_df['D2-Y2']  = Custm_df['D2'] -Custm_df['Y2']

        loss = Formula based on (D1,D2,Y1,Y2)

        return loss

我在编译完成时遇到错误

TypeError                                 Traceback (most recent call last)
<ipython-input-22-22841b8add1e> in <module>
     26 
     27 #model.compile(loss=custom_loss, optimizer=opt, metrics=['accuracy']) # loss 'adam' ,y_predicted
---> 28 model.compile(loss=keras_custom_loss_function, optimizer=opt, metrics=['accuracy']) # loss 'adam' ,y_predicted

~\AppData\Local\Continuum\anaconda3\lib\site-packages\keras\engine\training.py in compile(self, optimizer, loss, metrics, loss_weights, sample_weight_mode, weighted_metrics, target_tensors, **kwargs)
    227         #                   loss_weight_2 * output_2_loss_fn(...) +
    228         #                   layer losses.
--> 229         self.total_loss = self._prepare_total_loss(masks)
    230 
    231         # Functions for train, test and predict will

~\AppData\Local\Continuum\anaconda3\lib\site-packages\keras\engine\training.py in _prepare_total_loss(self, masks)
    690 
    691                     output_loss = loss_fn(
--> 692                         y_true, y_pred, sample_weight=sample_weight)
    693 
    694                 if len(self.outputs) > 1:

~\AppData\Local\Continuum\anaconda3\lib\site-packages\keras\losses.py in __call__(self, y_true, y_pred, sample_weight)
     69         scope_name = 'lambda' if self.name == '<lambda>' else self.name
     70         with K.name_scope(scope_name):
---> 71             losses = self.call(y_true, y_pred)
     72             return losses_utils.compute_weighted_loss(
     73                 losses, sample_weight, reduction=self.reduction)

~\AppData\Local\Continuum\anaconda3\lib\site-packages\keras\losses.py in call(self, y_true, y_pred)
    130             Loss values per sample.
    131         """
--> 132         return self.fn(y_true, y_pred, **self._fn_kwargs)
    133 
    134     def get_config(self):

<ipython-input-21-e80cbe1c715d> in keras_custom_loss_function(Y_true, Y_pred)
     39 
     40     #label_numpy = Y_pred.eval()
---> 41     en_Y_pred = pd.get_dummies(Y_pred)
     42     print("second after")
     43     en_Y_true.columns = ["D1","D2"]

~\AppData\Local\Continuum\anaconda3\lib\site-packages\pandas\core\reshape\reshape.py in get_dummies(data, prefix, prefix_sep, dummy_na, columns, sparse, drop_first, dtype)
    864                                  sparse=sparse,
    865                                  drop_first=drop_first,
--> 866                                  dtype=dtype)
    867     return result
    868 

~\AppData\Local\Continuum\anaconda3\lib\site-packages\pandas\core\reshape\reshape.py in _get_dummies_1d(data, prefix, prefix_sep, dummy_na, sparse, drop_first, dtype)
    872     from pandas.core.reshape.concat import concat
    873     # Series avoids inconsistent NaN handling
--> 874     codes, levels = _factorize_from_iterable(Series(data))
    875 
    876     if dtype is None:

~\AppData\Local\Continuum\anaconda3\lib\site-packages\pandas\core\series.py in __init__(self, data, index, dtype, name, copy, fastpath)
    227             elif (isinstance(data, compat.Iterable)
    228                   and not isinstance(data, compat.Sized)):
--> 229                 data = list(data)
    230             else:
    231 

~\AppData\Local\Continuum\anaconda3\lib\site-packages\tensorflow\python\framework\ops.py in __iter__(self)
    484     if shape[0] is None:
    485       raise TypeError(
--> 486           "Cannot iterate over a tensor with unknown first dimension.")
    487     for i in xrange(shape[0]):
    488       yield self[i]

TypeError: Cannot iterate over a tensor with unknown first dimension.

将自定义损失函数输出y test和y pred转换为numpy数组

0 个答案: