ValueError: Rank mismatch: Rank of labels (received 2) should equal rank of logits minus 1 (received

多分类任务时， tf.nn.sparse_softmax_cross_entropy_with_logits

for batchTrain in nextBatch(trainX, trainlabels, config.batchSize):
    print(batchTrain[0]) # 32 *type_num 
    print(batchTrain[1]) # 32 * 1
    # print(len(batchTrain[0]))
    print(type(batchTrain[1]))
    loss= tf.nn.sparse_softmax_cross_entropy_with_logits(labels=batchTrain[1], logits=batchTrain[0])
    # loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=tf.reshape(batchTrain[1],[-1]), logits=batchTrain[0])
    print(loss)
    break

其中
batchTrain[0]维度为32*type_num

[[0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 ...
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]
 [0. 0. 0. ... 0. 0. 0.]]

batchTrain[1]的维度为32*1

[[8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]
 [8]]

求损失函数时，labels应比logits少一个维度，上面两者维度都为2

 loss= tf.nn.sparse_softmax_cross_entropy_with_logits(labels=batchTrain[1], logits=batchTrain[0])

改为

loss = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=tf.reshape(batchTrain[1],[-1]), logits=batchTrain[0])