The ml-course1 from infilect

ml-course1's People

Contributors

Stargazers

Watchers

ml-course1's Issues

Unable to restore the convnet model trained on MNIST

Hi,

I wrote below snippet for restoring model but I get error when I apply softmax over logits.
"ValueError: Argument must be a dense tensor: got shape [1, 128, 10], but wanted [1]."

with tf.Session() as sess:

meta_file = os.path.join(checkpoints_dir, 'mnist-convnet-419.meta')  # partially documented: saver.save stores this
saver = tf.train.import_meta_graph(meta_file, clear_devices=True)
saver.restore(sess, os.path.join(checkpoints_dir, 'mnist-convnet-419'))

X = sess.graph.get_tensor_by_name("data/X_placeholder:0")
Y = sess.graph.get_tensor_by_name("data/Y_placeholder:0")
dropout = sess.graph.get_tensor_by_name("dropout:0")

w = sess.graph.get_tensor_by_name("softmax_linear/weights:0")
b = sess.graph.get_tensor_by_name("softmax_linear/biases:0")
fc = sess.graph.get_tensor_by_name("fc/relu:0")
fc = tf.nn.dropout(fc, dropout, name='relu_dropout')

logits = tf.matmul(fc, w) + b


# test the model
n_batches = int(mnist.test.num_examples / BATCH_SIZE)
total_correct_preds = 0
for i in range(n_batches):
    X_batch, Y_batch = mnist.test.next_batch(BATCH_SIZE)
    logits_batch = sess.run([logits], feed_dict={X: X_batch, Y: Y_batch, dropout: DROPOUT})
    preds = tf.nn.softmax(logits_batch)
    correct_preds = tf.equal(tf.argmax(preds, 1), tf.argmax(Y_batch, 1))
    accuracy = tf.reduce_sum(tf.cast(correct_preds, tf.float32))
    total_correct_preds += sess.run(accuracy)

print("Accuracy {0}".format(total_correct_preds / mnist.test.num_examples))

Recommend Projects