def vad_test(m_eval, sess_eval, batch_size_eval, eval_file_dir, norm_dir,
data_len, eval_type):
eval_input_dir = eval_file_dir
eval_output_dir = eval_file_dir + '/Labels'
pad_size = batch_size_eval - data_len % batch_size_eval
if eval_type != 2:
eval_data_set = dr.DataReader(eval_input_dir,
eval_output_dir,
norm_dir,
w=19,
u=9,
name="eval")
else:
eval_data_set = dnn_dr.DataReader(eval_input_dir,
eval_output_dir,
norm_dir,
w=19,
u=9,
name="eval")
final_softout, final_label = evaluation(m_eval, eval_data_set, sess_eval,
batch_size_eval, eval_type)
return final_softout, final_label
def do_test(fname_model, test_file_dir, norm_dir, data_len, is_default,
model_type):
eval_input_dir = test_file_dir
eval_output_dir = test_file_dir + '/Labels'
graph = load_graph(fname_model)
w = 19
u = 9
# [print(n.name) for n in graph.as_graph_def().node]
# for op in graph.get_operations():
# print(op.name)
final_softout = []
final_label = []
if model_type == 0: # acam
import data_reader_bDNN_v2 as dr
print(os.path.abspath('./configure/ACAM'))
sys.path.insert(0, os.path.abspath('./configure/ACAM'))
import config as cg
if is_default:
w = 19
u = 9
valid_batch_size = 4096
else:
w = cg.w
u = cg.u
valid_batch_size = cg.batch_size
valid_data_set = dr.DataReader(eval_input_dir,
eval_output_dir,
norm_dir,
w=w,
u=u,
name="eval")
node_inputs = graph.get_tensor_by_name('prefix/model_1/inputs:0')
node_labels = graph.get_tensor_by_name('prefix/model_1/labels:0')
node_keep_probability = graph.get_tensor_by_name(
'prefix/model_1/keep_probabilty:0')
node_logits = graph.get_tensor_by_name('prefix/model_1/logits:0')
node_raw_labels = graph.get_tensor_by_name(
'prefix/model_1/raw_labels:0')
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
feed_dict = {
node_inputs: valid_inputs,
node_labels: valid_labels,
node_keep_probability: 1
}
if valid_data_set.eof_checker():
final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
final_label = np.reshape(np.asarray(final_label), [-1, 1])
valid_data_set.reader_initialize()
# print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
break
with tf.Session(graph=graph) as sess:
logits, raw_labels = sess.run([node_logits, node_raw_labels],
feed_dict=feed_dict)
soft_pred = bdnn_prediction(valid_batch_size,
logits,
threshold=0.6,
w=w,
u=u)[1]
raw_labels = raw_labels.reshape((-1, 1))
final_softout.append(soft_pred)
final_label.append(raw_labels)
# if valid_data_set.eof_checker():
# final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
# final_label = np.reshape(np.asarray(final_label), [-1, 1])
# valid_data_set.reader_initialize()
# # print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
# break
return final_softout[0:data_len, :], final_label[0:data_len, :]
if model_type == 1: # bdnn
import data_reader_bDNN_v2 as dr
print(os.path.abspath('./configure/bDNN'))
sys.path.insert(0, os.path.abspath('./configure/bDNN'))
import config as cg
if is_default:
w = 19
u = 9
valid_batch_size = 4096
else:
w = cg.w
u = cg.u
valid_batch_size = cg.batch_size
valid_data_set = dr.DataReader(
eval_input_dir, eval_output_dir, norm_dir, w=w, u=u,
name="eval") # training data reader initialization
node_inputs = graph.get_tensor_by_name('prefix/model_1/inputs:0')
node_labels = graph.get_tensor_by_name('prefix/model_1/labels:0')
node_keep_probability = graph.get_tensor_by_name(
'prefix/model_1/keep_probabilty:0')
node_logits = graph.get_tensor_by_name('prefix/model_1/logits:0')
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
feed_dict = {
node_inputs: valid_inputs,
node_labels: valid_labels,
node_keep_probability: 1
}
if valid_data_set.eof_checker():
final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
final_label = np.reshape(np.asarray(final_label), [-1, 1])
valid_data_set.reader_initialize()
# print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
break
with tf.Session(graph=graph) as sess:
logits, labels = sess.run([node_logits, node_labels],
feed_dict=feed_dict)
soft_pred = bdnn_prediction(valid_batch_size,
logits,
threshold=0.6,
w=w,
u=u)[1]
raw_indx = int(np.floor(labels.shape[1] / 2))
raw_labels = labels[:, raw_indx]
raw_labels = raw_labels.reshape((-1, 1))
final_softout.append(soft_pred)
final_label.append(raw_labels)
return final_softout[0:data_len, :], final_label[0:data_len, :]
if model_type == 2: # dnn
import data_reader_DNN_v2 as dnn_dr
print(os.path.abspath('./configure/DNN'))
sys.path.insert(0, os.path.abspath('./configure/DNN'))
import config as cg
if is_default:
w = 19
u = 9
valid_batch_size = 4096
else:
w = cg.w
u = cg.u
valid_batch_size = cg.batch_size
valid_data_set = dnn_dr.DataReader(eval_input_dir,
eval_output_dir,
norm_dir,
w=w,
u=u,
name="eval")
node_inputs = graph.get_tensor_by_name('prefix/model_1/inputs:0')
node_labels = graph.get_tensor_by_name('prefix/model_1/labels:0')
node_keep_probability = graph.get_tensor_by_name(
'prefix/model_1/keep_probabilty:0')
node_softpred = graph.get_tensor_by_name('prefix/model_1/soft_pred:0')
node_raw_labels = graph.get_tensor_by_name(
'prefix/model_1/raw_labels:0')
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
