TensorFlow concat中的排名不匹配錯誤

Question

我一直在嘗試制作一個簡單的2層神經網絡。 我研究了tensorflow api和官方教程，制作了一個分層模型，但是在神經網絡中遇到了麻煩。 這是導致錯誤的代碼的一部分：

with graph.as_default():
    tf_train_dataset = tf.placeholder(tf.float32, shape=(batch_size, image_size * image_size))
    tf_train_labels = tf.placeholder(tf.int32, shape=(batch_size, num_labels))
    tf_valid_dataset = tf.constant(valid_dataset)
    tf_test_dataset = tf.constant(test_dataset)

    weights0 = tf.Variable(tf.truncated_normal([image_size**2, num_labels]))
    biases0 = tf.Variable(tf.zeros([num_labels]))

    hidden1 = tf.nn.relu(tf.matmul(tf_test_dataset, weights0) + biases0)

    weights1 = tf.Variable(tf.truncated_normal([num_labels, image_size * image_size]))
    biases1 = tf.Variable(tf.zeros([image_size**2]))

    hidden2 = tf.nn.relu(tf.matmul(hidden1, weights1) + biases1)


    logits = tf.matmul(hidden2, weights0) + biases0

    labels = tf.expand_dims(tf_train_labels, 1)

    indices = tf.expand_dims(tf.range(0, batch_size), 1)

    concated = tf.concat(1, [indices, tf.cast(labels,tf.int32)])

    onehot_labels = tf.sparse_to_dense(concated, tf.pack([batch_size, num_labels]), 1.0, 0.0)


    loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits, onehot_labels))

    optimizer = tf.train.GradientDescentOptimizer(0.5).minimize(loss)

    train_prediction = tf.nn.softmax(logits)
    valid_prediction = tf.nn.softmax(tf.matmul(tf.nn.relu(tf.matmul(tf.nn.relu(tf.matmul(tf_valid_dataset,weights0) + biases0),weights1)+biases1),weights0)+biases0)
    test_prediction = tf.nn.softmax(tf.matmul(tf.nn.relu(tf.matmul(tf.nn.relu(tf.matmul(tf_test_dataset,weights0) + biases0),weights1)+biases1),weights0)+biases0)

錯誤是：

Traceback (most recent call last):
    File "./test1.py", line 60, in <module>
    concated = tf.concat(1, [indices, tf.cast(labels,tf.int32)])
    File "/Users/username/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 309, in concat
    name=name)
    File "/Users/username/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/gen_array_ops.py", line 70, in _concat
    name=name)
    File "/Users/username/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/op_def_library.py", line 664, in apply_op
    op_def=op_def)
    File "/Users/username/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1836, in create_op
    set_shapes_for_outputs(ret)
    File "/Users/username/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/ops.py", line 1476, in set_shapes_for_outputs
    shapes = shape_func(op)
    File "/Users/username/tensorflow/lib/python2.7/site-packages/tensorflow/python/ops/array_ops.py", line 364, in _ConcatShape
    concat_dim + 1:].merge_with(value_shape[concat_dim + 1:])
    File "/Users/username/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/tensor_shape.py", line 527, in merge_with
    self.assert_same_rank(other)
    File "/Users/username/tensorflow/lib/python2.7/site-packages/tensorflow/python/framework/tensor_shape.py", line 570, in assert_same_rank
    "Shapes %s and %s must have the same rank" % (self, other))
    ValueError: Shapes TensorShape([]) and TensorShape([Dimension(10)]) must have the same rank

這是完整的代碼： http : //pastebin.com/sX7RqbAf

我已經使用過TensorFlow和Python 2.7。 我對神經網絡和機器學習非常陌生，所以請您多謝我，如有任何錯誤，請多謝。

Answer 1

在您的示例中：

• tf_train_labels形狀為[batch_size, num_labels]
• 因此， labels形狀為[batch_size, 1, num_labels]
• indices形狀為[batch_size, 1]

因此，當您編寫：

concated = tf.concat(1, [indices, tf.cast(labels,tf.int32)])

它會引發錯誤，因為labels和indices的第三個維度是不同的。 labels具有尺寸num_labels （大概為10）的第三維， indices沒有第三維。