Python 无法在keras中定义自定义损失函数
使用:Python 无法在keras中定义自定义损失函数,python,tensorflow,keras,jupyter-notebook,Python,Tensorflow,Keras,Jupyter Notebook,使用: Keras 2.1.5 ipython 6.2.1 ipython_genutils 0.2.0 Python 3 Windows 7(x64)-8GB ram-仅限CPU 我正试图在jupyter笔记本上制作一个神经网络,它工作得很好,直到我决定编写一个自定义损失函数。对于初学者,我决定编写一个简单的均方损失函数。 损失函数为: def msqeloss(y_true, y_pred): return K.sum(K.square(y_true-y_pred),0) 我将此
def msqeloss(y_true, y_pred):
return K.sum(K.square(y_true-y_pred),0)
我将此函数输入到模型中,如下所示:
...
model.add(Dropout(0.1))
model.add(Flatten())
model.add(Dense(120,activation='relu'))
model.add(Dropout(0.1))
model.add(Dense(3,activation='softmax'))
model.compile(optimizer='RMSprop',loss=msqeloss,metrics=['accuracy'])
...
完整堆栈错误如下所示:
AttributeError Traceback (most recent call last)
<ipython-input-10-86f0a38b9a14> in <module>()
42 model.add(Dropout(0.1))
43 model.add(Dense(3,activation='softmax'))
---> 44 model.compile(optimizer='RMSprop',loss=msqeloss,metrics=['accuracy'])
45 csv_logger = CSVLogger('meanlog_loss_log'+str(index)+'.csv', append=True, separator=';')
46 model.fit(Xtrain,one_hot_labels,batch_size=10,epochs=Epochs,validation_data=(Xtest,np_utils.to_categorical(ytest, num_classes=3)),callbacks=[csv_logger])
~\Anaconda3\lib\site-packages\keras\models.py in compile(self, optimizer, loss, metrics, sample_weight_mode, weighted_metrics, target_tensors, **kwargs)
822 weighted_metrics=weighted_metrics,
823 target_tensors=target_tensors,
--> 824 **kwargs)
825 self.optimizer = self.model.optimizer
826 self.loss = self.model.loss
~\Anaconda3\lib\site-packages\keras\engine\training.py in compile(self, optimizer, loss, metrics, loss_weights, sample_weight_mode, weighted_metrics, target_tensors, **kwargs)
828 with K.name_scope(self.output_names[i] + '_loss'):
829 output_loss = weighted_loss(y_true, y_pred,
--> 830 sample_weight, mask)
831 if len(self.outputs) > 1:
832 self.metrics_tensors.append(output_loss)
~\Anaconda3\lib\site-packages\keras\engine\training.py in weighted(y_true, y_pred, weights, mask)
427 """
428 # score_array has ndim >= 2
--> 429 score_array = fn(y_true, y_pred)
430 if mask is not None:
431 # Cast the mask to floatX to avoid float64 upcasting in Theano
<ipython-input-9-d8157a575c41> in msqeloss(y_true, y_pred)
1 def msqeloss(y_true, y_pred):
----> 2 return K.sum(K.square(y_true-y_pred),0)
AttributeError: 'int' object has no attribute 'sum'
AttributeError回溯(最近一次调用)
在()
42型号。添加(辍学(0.1))
43型号。添加(密集型(3,激活=softmax'))
--->44 model.compile(优化器='RMSprop',loss=msqeloss,metrics=['accurity']))
45 csv_logger=CSVLogger('meanlog_loss_log'+str(index)+'.csv',append=True,separator=';'))
46 model.fit(Xtrain,one_hot_标签,批处理大小=10,epochs=epochs,验证数据=(Xtest,np_utils.to_category(ytest,num_classes=3)),回调=[csv_logger])
编译中的~\Anaconda3\lib\site packages\keras\models.py(self、优化器、损耗、度量、样本权重模式、加权度量、目标张量、**kwargs)
822加权度量=加权度量,
823目标张量=目标张量,
-->824**夸尔格)
825 self.optimizer=self.model.optimizer
826 self.loss=self.model.loss
编译中的~\Anaconda3\lib\site packages\keras\engine\training.py(self、优化器、损耗、度量、损耗权重、样本权重模式、加权度量、目标张量、**kwargs)
828带有K.name \-u作用域(self.output \-u names[i]+'.\u loss'):
829输出损耗=加权损耗(y_真,y_pred,
-->830样品(重量,面罩)
831如果透镜(自输出)>1:
832自度量张量追加(输出损失)
~\Anaconda3\lib\site packages\keras\engine\training.py加权(y_真、y_pred、权重、掩码)
427 """
428#分数_阵列的ndim>=2
-->429分数数组=fn(y_真,y_pred)
430如果掩码不是无:
431#将遮罩投射到floatX上,以避免float64在场景中向上投射
在msqeloss中(y_true,y_pred)
1 def msqeloss(y_true,y_pred):
---->2返回K.sum(K.square(y_true-y_pred),0)
AttributeError:“int”对象没有属性“sum”
我是Keras的新手,任何帮助都将不胜感激。请帮我解决这个问题,因为它已经困扰我两天了。
提前感谢。您在哪里定义了
K
?是的,正如@Kapil所提到的,似乎在您的代码中的某个地方,您将K
从keras.backend
重新定义为int
变量……非常感谢您。即使在这里问这个问题,我现在也觉得很傻。非常感谢!