one_hot_labels = valid_labels.reshape((-1, 1))
one_hot_labels = utils.dense_to_one_hot(one_hot_labels,
num_classes=2)
feed_dict = {
node_inputs: valid_inputs,
node_labels: one_hot_labels,
node_keep_probability: 1
}
if valid_data_set.eof_checker():
final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
final_label = np.reshape(np.asarray(final_label), [-1, 1])
valid_data_set.reader_initialize()
# print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
break
print("aa")
session_conf = tf.ConfigProto(device_count={
'CPU': 1,
'GPU': 0
},
allow_soft_placement=True,
log_device_placement=False)
with tf.Session(graph=graph, config=session_conf) as sess:
soft_pred, raw_labels = sess.run(
[node_softpred, node_raw_labels], feed_dict=feed_dict)
raw_labels = raw_labels.reshape((-1, 1))
final_softout.append(soft_pred)
final_label.append(raw_labels)
return final_softout[0:data_len, :], final_label[0:data_len, :]
if model_type == 3: # lstm
import data_reader_RNN as rnn_dr
print(os.path.abspath('./configure/LSTM'))
sys.path.insert(0, os.path.abspath('./configure/LSTM'))
import config as cg
if is_default:
target_delay = 5
seq_size = 20
batch_num = 200
valid_batch_size = seq_size * batch_num
else:
target_delay = cg.target_delay
seq_size = cg.seq_len
batch_num = cg.num_batches
valid_batch_size = seq_size * batch_num
valid_data_set = rnn_dr.DataReader(eval_input_dir,
eval_output_dir,
norm_dir,
target_delay=target_delay,
name="eval")
node_inputs = graph.get_tensor_by_name('prefix/model_1/inputs:0')
node_labels = graph.get_tensor_by_name('prefix/model_1/labels:0')
node_keep_probability = graph.get_tensor_by_name(
'prefix/model_1/keep_probabilty:0')
node_softpred = graph.get_tensor_by_name('prefix/model_1/soft_pred:0')
node_raw_labels = graph.get_tensor_by_name(
'prefix/model_1/raw_labels:0')
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
one_hot_labels = valid_labels.reshape((-1, 1))
one_hot_labels = utils.dense_to_one_hot(one_hot_labels,
num_classes=2)
feed_dict = {
node_inputs: valid_inputs,
node_labels: one_hot_labels,
node_keep_probability: 1
}
if valid_data_set.eof_checker():
final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
final_label = np.reshape(np.asarray(final_label), [-1, 1])
valid_data_set.reader_initialize()
# print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
break
with tf.Session(graph=graph) as sess:
soft_pred, raw_labels = sess.run(
[node_softpred, node_raw_labels], feed_dict=feed_dict)
raw_labels = raw_labels.reshape((-1, 1))
final_softout.append(soft_pred)
final_label.append(raw_labels)
# if valid_data_set.eof_checker():
# final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
# final_label = np.reshape(np.asarray(final_label), [-1, 1])
# valid_data_set.reader_initialize()
# # print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
# break
return final_softout[0:data_len, :], final_label[0:data_len, :]
def main(prj_dir=None, model=None, mode=None):
# Configuration Part #
if mode is 'train':
import path_setting as ps
set_path = ps.PathSetting(prj_dir, model)
logs_dir = initial_logs_dir = set_path.logs_dir
input_dir = set_path.input_dir
output_dir = set_path.output_dir
norm_dir = set_path.norm_dir
valid_file_dir = set_path.valid_file_dir
sys.path.insert(0, prj_dir + '/configure/ACAM')
import config as cg
global initLr, dropout_rate, max_epoch, batch_size, valid_batch_size
initLr = cg.lr
dropout_rate = cg.dropout_rate
max_epoch = cg.max_epoch
batch_size = valid_batch_size = cg.batch_size
global w, u
w = cg.w
u = cg.u
global bdnn_winlen, bdnn_inputsize, bdnn_outputsize
bdnn_winlen = (((w - 1) / u) * 2) + 3
bdnn_inputsize = int(bdnn_winlen * num_features)
bdnn_outputsize = int(bdnn_winlen)
global glimpse_hidden, bp_hidden, glimpse_out, bp_out, nGlimpses,\
lstm_cell_size, action_hidden_1, action_hidden_2
glimpse_hidden = cg.glimpse_hidden
bp_hidden = cg.bp_hidden
glimpse_out = bp_out = cg.glimpse_out
nGlimpses = cg.nGlimpse # 7
lstm_cell_size = cg.lstm_cell_size
action_hidden_1 = cg.action_hidden_1 # default : 256
action_hidden_2 = cg.action_hidden_2 # default : 256
# Graph Part #
mean_acc_list = []
var_acc_list = []
print('Mode : ' + mode)
print("Graph initialization...")
with tf.device(device):
with tf.variable_scope("model", reuse=None):
m_train = Model(batch_size=batch_size,
reuse=None,
is_training=True)
# m_train(batch_size)
with tf.device(device):
with tf.variable_scope("model", reuse=True):
m_valid = Model(batch_size=valid_batch_size,
reuse=True,
is_training=False)
print("Done")
# Summary Part #
print("Setting up summary op...")
summary_ph = tf.placeholder(dtype=tf.float32)
with tf.variable_scope("Training_procedure"):
cost_summary_op = tf.summary.scalar("cost", summary_ph)
accuracy_summary_op = tf.summary.scalar("accuracy", summary_ph)
# train_summary_writer = tf.summary.FileWriter(logs_dir + '/train/', max_queue=4)
# valid_summary_writer = tf.summary.FileWriter(logs_dir + '/valid/', max_queue=4)
# summary_dic = summary_generation(valid_file_dir)
print("Done")
# Model Save Part #
print("Setting up Saver...")
