for itr in range(100):
print("\nFor the {0:03d}".format(itr))
awa_train_path = config.data_path + 'meta/path_label_list{0:03d}.txt'.format(itr)
# num_file : int
# count the number of input image files
with open(awa_train_path) as f:
for num_file, l in enumerate(f):
pass
"""
example) queue_data('/home/siit/navi/data/sample/meta/path_label_list.txt',
50, 1, 'val',multi_label=False)
"""
trainX, trainY = data_loader.queue_data(
awa_train_path, config.n_classes, config.batch_size, 'val', multi_label=False)
feat = []
lab = []
path_feat = {}
for i in range(num_file+1):
batch_x, batch_y = sess.run([trainX, trainY])
_, idx = np.nonzero(batch_y)
feature = sess.run(feat_layer, feed_dict={x: batch_x, y_: batch_y, keep_prob:1.0})
feat.append(feature[0][0][0])
lab.append(idx[0])
path_feat[]
if i%1000 == 0:
total_cost += cost_val
counter += 1
if np.mod(counter, config.print_freq) == 0:
print('Step:', '%05dk' % (counter),
'\tAvg. cost =', '{:.5f}'.format(cost_val),
'\tAcc: {:.5f}'.format(acc_))
writer.add_summary(acc, counter)
# Save the model
if np.mod(counter, config.save_freq) == 0:
if config.nsml:
nsml.save(counter)
if not os.path.exists(config.checkpoint_path):
os.mkdir(config.checkpoint_path)
saver.save(sess, os.path.join(config.checkpoint_path,
'vgg19_{0:03d}k'.format(int(counter/1000))))
print('Model ')
# -------------------- Testing -------------------- #
Xbatch, Ybatch, _ = data_loader.queue_data(
test_data, label_list, im_size)
accuracy_ = sess.run(accuracy, feed_dict = {X: Xbatch, Y: Ybatch})
print('Accuracy:', accuracy_)
train_data = [line for line in path_label_list
if 'train' in line]
test_data = [line for line in path_label_list
if 'test' in line]
num_file = len(train_data)
# print('Number of input files: \t{}'.format(num_file))
total_batch = int(num_file / batch_size)
total_cost = 0
final_acc = 0
for i in range(total_batch):
# Get the batch as [batch_size, 28,28] and [batch_size, n_classes] ndarray
Xbatch, Ybatch, _ = data_loader.queue_data(
train_data[i*batch_size:(i+1)*batch_size], label_list,
im_size)
_, cost_val, acc = sess.run([optimizer, cost, merged], feed_dict={X: Xbatch, Y: Ybatch})
total_cost += cost_val
print('Epoch:', '%04d' % (epoch + 1),
'\tAvg. cost =', '{:.3f}'.format(total_cost / total_batch))
writer.add_summary(acc, epoch)
# Save the model
if epoch % 5 == 0:
if not os.path.exists(config.checkpoint_path):
os.mkdir(config.checkpoint_path)
saver.save(sess, os.path.join(config.checkpoint_path,
test_data = all_data[:1000]
num_file = len(train_data)
if num_file == 0:
break
# print('Number of input files: \t{}'.format(num_file))
total_batch = int(num_file / batch_size)
total_cost = 0
final_acc = 0
for i in range(total_batch):
# Get the batch as [batch_size, 28,28] and [batch_size, n_classes] ndarray
Xbatch, Ybatch, _ = data_loader.queue_data(
train_data[i * batch_size:(i + 1) * batch_size], label_list,
im_size, config.lable_processed)
_, cost_val, acc = sess.run([optimizer, cost, merged],
feed_dict={
X: Xbatch,
Y: Ybatch
})
total_cost += cost_val
if np.mod(i, 10) == 0:
print('Epoch:', '%02d' % (epoch + 1), '\tAvg. cost =',
'{:.3f}'.format(total_cost / total_batch))
print('Epoch:', '%04d' % (epoch + 1), '\tAvg. cost =',
'{:.3f}'.format(total_cost / total_batch))