OP_REQUIRES失败无效参数：转置期望大小为5的向量。但是input（1）是大小为4的向量

Question

我有一个TF Estimator模型，其代码如下： '''

estimator_model = keras.estimator.model_to_estimator(keras_model=model, model_dir=outputdir)


def batchGen(features, labels):      
    num = 0
    for i in itertools.count(0):   
        if(num < len(features)):
           yield features[i], labels[i]
           num+=1 
        else:
           break  


def newFeatureFunc(x):
  x = x.decode("utf-8") 
  outT = tokenizer.texts_to_sequences(x, maxlen)
  outT = outT.astype(np.float32)
  return outT

def newLabelFunc(y):
    outT = utils.to_categorical(y, nb_classes)
    return outT

def train_fn_custom(features, labels, batch_size): 
    genSet = lambda : batchGen(features,labels)
    dataset=tf.data.Dataset.from_generator(genSet,(tf.string, tf.float32),(tf.TensorShape([]), tf.TensorShape([])))

    def _preprocess_function(features, labels):      
        output1 = tf.py_func(newFeatureFunc, [features], tf.float32)       
        output1.set_shape([,maxlen])        
        output2 = tf.py_func(newLabelFunc, [labels], tf.float32)
        output2.set_shape([,nb_classes])
        return output1, output2

    dataset = dataset.map(_preprocess_function)
    dataset = dataset.repeat(9)
    dataset = dataset.batch(batch_size)
    return dataset

batch_size = 1
customTrain = lambda:train_fn_custom(X1, y_train, batch_size)
estimator_model.train(input_fn=customTrain, steps=100)

'''

这会产生错误

  

OP_REQUIRES在transpose_op.cc:157处失败：参数无效：transpose期望一个大小为5的向量。但是input（1）是一个大小为4的向量

我无法指向导致此错误的代码的位置以及在我的估算器的上下文中的含义。有人可以提供一些见解吗？