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(initial_logs_dir)
print("Done")
# Session Part #
sess_config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)
sess_config.gpu_options.allow_growth = True
sess = tf.Session(config=sess_config)
if mode is 'train':
train_summary_writer = tf.summary.FileWriter(logs_dir + '/train/',
sess.graph,
max_queue=2)
valid_summary_writer = tf.summary.FileWriter(logs_dir + '/valid/',
max_queue=2)
if ckpt and ckpt.model_checkpoint_path: # model restore
print("Model restored...")
print(initial_logs_dir + ckpt_name)
if mode is 'train':
saver.restore(sess, ckpt.model_checkpoint_path)
else:
saver.restore(sess, initial_logs_dir + ckpt_name)
saver.save(sess, initial_logs_dir + "/model_ACAM.ckpt",
0) # model save
print("Done")
else:
sess.run(tf.global_variables_initializer()
) # if the checkpoint doesn't exist, do initialization
if mode is 'train':
train_data_set = dr.DataReader(
input_dir, output_dir, norm_dir, w=w, u=u,
name="train") # training data reader initialization
if mode is 'train':
for itr in range(max_epoch):
start_time = time.time()
train_inputs, train_labels = train_data_set.next_batch(batch_size)
feed_dict = {
m_train.inputs: train_inputs,
m_train.labels: train_labels,
m_train.keep_probability: dropout_rate
}
sess.run(m_train.train_op, feed_dict=feed_dict)
if itr % 10 == 0 and itr >= 0:
train_cost, train_reward, train_avg_b, train_rminusb, train_p_bps, train_lr \
= sess.run([m_train.cost, m_train.reward, m_train.avg_b, m_train.rminusb, m_train.p_bps,
m_train.print_lr]
, feed_dict=feed_dict)
duration = time.time() - start_time
print(
"Step: %d, cost: %.4f, accuracy: %4.4f, b: %4.4f, R-b: %4.4f, p_bps: %4.4f, lr: %7.6f (%.3f sec)"
% (itr, train_cost, train_reward, train_avg_b,
train_rminusb, train_p_bps, train_lr, duration))
train_cost_summary_str = sess.run(
cost_summary_op, feed_dict={summary_ph: train_cost})
train_accuracy_summary_str = sess.run(
accuracy_summary_op, feed_dict={summary_ph: train_reward})
train_summary_writer.add_summary(
train_cost_summary_str,
itr) # write the train phase summary to event files
train_summary_writer.add_summary(train_accuracy_summary_str,
itr)
# if train_data_set.eof_checker():
# if itr % val_freq == 0 and itr >= val_start_step:
if itr % 50 == 0 and itr > 0:
saver.save(sess, logs_dir + "/model.ckpt", itr) # model save
print('validation start!')
valid_accuracy, valid_cost = \
utils.do_validation(m_valid, sess, valid_file_dir, norm_dir,
type='ACAM')
print("valid_cost: %.4f, valid_accuracy=%4.4f" %
(valid_cost, valid_accuracy * 100))
valid_cost_summary_str = sess.run(
cost_summary_op, feed_dict={summary_ph: valid_cost})
valid_accuracy_summary_str = sess.run(
accuracy_summary_op,
feed_dict={summary_ph: valid_accuracy})
valid_summary_writer.add_summary(
valid_cost_summary_str,
itr) # write the train phase summary to event files
valid_summary_writer.add_summary(valid_accuracy_summary_str,
itr)
# mean_accuracy, var_accuracy = full_evaluation(m_valid, sess, valid_batch_size, valid_file_dir, valid_summary_writer, summary_dic, itr)
# if mean_accuracy >= 0.991:
#
# print('model was saved!')
# model_name = '/model' + str(int(mean_accuracy * 1e4)) + 'and'\
# + str(int(var_accuracy * 1e5)) + '.ckpt'
# saver.save(sess, save_dir + model_name, itr)
# mean_acc_list.append(mean_accuracy)
# var_acc_list.append(var_accuracy)
# train_data_set.initialize()
elif mode == 'test':
final_softout, final_label = utils.vad_test(m_valid, sess,
valid_batch_size,
test_file_dir, norm_dir,
data_len, eval_type)
# if data_len is None:
# return final_softout, final_label
# else:
# final_softout = final_softout[0:data_len, :]
# final_label = final_label[0:data_len, :]
# fpr, tpr, thresholds = metrics.roc_curve(final_label, final_softout, pos_label=1)
# eval_auc = metrics.auc(fpr, tpr)
# print(eval_auc)
# full_evaluation(m_valid, sess, valid_batch_size, test_file_dir, valid_summary_writer, summary_dic, 0)
# if visualization:
# global attention
# attention = np.asarray(attention)
# sio.savemat('attention.mat', {'attention' : attention})
# subprocess.call(['./visualize.sh'])
if data_len is None:
return final_softout, final_label
else:
return final_softout[0:data_len, :], final_label[0:data_len, :]
def full_evaluation(m_eval, sess_eval, batch_size_eval, eval_file_dir,
summary_writer, summary_dic, itr):
mean_cost = []
mean_accuracy = []
mean_auc = []
print("-------- Performance for each of noise types --------")
noise_list = os.listdir(eval_file_dir)
noise_list = sorted(noise_list)
summary_ph = summary_dic["summary_ph"]
for i in range(len(noise_list)):
noise_name = '/' + noise_list[i]
eval_input_dir = eval_file_dir + noise_name
eval_output_dir = eval_file_dir + noise_name + '/Labels'
eval_data_set = dr.DataReader(eval_input_dir,
eval_output_dir,
norm_dir,
w=w,
u=u,
name="eval")
