Tensorflow를 이용한 CNN모델 학습시 발생하는 TypeError

Question

Cifar-10데이터셋 불러와서 batch생성하고 학습시키려고 하는데 아래 사진과 같이 타입에러가 나타납니다. shape이 문제인건지 계산해봐도 도저히 모르겠습니다. 코드는 아래에 있습니다. 어디서 문제가 발생한건지 알려주시면 감사하겠습니다.import tensorflow as tf import random from tensorflow.keras.datasets.cifar10 import load_data (x_train, y_train), (x_test, y_test) &#61; load_data() y_train_one_hot &#61; tf.squeeze(tf.one_hot(y_train, 10), axis&#61;1) y_test_one_hot &#61; tf.squeeze(tf.one_hot(y_test, 10), axis&#61;1) tf.set_random_seed(777) learning_rate &#61; 0.001 training_epochs &#61; 15 batch_size &#61; 100 keep_prob&#61;tf.placeholder(tf.float32) X&#61;tf.placeholder(tf.float32,[None,32,32,3]) Y&#61;tf.placeholder(tf.float32,[None,10]) #L1 ImgIn shape&#61;(?, 32, 32, 3) W1&#61;tf.Variable(tf.random_normal([3,3,3,32], stddev&#61;0.01)) #3*3,channel&#61;1,fiter32개 L1&#61;tf.nn.conv2d(X, W1, strides&#61;[1,1,1,1], padding&#61;'SAME') L1&#61;tf.nn.relu(L1) L1&#61;tf.nn.max_pool(L1,ksize&#61;[1,2,2,1], strides&#61;[1,2,2,1], padding&#61;'SAME') L1&#61;tf.nn.dropout(L1, keep_prob&#61;keep_prob) # Conv ->(?, 32, 32, 32) # Pool ->(?, 16, 16, 32) #L2 ImgIn shape&#61;(?, 16, 16, 32) W2&#61;tf.Variable(tf.random_normal([3,3,32,64],stddev&#61;0.01)) L2&#61;tf.nn.conv2d(L1, W2, strides&#61;[1,1,1,1], padding&#61;'SAME') L2&#61;tf.nn.relu(L2) L2&#61;tf.nn.max_pool(L2, ksize&#61;[1,2,2,1], strides&#61;[1,2,2,1], padding&#61;'SAME') L2&#61;tf.nn.dropout(L2, keep_prob&#61;keep_prob) # Conv ->(?, 16, 16, 64) # Pool ->(?, 8, 8, 64) #L3 ImgIn shape&#61;(?, 8, 8, 64) W3 &#61; tf.Variable(tf.random_normal([3, 3, 64, 128], stddev&#61;0.01)) L3 &#61; tf.nn.conv2d(L2, W3, strides&#61;[1, 1, 1, 1], padding&#61;'SAME') L3 &#61; tf.nn.relu(L3) L3 &#61; tf.nn.max_pool(L3, ksize&#61;[1, 2, 2, 1], strides&#61;[1, 2, 2, 1], padding&#61;'SAME') L3 &#61; tf.nn.dropout(L3, keep_prob&#61;keep_prob) L3_flat &#61; tf.reshape(L3, [-1, 128 * 4 * 4]) # Conv ->(?, 8, 8, 128) # Pool ->(?, 4, 4, 128) # Reshape ->(?, 4 * 4 * 128) # Flatten them for FC W4 &#61; tf.get_variable(&#34;W4&#34;, shape&#61;[128 * 4 * 4, 625], initializer&#61;tf.contrib.layers.xavier_initializer()) b4 &#61; tf.Variable(tf.random_normal([625])) L4 &#61; tf.nn.relu(tf.matmul(L3_flat, W4) &#43; b4) L4 &#61; tf.nn.dropout(L4, keep_prob&#61;keep_prob) W5 &#61; tf.get_variable(&#34;W5&#34;, shape&#61;[625, 10], initializer&#61;tf.contrib.layers.xavier_initializer()) b5 &#61; tf.Variable(tf.random_normal([10])) logits &#61; tf.matmul(L4, W5) &#43; b5 cost&#61;tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits&#61;logits, labels&#61;Y)) optimizer&#61;tf.train.AdamOptimizer(learning_rate&#61;learning_rate).minimize(cost) def next_batch_train(num, starting_point): x_train_batch &#61; x_train[starting_point:num&#43;starting_point] y_train_batch &#61; y_train_one_hot[starting_point:num&#43;starting_point] return x_train_batch, y_train_batch sess&#61;tf.Session() sess.run(tf.global_variables_initializer()) for epoch in range(training_epochs): avg_cost&#61;0 total_batch&#61;int(len(x_train)/batch_size) starting_point&#61;0 for i in range(total_batch): batch_xs, batch_ys&#61;next_batch_train(batch_size, starting_point) c,_&#61;sess.run([cost,optimizer],feed_dict &#61; {X: batch_xs, Y: batch_ys, keep_prob: 0.5}) avg_cost&#43;&#61;c/total_batch starting_point &#43;&#61; 100 print('Epoch:', '%04d' % (epoch &#43; 1), 'cost &#61;', '{:.9f}'.format(avg_cost)) prediction&#61;tf.equal(tf.argmax(logits,1),tf.argmax(Y,1)) accuracy&#61;tf.reduce_mean(tf.cast(prediction,tf.float32)) print('Accuracy:', sess.run(accuracy, feed_dict&#61;{ X: x_test, Y: y_test_one_hot, keep_prob:1}))

침묵술사 · Answer

에러에서 Tensor Object로 넘기면 안된다고 하는데 다른 걸로 변환해서 넘기심이?

Tensorflow를 이용한 CNN모델 학습시 발생하는 TypeError

1개의 답변

Q&A 목록