def evaluate(sess, x_, y_):
"""评估在某一数据上的准确率和损失"""
data_len = len(x_)
batch_eval = batch_iter(x_, y_, 128)
total_loss = 0.0
total_acc = 0.0
for x_batch, y_batch in batch_eval:
batch_len = len(x_batch)
feed_dict = feed_data(x_batch, y_batch, 1.0)
loss, acc = sess.run([model.loss, model.acc], feed_dict=feed_dict)
total_loss += loss * batch_len
total_acc += acc * batch_len
return total_loss / data_len, total_acc / data_len
def evalute_dev_set(run_number, model_number):
checkpoint_file = cwd + "/runs/{}/checkpoints/model-{}".format(
run_number, model_number)
graph = tf.Graph()
with graph.as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
# Load the saved meta graph and restore variables
saver = tf.train.import_meta_graph(
"{}.meta".format(checkpoint_file))
saver.restore(sess, checkpoint_file)
# Get the placeholders from the graph by name
input = graph.get_operation_by_name("true_images").outputs[0]
# Tensors we want to evaluate
scores = graph.get_operation_by_name(
"discr/output/score_pred"
).outputs[
0] #this has to be changed to output/score_pred with the new version of model
# Generate batches for evaluation
batches = preprocessing.batch_iter(scored_data,
FLAGS.batch_size,
1,
shuffle=False)
# Predicting the score from the model
acc = 0
n = 0
for test_batch in batches:
batch_imgs = np.reshape(np.concatenate(test_batch[:, 0]),
(-1, 1000, 1000))
batch_score = np.reshape(test_batch[:, 1], (-1))
batch_pred_scores = sess.run(scores, {input: batch_imgs})
for i in range(len(batch_pred_scores)):
tmp = max(0.0, min(8.0, batch_pred_scores[i]))
acc += abs(tmp - batch_score[i])
n += 1
# Return the MAE on the dev set.
return (acc / n)
saver = tf.train.import_meta_graph(meta_graph_path)
saver.restore(sess, checkpoint_folder)
graph = tf.get_default_graph()
graph_input_y = graph.get_tensor_by_name("input_y:0")
graph_input_x = graph.get_tensor_by_name("input_x:0")
graph_batch_size = graph.get_tensor_by_name("batch_size:0")
graph_predictions = graph.get_tensor_by_name("output_layer/predictions:0")
graph_losses = graph.get_tensor_by_name("output_layer/losses/Reshape_2:0")
nlpdata = datasets.NLUProject1Dataset()
nlpdatatest = datasets.NLUProject1TestDataset()
print(len(nlpdatatest.test_x))
batches = prep.batch_iter(list(zip(nlpdatatest.test_x, nlpdatatest.test_y)), con.BATCH_SIZE, con.n_epochs)
#print(len(batches))
for batch in batches:
x_batch, y_batch = zip(*batch)
feed_dict = {
graph_input_x: x_batch,
graph_input_y: y_batch,
graph_batch_size: len(x_batch)
}
losses, input_y = sess.run(
[graph_losses, graph_input_y],
feed_dict)
time_str = datetime.datetime.now().isoformat()
step, summaries, loss, accuracy = sess.run(
[global_step, dev_summary_op, cnn.loss, cnn.accuracy],
feed_dict)
acc.append(accuracy)
losses.append(loss)
time_str = datetime.datetime.now().isoformat()
print("batch " + str(i + 1) + " in dev >>" +
" {}: loss {:g}, acc {:g}".format(
time_str, loss, accuracy))
if writer:
writer.add_summary(summaries, step)
print("\nMean accuracy=" + str(sum(acc) / len(acc)))
print("Mean loss=" + str(sum(losses) / len(losses)))
# Generate batches in one-hot-encoding format
batches = preprocessing.batch_iter(x_train, y_train, FLAGS.batch_size,
FLAGS.num_epochs)