eval_cost, eval_accuracy, eval_list, eval_auc, eval_auc_list = evaluation(
m_eval, eval_data_set, sess_eval, batch_size_eval)
print("--noise type : " + noise_list[i])
print("cost: %.4f, accuracy across all SNRs: %.4f" %
(eval_cost, eval_accuracy * 100))
print('accuracy wrt SNR:')
print('SNR_-5 : %.4f, SNR_0 : %.4f, SNR_5 : %.4f, SNR_10 : %.4f' %
(eval_list[0] * 100, eval_list[1] * 100, eval_list[2] * 100,
eval_list[3] * 100))
print('AUC wrt SNR:')
print('SNR_-5 : %.4f, SNR_0 : %.4f, SNR_5 : %.4f, SNR_10 : %.4f' %
(eval_auc_list[0] * 100, eval_auc_list[1] * 100,
eval_auc_list[2] * 100, eval_auc_list[3] * 100))
print('')
eval_summary_list = [eval_cost] + eval_list + [eval_accuracy]
for j, summary_name in enumerate(summary_list):
summary_str = sess_eval.run(
summary_dic[noise_list[i] + "_" + summary_name],
feed_dict={summary_ph: eval_summary_list[j]})
summary_writer.add_summary(summary_str, itr)
mean_cost.append(eval_cost)
mean_accuracy.append(eval_accuracy)
mean_auc.append(eval_auc)
mean_cost = np.mean(np.asarray(mean_cost))
var_accuracy = np.var(np.asarray(mean_accuracy))
mean_accuracy = np.mean(np.asarray(mean_accuracy))
mean_auc = np.mean(np.asarray(mean_auc))
summary_writer.add_summary(
sess_eval.run(summary_dic["cost_across_all_noise_types"],
feed_dict={summary_ph: mean_cost}), itr)
summary_writer.add_summary(
sess_eval.run(summary_dic["accuracy_across_all_noise_types"],
feed_dict={summary_ph: mean_accuracy}), itr)
summary_writer.add_summary(
sess_eval.run(summary_dic["variance_across_all_noise_types"],
feed_dict={summary_ph: var_accuracy}), itr)
print("-------- Performance across all of noise types --------")
print("cost : %.4f" % mean_cost)
print("******* averaged accuracy across all noise_types : %.4f *******" %
(mean_accuracy * 100))
print("******* averaged auc across all noise_types : %.7f *******" %
(mean_auc * 100))
print(
"******* variance of accuracies across all noise_types : %6.6f *******"
% var_accuracy)
return mean_auc, var_accuracy
def main(save_dir, prj_dir=None, model=None, mode=None, dev="/gpu:2"):
# Configuration Part #
# os.environ["CUDA_VISIBLE_DEVICES"] = '3'
device = dev
os.environ["CUDA_VISIBLE_DEVICES"] = device[-1]
if mode is 'train':
import path_setting as ps
set_path = ps.PathSetting(prj_dir, model, save_dir)
logs_dir = initial_logs_dir = set_path.logs_dir
input_dir = set_path.input_dir
output_dir = set_path.output_dir
norm_dir = set_path.norm_dir
valid_file_dir = set_path.valid_file_dir
sys.path.insert(0, prj_dir + '/configure/ACAM')
import config as cg
global initLr, dropout_rate, max_epoch, batch_size, valid_batch_size
initLr = cg.lr
dropout_rate = cg.dropout_rate
max_epoch = cg.max_epoch
batch_size = valid_batch_size = cg.batch_size
global w, u
w = cg.w
u = cg.u
global bdnn_winlen, bdnn_inputsize, bdnn_outputsize
bdnn_winlen = (((w - 1) / u) * 2) + 3
bdnn_inputsize = int(bdnn_winlen * num_features)
bdnn_outputsize = int(bdnn_winlen)
global glimpse_hidden, bp_hidden, glimpse_out, bp_out, nGlimpses,\
lstm_cell_size, action_hidden_1, action_hidden_2
glimpse_hidden = cg.glimpse_hidden
bp_hidden = cg.bp_hidden
glimpse_out = bp_out = cg.glimpse_out
nGlimpses = cg.nGlimpse # 7
lstm_cell_size = cg.lstm_cell_size
action_hidden_1 = cg.action_hidden_1 # default : 256
action_hidden_2 = cg.action_hidden_2 # default : 256
# Graph Part #
mean_acc_list = []
var_acc_list = []
print('Mode : ' + mode)
print("Graph initialization...")
with tf.device(device):
with tf.variable_scope("model", reuse=None):
m_train = Model(batch_size=batch_size,
reuse=None,
is_training=True)
# m_train(batch_size)
with tf.device(device):
with tf.variable_scope("model", reuse=True):
m_valid = Model(batch_size=valid_batch_size,
reuse=True,
is_training=False)
print("Done")
# Summary Part #
print("Setting up summary op...")
summary_ph = tf.placeholder(dtype=tf.float32)
with tf.variable_scope("Training_procedure"):
cost_summary_op = tf.summary.scalar("cost", summary_ph)
accuracy_summary_op = tf.summary.scalar("accuracy", summary_ph)
# train_summary_writer = tf.summary.FileWriter(logs_dir + '/train/', max_queue=4)
# valid_summary_writer = tf.summary.FileWriter(logs_dir + '/valid/', max_queue=4)
# summary_dic = summary_generation(valid_file_dir)
print("Done")
# Model Save Part #
print("Setting up Saver...")
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(initial_logs_dir)
print("Done")
# Session Part #
sess_config = tf.ConfigProto(allow_soft_placement=True,
log_device_placement=False)
sess_config.gpu_options.allow_growth = True
sess = tf.Session(config=sess_config)
if mode is 'train':
train_summary_writer = tf.summary.FileWriter(logs_dir + '/train/',
sess.graph,
max_queue=2)
valid_summary_writer = tf.summary.FileWriter(logs_dir + '/valid/',
max_queue=2)
if ckpt and ckpt.model_checkpoint_path: # model restore
print("Model restored...")
