java 张量运算,乘以两个张量

我在Tensorflow中有两个张量，它们有以下两种形状：

print(tf.valid_dataset.get_shape())

print(weights1.get_shape())

有结果：

(10000, 784)

(784, 1024)

但是，如果我试图将它们相乘，就像这样：

tf.matmul(tf_valid_dataset, weights1)

我明白了：

Tensor("Variable:0", shape=(784, 1024), dtype=float32_ref) must be from the same graph as Tensor("Const:0", shape=(10000, 784), dtype=float32).

由于我将它们放在它们都具有 784 大小的一侧，这对我来说似乎是正确的 .

知道什么可能是错的吗？

编辑：

我在print语句之前的代码是这样的：

num_hidden_nodes=1024

batch_size = 128

learning_rate = 0.5

graph = tf.Graph()

with graph.as_default():

tf_train_dataset = tf.placeholder(tf.float32, shape=(batch_size, image_size*image_size))

tf_train_labels = tf.placeholder(tf.float32, shape=(batch_size, num_labels))

tf.valid_dataset = tf.constant(valid_dataset)

tf.test_dataset = tf.constant(test_dataset)

weights1 = tf.Variable(tf.truncated_normal([image_size * image_size, num_hidden_nodes]))

biases1 = tf.Variable(tf.zeros([num_hidden_nodes]))

weights2 = tf.Variable(tf.truncated_normal([num_hidden_nodes, num_labels]))

biases2 = tf.Variable(tf.zeros([num_labels]))

weights = [weights1, biases1, weights2, biases2]

lay1_train = tf.nn.relu(tf.matmul(tf_train_dataset, weights1) + biases1)

logits = tf.matmul(lay1_train, weights2) + biases2

loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logits, labels=tf_train_labels))

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