# Training loop. For each batch...
for batch in batches:
x_batch, y_batch = zip(*batch)
train_step(x_batch, y_batch)
current_step = tf.train.global_step(sess, global_step)
if current_step % FLAGS.evaluate_every == 0:
print("\nEvaluation:")
dev_step(x_dev, y_dev, writer=dev_summary_writer)
print("")
if current_step % FLAGS.checkpoint_every == 0:
path = saver.save(sess,
checkpoint_prefix,
global_step=current_step)
print("Saved model checkpoint to {}\n".format(path))
sess = tf.Session(config=session_conf)
with sess.as_default():
# Load the saved meta graph and restore variables
saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
saver.restore(sess, checkpoint_file)
# Get the placeholders from the graph by name
input = graph.get_operation_by_name("true_images").outputs[0]
# Tensors we want to evaluate
scores = graph.get_operation_by_name(
"discr/output/score_pred").outputs[0]
# Generate batches from the query dataset for one epoch
batches = preprocessing.batch_iter(query_data,
FLAGS.batch_size,
1,
shuffle=False)
# If necessary Create the directory predictions
timestamp = str(run_number)
out_dir = os.path.abspath(os.path.join(os.path.curdir, "predictions"))
if not os.path.exists(out_dir):
os.makedirs(out_dir)
# Defining the csv file name
out_file = os.path.abspath(
os.path.join(out_dir, "{}".format(timestamp)))
print("Writing to {}\n".format(out_file))
with open("{}.csv".format(out_file), "w") as file:
# Create the required header
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
# load the saved meta graph and restore variables
saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
saver.restore(sess, checkpoint_file)
# Get the placeholders from the graph by name
x = graph.get_operation_by_name("X").outputs[0]
dropout_keep_prob = graph.get_operation_by_name("dropout_keep_prob").outputs[0]
# Tensors we wanna evaluate
predictions = graph.get_operation_by_name("accuracy/predictions").outputs[0]
# Generate batches for one epoch
batches = preprocessing.batch_iter(x_test, FLAGS.batch_size, 1, shuffle=False)
# Collect the predictions here
all_predictions = []
for x_test_batch in batches:
batch_pred = sess.run(predictions, {x: x_test_batch, dropout_keep_prob: 1.0})
all_predictions = np.concatenate([all_predictions, batch_pred])
correct_pred = float(sum(all_predictions == y_test))
print("Total number of text examples: {}".format(len(y_test)))
print("Accuracy: {:g}".format(correct_pred/float(len(y_test))))
sess.run(assign_op)
W_fc = cnn.W_fc.eval()
l2_norm = np.linalg.norm(W_fc)
# print(">>>>>>> NEW L2 NORM IS:" + str(l2_norm))
def validation_step(x_batch, y_batch, writer=None):
feed_dict = {cnn.x: x_batch, cnn.y: y_batch, cnn.dropout_keep_prob: 1.0}
step, summaries, loss, accuracy = sess.run([global_step, dev_summary_op, cnn.loss, cnn.accuracy],
feed_dict=feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
if writer:
writer.add_summary(summaries, step)
# Generate batches
batches = preprocessing.batch_iter(list(zip(x_train, y_train)), tf.flags.FLAGS.batch_size,
tf.flags.FLAGS.n_epochs)
# Training loop. For each batch...
for batch in batches:
x_batch, y_batch = zip(*batch)
train_step(x_batch, y_batch)
current_step = tf.train.global_step(sess, global_step)
# if current_step % tf.flags.FLAGS.evaluate_every == 0:
# print("\nEvaluation on Validation set:")
# validation_step(x_valid, y_valid, writer=dev_summary_writer)
# print("")
# if current_step % tf.flags.FLAGS.checkpoint_every == 0:
# path = saver.save(sess, checkpoint_prefix, global_step=current_step)
# print("Saved model checkpoint to {}\n".format(path))
print("\nEvaluation on Validation set:")
validation_step(x_valid, y_valid, writer=dev_summary_writer)
#testing step
def test_step(x_batch, y_batch, writer=None):
feed_dict = {
cnn.input_x:x_batch,
cnn.input_y:y_batch,
cnn.dropout_keep_prob:1.0
}
step, summaries, loss, accuracy = sess.run([global_step, test_summary_op, cnn.loss, cnn.accuracy], feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
if writer:
writer.add_summary(summaries, step)
return loss, accuracy
batches = preprocessing.batch_iter(list(zip(x_train, y_train)), batch_size, num_epochs)
#存储训练集和样本集的loss和accuracy，以供之后的画图
train_loss_all = []
train_accuracy_all = []
test_loss_all = []
test_accuracy_all = []
#进行训练 training loop for every step
for batch in batches:
x_batch, y_batch = zip(*batch)
loss_train, accuracy_train = train_step(x_batch, y_batch)
train_loss_all.append(loss_train)
train_accuracy_all.append(accuracy_train)
current_step = tf.train.global_step(sess, global_step) #将Session和global_step
#每evaluateevery次进行一次测试
if current_step % evaluate_every == 0:
def train():
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
if os.path.exists('tensorboard'):
shutil.rmtree('tensorboard')
tensorboard_dir = 'tensorboard/textcnn'
if not os.path.exists(tensorboard_dir):
os.makedirs(tensorboard_dir)
tf.summary.scalar("loss", model.loss)
tf.summary.scalar("accuracy", model.acc)
merged_summary = tf.summary.merge_all()
writer = tf.summary.FileWriter(tensorboard_dir)
# 配置 Saver
saver = tf.train.Saver()
if not os.path.exists(save_dir):
os.makedirs(save_dir)
print("Loading training and validation data...")