print(initial_logs_dir + ckpt_name)
if mode is 'train':
saver.restore(sess, ckpt.model_checkpoint_path)
else:
saver.restore(sess, initial_logs_dir + ckpt_name)
saver.save(sess, initial_logs_dir + "/model_ACAM.ckpt",
0) # model save
print("Done")
else:
sess.run(tf.global_variables_initializer()
) # if the checkpoint doesn't exist, do initialization
if mode is 'train':
train_data_set = dr.DataReader(
input_dir, output_dir, norm_dir, w=w, u=u,
name="train") # training data reader initialization
if mode is 'train':
file_len = train_data_set.get_file_len()
MAX_STEP = max_epoch * file_len
print(get_num_params())
for itr in range(MAX_STEP):
start_time = time.time()
train_inputs, train_labels = train_data_set.next_batch(batch_size)
feed_dict = {
m_train.inputs: train_inputs,
m_train.labels: train_labels,
m_train.keep_probability: dropout_rate
}
sess.run(m_train.train_op, feed_dict=feed_dict)
if itr % 100 == 0 and itr >= 0:
train_cost, train_reward, train_avg_b, train_rminusb, train_p_bps, train_lr, train_res \
= sess.run([m_train.cost, m_train.reward, m_train.avg_b, m_train.rminusb, m_train.p_bps,
m_train.print_lr,m_train.result]
, feed_dict=feed_dict)
duration = time.time() - start_time
print(
"Step: %d, cost: %.4f, accuracy: %4.4f, b: %4.4f, R-b: %4.4f, p_bps: %4.4f, lr: %7.6f (%.3f sec)"
% (itr, train_cost, train_reward, train_avg_b,
train_rminusb, train_p_bps, train_lr, duration))
# np.save('pre/'+train_data_set.get_cur_file_name().split('/')[-1], train_res)
train_cost_summary_str = sess.run(
cost_summary_op, feed_dict={summary_ph: train_cost})
train_accuracy_summary_str = sess.run(
accuracy_summary_op, feed_dict={summary_ph: train_reward})
train_summary_writer.add_summary(train_cost_summary_str, itr)
train_summary_writer.add_summary(train_accuracy_summary_str,
itr)
if itr % file_len == 0 and itr > 0:
saver.save(sess, logs_dir + "/model.ckpt", itr) # model save
print('validation start!')
valid_accuracy, valid_cost = \
utils.do_validation(m_valid, sess, valid_file_dir, norm_dir,
type='ACAM')
print("valid_cost: %.4f, valid_accuracy=%4.4f" %
(valid_cost, valid_accuracy * 100))
valid_cost_summary_str = sess.run(
cost_summary_op, feed_dict={summary_ph: valid_cost})
valid_accuracy_summary_str = sess.run(
accuracy_summary_op,
feed_dict={summary_ph: valid_accuracy})
valid_summary_writer.add_summary(
valid_cost_summary_str,
itr) # write the train phase summary to event files
valid_summary_writer.add_summary(valid_accuracy_summary_str,
itr)
gs.freeze_graph(prj_dir + '/logs/ACAM',
prj_dir + '/saved_model/graph/ACAM',
'model_1/logits,model_1/raw_labels')
elif mode == 'test':
final_softout, final_label = utils.vad_test(m_valid, sess,
valid_batch_size,
test_file_dir, norm_dir,
data_len, eval_type)
if data_len is None:
return final_softout, final_label
else:
return final_softout[0:data_len, :], final_label[0:data_len, :]
def do_test(fname_model, test_file_dir, norm_dir, prj_dir, is_default,
model_type):
eval_input_dir = test_file_dir
graph = load_graph(fname_model)
w = 19
u = 9
# [print(n.name) for n in graph.as_graph_def().node]
# for op in graph.get_operations():
# print(op.name)
final_softout = []
final_label = []
if model_type == 0: # acam
import data_reader_bDNN_v2 as dr
print(prj_dir + '/configure/ACAM')
sys.path.insert(0, os.path.abspath(prj_dir + 'ACAM'))
# import config as cg
#
# if is_default:
# w = 19
# u = 9
# valid_batch_size = 4096
# else:
# w = cg.w
# u = cg.u
# valid_batch_size = cg.batch_size
valid_data_set = dr.DataReader(eval_input_dir + "/train/feature_mrcg",
eval_input_dir + "/train/label",
norm_dir,
w=w,
u=u,
name="eval")
node_inputs = graph.get_tensor_by_name('prefix/model_1/inputs:0')
node_labels = graph.get_tensor_by_name('prefix/model_1/labels:0')
node_keep_probability = graph.get_tensor_by_name(
'prefix/model_1/keep_probabilty:0')
node_logits = graph.get_tensor_by_name('prefix/model_1/logits:0')
node_raw_labels = graph.get_tensor_by_name(
'prefix/model_1/raw_labels:0')
acc_sum = 0
file_num = 0
acc_sum = 0
tp_sum = 0
tn_sum = 0
fp_sum = 0
fn_sum = 0
frame_num = 0
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(7)
feed_dict = {
node_inputs: valid_inputs,
node_labels: valid_labels,
node_keep_probability: 1
}
if valid_data_set.eof_checker():
final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
final_label = np.reshape(np.asarray(final_label), [-1, 1])
valid_data_set.reader_initialize()
# print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
break
with tf.Session(graph=graph) as sess:
logits, raw_labels = sess.run([node_logits, node_raw_labels],
feed_dict=feed_dict)
raw_labels = raw_labels.reshape((-1, 1))
soft_pred = bdnn_prediction(len(raw_labels),
logits,
threshold=0.6,
w=w,
u=u)[0]
eql = (raw_labels == soft_pred)
acc = np.mean(eql)
acc_sum += acc
file_num += 1
frame_s = len(raw_labels)
frame_num += len(raw_labels)
tn = 0
tp = 0
fp = 0
fn = 0
for i in range(len(soft_pred)):
if soft_pred[i] == 0 and raw_labels[i] == 0:
tn += 1
elif soft_pred[i] == 0 and raw_labels[i] == 1:
fn += 1
elif soft_pred[i] == 1 and raw_labels[i] == 0:
fp += 1
elif soft_pred[i] == 1 and raw_labels[i] == 1:
tp += 1
final_softout.append(soft_pred)
final_label.append(raw_labels)
tn_sum += tn
tp_sum += tp
fn_sum += fn
fp_sum += fp
# if acc <= 0.7:
# file_name =valid_data_set.get_cur_file_name().split('/')[-1]
# obj_name = file_name.split('.')[0]
# wav_path = "/mnt/E_DRIVE/Lipr_with_label/train/low"
# shutil.copy("/mnt/E_DRIVE/Lipr_with_label/train/" + obj_name + '.wav', wav_path)
# np.save(os.path.join("/mnt/E_DRIVE/Lipr_with_label/train/low", obj_name + '.label.npy'),
# original_label(soft_pred, "/mnt/E_DRIVE/Lipr_with_label/train/" + obj_name + '.wav'))
print(" train_accuracy=%4.4f" % (acc * 100))
# print("path is " + valid_data_set.get_cur_file_name())
print(
"true_positive: %f, false positive: %f, true negative: %f, false negative: %f"
% (tp / frame_s, fp / frame_s, tn / frame_s, fn / frame_s))
