print(" dist_no_pair: %f" % d_notpair_val_f)
summary = tf.Summary()
summary.value.add(tag="Loss", simple_value=loss_val_f)
val_writer.add_summary(summary, iter_show)
if epoch + 1 == 1 or (epoch + 1) % print_freq == 0:
print("Epoch %d of %d took %fs" %
(epoch + 1, n_epoch, time.time() - start_time))
test_triplet_loss, test_dpair, test_nopair, batch_test_size = 0, 0, 0, 10
test_iters = 2
#test_street_loss, test_shop_loss, test_nopair_loss = 0, 0, 0
#test_street_acc, test_shop_acc, test_nopair_acc = 0, 0, 0
for iter_test in xrange(test_iters):
street_test_batch, shop_test_batch, nopair_test_batch, y_street_test_batch, y_shop_test_batch, y_nopair_test_batch = input.load_test_images(
batch_test_size)
feed_dict = {
x_street: street_test_batch,
x_shop: shop_test_batch,
x_nopair: nopair_test_batch,
train_mode: False
}
test_err, d_test_pair, d_test_nopair = sess.run(
[triplet_loss, dist_pair, dist_nopair], feed_dict=feed_dict)
test_triplet_loss += test_err
test_dpair += d_test_pair
test_nopair += d_test_nopair
#
#print "y-batch[index]:"
#print y_batch[index]
#print "y_batch[index]==1:"
#print y_batch[index] == 1
#print "y_batch:"
#print y_batch
#print y_batch[xrange(batch_size),index.reshape(-1)]
acc = np.mean(y_batch[xrange(batch_size),
index.reshape(-1)].reshape(batch_size, -1) == 1)
if i % 10 == 0:
print_time = time.strftime(ISOTIMEFORMAT, time.localtime())
print print_time
print('Train iteration:{0}, loss:{1}, accuracy:{2}, lr:{3}'.format(
i, cost, acc, lr))
if i % 100 == 0:
x_batch_test, y_batch_test = input.load_test_images(test_num=64)
prob_test = sess.run(nin.prob,
feed_dict={
images: x_batch_test,
train_mode: False
})
test_num = y_batch_test.shape[0]
index_test = np.argsort(prob_test, axis=1)[:, 49:50]
acc = np.mean(y_batch_test[xrange(test_num),
index_test.reshape(-1)].reshape(
test_num, -1) == 1)
print('Test After iterations:{0}, Test Accuracy:{1}'.format(
i, acc))
#for i in xrange(1000):
#for iter in xrange(iter_per_epoch):
#x_batch, y_batch = input.load_batchsize_images(batch_size)
#dp_dict = tl.utils.dict_to_one( network.all_drop ) # disable all dropout/dropconnect/denoising layers
#feed_dict = {x: x_batch, y_: y_batch}
#feed_dict.update(dp_dict)
#err, ac = sess.run([cost, acc], feed_dict=feed_dict)
#assert not np.isnan(err), 'Model diverged with cost = NaN'
#train_loss += err; train_acc += ac;
#print(" train loss: %f" % (train_loss/ iter_per_epoch))
#print(" train acc: %f" % (train_acc/ iter_per_epoch))
test_loss, test_acc, test_top5_acc, test_top10_acc, batch_test_size = 0, 0, 0, 0, 128
test_iters = 312
for iter_test in xrange(test_iters):
#print iter_test
x_test_batch, y_test_batch = input.load_test_images(
batch_test_size)
dp_dict = tl.utils.dict_to_one(
network.all_drop
) # disable all dropout/dropconnect/denoising layers
feed_dict = {x: x_test_batch, y_: y_test_batch}
feed_dict.update(dp_dict)
err, ac, top5_ac, top10_ac = sess.run(
[cost, acc, top5_acc, top10_acc], feed_dict=feed_dict)
test_loss += err
test_acc += ac
test_top5_acc += top5_ac
test_top10_acc += top10_ac
print(" Test Loss: %f" % (test_loss / test_iters))
print(" Test Top1-Acc: %f" % (test_acc / test_iters))
print(" Test Top5-Acc: %f" % (test_top5_acc / test_iters))
print(" Test Top10-Acc: %f" % (test_top10_acc / test_iters))