# 载入训练集与验证集
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id,
config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
for epoch in range(config.num_epochs):
print('Epoch:', epoch + 1)
batch_train = batch_iter(x_train, y_train, config.batch_size)
for x_batch, y_batch in batch_train:
feed_dict = feed_data(x_batch, y_batch, config.dropout_keep_prob)
if total_batch % config.save_per_batch == 0:
# 每多少轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
feed_dict[model.keep_prob] = 1.0
loss_train, acc_train = session.run([model.loss, model.acc],
feed_dict=feed_dict)
loss_val, acc_val = evaluate(session, x_val, y_val) # todo
if acc_val > best_acc_val:
# 保存最好结果
best_acc_val = acc_val
last_improved = total_batch
saver.save(sess=session, save_path=save_path)
improved_str = '*'
else:
improved_str = ''
time_dif = get_time_dif(start_time)
msg = 'Iter: {0:>6}, Train Loss:{1:>6.2}, Train Acc:{2:>7.2%},' \
+ ' Val Loss:{3:>6.2}, Val Acc:{4:>7.2%}, Time:{5} {6}'
print(
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
feed_dict[model.keep_prob] = config.dropout_keep_prob
session.run(model.optim, feed_dict=feed_dict) # 运行优化
total_batch += 1
if total_batch - last_improved > require_improvement:
# 验证集正确率长期不提升，提前结束训练
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
##### TRAIN / TEST SPLIT #####
# Randomly shuffle data
np.random.seed(10)
# Train/test split for score
shuffled_indices = np.random.permutation(len(scored_data[:,0]))
dev_sample_index = int(FLAGS.dev_sample_percentage * float(len(shuffled_indices)))
test_indices = shuffled_indices[:dev_sample_index]
train_indices = shuffled_indices[dev_sample_index:]
train_score = scored_data[train_indices,:]
test_scored = scored_data[test_indices,:]
# Needed for dev step - but actually we only evaluated on the first 64 (not the full 96 test set) for tensorboard.
test_score_imgs = np.reshape(np.concatenate(test_scored[:, 0]), (-1,1000,1000))
test_score = np.reshape(test_scored[:, 1], (-1))
# Generate training batches for scored images
batches_scored = preprocessing.batch_iter(train_score, FLAGS.batch_size, FLAGS.num_epochs)
# Train/test split for labels
shuffled_indices = np.random.permutation(len(labeled_data[:,0]))
dev_sample_index = int(FLAGS.dev_sample_percentage * float(len(shuffled_indices)))
test_indices = shuffled_indices[:dev_sample_index]
train_indices = shuffled_indices[dev_sample_index:]
train_label = labeled_data[train_indices,:]
test_labeled = labeled_data[test_indices,:]
# Generate training batches for labeled images
batches_labeled = preprocessing.batch_iter(train_label, FLAGS.batch_size, FLAGS.num_epochs)
##### PRINTING THE USED PARAMETERS TO THE LOG FILE #####
print("\nParameters:")
for attr, value in sorted(FLAGS.__flags.items()):