# print("accuracy: %f, true_positive: %f, false positive: %f, true negative: %f, false negative: %f" % (
# acc, tp / frame_s, fp / frame_s, tn / frame_s, fn / frame_s))
print("total accuracy: " + str(acc_sum / file_num))
print(
"total: true_positive: %f, false positive: %f, true negative: %f, false negative: %f"
% (tp_sum / frame_num, fp_sum / frame_num, tn_sum / frame_num,
fn_sum / frame_num))
return final_softout[:, :], final_label[:, :]
if model_type == 1: # bdnn
import data_reader_bDNN_v2 as dr
print(prj_dir + '/configure/bDNN')
sys.path.insert(0, os.path.abspath(prj_dir + 'bDNN'))
import config as cg
if is_default:
w = 19
u = 9
valid_batch_size = 4096
else:
w = cg.w
u = cg.u
valid_batch_size = cg.batch_size
valid_data_set = dr.DataReader(
eval_input_dir, eval_output_dir, norm_dir, w=w, u=u,
name="eval") # training data reader initialization
node_inputs = graph.get_tensor_by_name('prefix/model_1/inputs:0')
node_labels = graph.get_tensor_by_name('prefix/model_1/labels:0')
node_keep_probability = graph.get_tensor_by_name(
'prefix/model_1/keep_probabilty:0')
node_logits = graph.get_tensor_by_name('prefix/model_1/logits:0')
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
feed_dict = {
node_inputs: valid_inputs,
node_labels: valid_labels,
node_keep_probability: 1
}
if valid_data_set.eof_checker():
final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
final_label = np.reshape(np.asarray(final_label), [-1, 1])
valid_data_set.reader_initialize()
# print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
break
with tf.Session(graph=graph) as sess:
logits, labels = sess.run([node_logits, node_labels],
feed_dict=feed_dict)
soft_pred = bdnn_prediction(valid_batch_size,
logits,
threshold=0.6,
w=w,
u=u)[1]
raw_indx = int(np.floor(labels.shape[1] / 2))
raw_labels = labels[:, raw_indx]
raw_labels = raw_labels.reshape((-1, 1))
final_softout.append(soft_pred)
final_label.append(raw_labels)
return final_softout[0:data_len, :], final_label[0:data_len, :]
if model_type == 2: # dnn
import data_reader_DNN_v2 as dnn_dr
print(prj_dir + '/configure/DNN')
sys.path.insert(0, os.path.abspath(prj_dir + 'DNN'))
import config as cg
if is_default:
w = 19
u = 9
valid_batch_size = 4096
else:
w = cg.w
u = cg.u
valid_batch_size = cg.batch_size
valid_data_set = dnn_dr.DataReader(eval_input_dir,
eval_output_dir,
norm_dir,
w=w,
u=u,
name="eval")
node_inputs = graph.get_tensor_by_name('prefix/model_1/inputs:0')
node_labels = graph.get_tensor_by_name('prefix/model_1/labels:0')
node_keep_probability = graph.get_tensor_by_name(
'prefix/model_1/keep_probabilty:0')
node_softpred = graph.get_tensor_by_name('prefix/model_1/soft_pred:0')
node_raw_labels = graph.get_tensor_by_name(
'prefix/model_1/raw_labels:0')
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
one_hot_labels = valid_labels.reshape((-1, 1))
one_hot_labels = utils.dense_to_one_hot(one_hot_labels,
num_classes=2)
feed_dict = {
node_inputs: valid_inputs,
node_labels: one_hot_labels,
node_keep_probability: 1
}
if valid_data_set.eof_checker():
final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
final_label = np.reshape(np.asarray(final_label), [-1, 1])
valid_data_set.reader_initialize()
# print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
break
with tf.Session(graph=graph) as sess:
soft_pred, raw_labels = sess.run(
[node_softpred, node_raw_labels], feed_dict=feed_dict)
raw_labels = raw_labels.reshape((-1, 1))
final_softout.append(soft_pred)
final_label.append(raw_labels)
return final_softout[0:data_len, :], final_label[0:data_len, :]
if model_type == 3: # lstm
import data_reader_RNN as rnn_dr
print(prj_dir + '/configure/LSTM')
sys.path.insert(0, os.path.abspath(prj_dir + 'LSTM'))
import config as cg
if is_default:
target_delay = 5
seq_size = 20
batch_num = 200
valid_batch_size = seq_size * batch_num
else:
target_delay = cg.target_delay
seq_size = cg.seq_len
valid_batch_size = seq_size
valid_data_set = rnn_dr.DataReader(eval_input_dir +
"/train/feature_mrcg",
eval_input_dir + '/train/label',
norm_dir,
target_delay=target_delay,
name="eval")
node_inputs = graph.get_tensor_by_name('prefix/model_1/inputs:0')
node_labels = graph.get_tensor_by_name('prefix/model_1/labels:0')
node_keep_probability = graph.get_tensor_by_name(
'prefix/model_1/keep_probabilty:0')
node_softpred = graph.get_tensor_by_name('prefix/model_1/soft_pred:0')
node_raw_labels = graph.get_tensor_by_name(
'prefix/model_1/raw_labels:0')
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
one_hot_labels = valid_labels.reshape((-1, 1))
one_hot_labels = utils.dense_to_one_hot(one_hot_labels,
num_classes=2)
feed_dict = {
node_inputs: valid_inputs,
node_labels: one_hot_labels,
node_keep_probability: 1
}
if valid_data_set.eof_checker():
# final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
# final_label = np.reshape(np.asarray(final_label), [-1, 1])
valid_data_set.reader_initialize()
break
with tf.Session(graph=graph) as sess:
soft_pred, raw_labels = sess.run(
[node_softpred, node_raw_labels], feed_dict=feed_dict)
raw_labels = raw_labels.reshape((-1, 1))
final_softout.append(soft_pred)
final_label.append(raw_labels)
# if valid_data_set.eof_checker():
# final_softout = np.reshape(np.asarray(final_softout), [-1, 1])
# final_label = np.reshape(np.asarray(final_label), [-1, 1])
# valid_data_set.reader_initialize()
# # print('Valid data reader was initialized!') # initialize eof flag & num_file & start index
# break
return final_softout, final_label
def do_validation(m_valid, sess, valid_file_dir, norm_dir, type='DNN'):
# dataset reader setting #
# sys.path.insert(0, prj_dir + '/configure/DNN')
if type is 'DNN':
sys.path.insert(0, os.path.abspath('../../configure/DNN'))
import config as cg
valid_batch_size = cg.batch_size
valid_data_set = dnn_dr.DataReader(valid_file_dir,
valid_file_dir + '/Labels',
norm_dir,
w=cg.w,
u=cg.u,
name="eval")
avg_valid_accuracy = 0.
avg_valid_cost = 0.
itr_sum = 0.
accuracy_list = [0 for i in range(valid_data_set._file_len)]
cost_list = [0 for i in range(valid_data_set._file_len)]
itr_file = 0
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
if valid_data_set.file_change_checker():
# print(itr_file)
accuracy_list[itr_file] = avg_valid_accuracy / itr_sum
cost_list[itr_file] = avg_valid_cost / itr_sum
avg_valid_cost = 0.
avg_valid_accuracy = 0.
itr_sum = 0
itr_file += 1
valid_data_set.file_change_initialize()
if valid_data_set.eof_checker():
valid_data_set.reader_initialize()
print('Valid data reader was initialized!'
) # initialize eof flag & num_file & start index
break
one_hot_labels = valid_labels.reshape((-1, 1))
one_hot_labels = dense_to_one_hot(one_hot_labels, num_classes=2)
feed_dict = {
m_valid.inputs: valid_inputs,
m_valid.labels: one_hot_labels,
m_valid.keep_probability: 1
}
# valid_cost, valid_softpred, valid_raw_labels\
# = sess.run([m_valid.cost, m_valid.softpred, m_valid.raw_labels], feed_dict=feed_dict)
#
# fpr, tpr, thresholds = metrics.roc_curve(valid_raw_labels, valid_softpred, pos_label=1)
# valid_auc = metrics.auc(fpr, tpr)
valid_cost, valid_accuracy = sess.run(
[m_valid.cost, m_valid.accuracy], feed_dict=feed_dict)
avg_valid_accuracy += valid_accuracy
avg_valid_cost += valid_cost
itr_sum += 1
total_avg_valid_accuracy = np.asscalar(
np.mean(np.asarray(accuracy_list)))
total_avg_valid_cost = np.asscalar(np.mean(np.asarray(cost_list)))
elif type is 'bDNN':
sys.path.insert(0, os.path.abspath('../../configure/bDNN'))
import config as cg
valid_batch_size = cg.batch_size
valid_data_set = dr.DataReader(valid_file_dir,
valid_file_dir + '/Labels',
norm_dir,
w=cg.w,
u=cg.u,
name="eval")
avg_valid_accuracy = 0.
avg_valid_cost = 0.
itr_sum = 0.
accuracy_list = [0 for i in range(valid_data_set._file_len)]
cost_list = [0 for i in range(valid_data_set._file_len)]
itr_file = 0
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
if valid_data_set.file_change_checker():
# print(itr_file)
accuracy_list[itr_file] = avg_valid_accuracy / itr_sum
cost_list[itr_file] = avg_valid_cost / itr_sum
avg_valid_cost = 0.
avg_valid_accuracy = 0.
itr_sum = 0
itr_file += 1
valid_data_set.file_change_initialize()
if valid_data_set.eof_checker():
valid_data_set.reader_initialize()
print('Valid data reader was initialized!'
) # initialize eof flag & num_file & start index
break
feed_dict = {
m_valid.inputs: valid_inputs,
m_valid.labels: valid_labels,
m_valid.keep_probability: 1
}
valid_cost, valid_logits = sess.run([m_valid.cost, m_valid.logits],
feed_dict=feed_dict)
valid_pred, soft_pred = bdnn_prediction(valid_batch_size,
valid_logits,
threshold=0.6)
# print(np.sum(valid_pred))
raw_indx = int(np.floor(valid_labels.shape[1] / 2))
raw_labels = valid_labels[:, raw_indx]
raw_labels = raw_labels.reshape((-1, 1))
valid_accuracy = np.equal(valid_pred, raw_labels)
valid_accuracy = valid_accuracy.astype(int)
valid_accuracy = np.sum(valid_accuracy) / valid_batch_size
avg_valid_cost += valid_cost
avg_valid_accuracy += valid_accuracy
itr_sum += 1
total_avg_valid_accuracy = np.asscalar(
np.mean(np.asarray(accuracy_list)))
total_avg_valid_cost = np.asscalar(np.mean(np.asarray(cost_list)))
elif type is 'ACAM':
sys.path.insert(0, os.path.abspath('../../configure/ACAM'))
import config as cg
valid_batch_size = cg.batch_size
valid_data_set = dr.DataReader(valid_file_dir,
valid_file_dir + '/Labels',
norm_dir,
w=cg.w,
u=cg.u,
name="eval")
avg_valid_accuracy = 0.
avg_valid_cost = 0.
itr_sum = 0.
accuracy_list = [0 for i in range(valid_data_set._file_len)]
cost_list = [0 for i in range(valid_data_set._file_len)]
itr_file = 0
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
if valid_data_set.file_change_checker():
# print(itr_file)
accuracy_list[itr_file] = avg_valid_accuracy / itr_sum
cost_list[itr_file] = avg_valid_cost / itr_sum
avg_valid_cost = 0.
avg_valid_accuracy = 0.
itr_sum = 0
itr_file += 1
valid_data_set.file_change_initialize()
if valid_data_set.eof_checker():
valid_data_set.reader_initialize()
print('Valid data reader was initialized!'
) # initialize eof flag & num_file & start index
break
feed_dict = {
m_valid.inputs: valid_inputs,
m_valid.labels: valid_labels,
m_valid.keep_probability: 1
}
# valid_cost, valid_softpred, valid_raw_labels\
# = sess.run([m_valid.cost, m_valid.softpred, m_valid.raw_labels], feed_dict=feed_dict)
#
# fpr, tpr, thresholds = metrics.roc_curve(valid_raw_labels, valid_softpred, pos_label=1)
# valid_auc = metrics.auc(fpr, tpr)
valid_cost, valid_accuracy = sess.run(
[m_valid.cost, m_valid.reward], feed_dict=feed_dict)
avg_valid_accuracy += valid_accuracy
avg_valid_cost += valid_cost
itr_sum += 1
total_avg_valid_accuracy = np.asscalar(
np.mean(np.asarray(accuracy_list)))
total_avg_valid_cost = np.asscalar(np.mean(np.asarray(cost_list)))
elif type is 'LSTM':
sys.path.insert(0, os.path.abspath('../../configure/LSTM'))
import config as cg
valid_batch_size = cg.seq_len * cg.num_batches
valid_data_set = rnn_dr.DataReader(valid_file_dir,
valid_file_dir + '/Labels',
norm_dir,
target_delay=cg.target_delay,
name="eval")
avg_valid_accuracy = 0.
avg_valid_cost = 0.
itr_sum = 0.
accuracy_list = [0 for i in range(valid_data_set._file_len)]
cost_list = [0 for i in range(valid_data_set._file_len)]
itr_file = 0
while True:
valid_inputs, valid_labels = valid_data_set.next_batch(
valid_batch_size)
if valid_data_set.file_change_checker():
# print(itr_file)
accuracy_list[itr_file] = avg_valid_accuracy / itr_sum
cost_list[itr_file] = avg_valid_cost / itr_sum
avg_valid_cost = 0.
avg_valid_accuracy = 0.
itr_sum = 0
itr_file += 1
valid_data_set.file_change_initialize()
if valid_data_set.eof_checker():
valid_data_set.reader_initialize()
print('Valid data reader was initialized!'
) # initialize eof flag & num_file & start index
break
one_hot_labels = valid_labels.reshape((-1, 1))
one_hot_labels = dense_to_one_hot(one_hot_labels, num_classes=2)
feed_dict = {
m_valid.inputs: valid_inputs,
m_valid.labels: one_hot_labels,
m_valid.keep_probability: 1
}
# valid_cost, valid_softpred, valid_raw_labels\
# = sess.run([m_valid.cost, m_valid.softpred, m_valid.raw_labels], feed_dict=feed_dict)
#
# fpr, tpr, thresholds = metrics.roc_curve(valid_raw_labels, valid_softpred, pos_label=1)
# valid_auc = metrics.auc(fpr, tpr)
valid_cost, valid_accuracy = sess.run(
[m_valid.cost, m_valid.accuracy], feed_dict=feed_dict)
avg_valid_accuracy += valid_accuracy
avg_valid_cost += valid_cost
itr_sum += 1
total_avg_valid_accuracy = np.asscalar(
np.mean(np.asarray(accuracy_list)))
total_avg_valid_cost = np.asscalar(np.mean(np.asarray(cost_list)))
return total_avg_valid_accuracy, total_avg_valid_cost
def main(prj_dir=None, model=None, mode=None):
# Configuration Part #
if mode is 'train':
import path_setting as ps
set_path = ps.PathSetting(prj_dir, model)
logs_dir = initial_logs_dir = set_path.logs_dir
input_dir = set_path.input_dir
output_dir = set_path.output_dir
norm_dir = set_path.norm_dir
valid_file_dir = set_path.valid_file_dir
sys.path.insert(0, prj_dir+'/configure/bDNN')
import config as cg
global initLr, dropout_rate, max_epoch, batch_size, valid_batch_size
initLr = cg.lr
dropout_rate = cg.dropout_rate
max_epoch = cg.max_epoch
batch_size = valid_batch_size = cg.batch_size
global w, u
w = cg.w
u = cg.u
global bdnn_winlen, bdnn_inputsize, bdnn_outputsize
bdnn_winlen = (((w-1) / u) * 2) + 3
bdnn_inputsize = int(bdnn_winlen * num_features)
bdnn_outputsize = int(bdnn_winlen)
global num_hidden_1, num_hidden_2
num_hidden_1 = cg.num_hidden_1
num_hidden_2 = cg.num_hidden_2
# Graph Part #
print("Graph initialization...")
with tf.device(device):
with tf.variable_scope("model", reuse=None):
m_train = Model(is_training=True)
with tf.variable_scope("model", reuse=True):
m_valid = Model(is_training=False)
print("Done")
# Summary Part #
print("Setting up summary op...")
summary_ph = tf.placeholder(dtype=tf.float32)
with tf.variable_scope("Training_procedure"):
cost_summary_op = tf.summary.scalar("cost", summary_ph)
accuracy_summary_op = tf.summary.scalar("accuracy", summary_ph)
# train_summary_writer = tf.summary.FileWriter(logs_dir + '/train/', max_queue=4)
# valid_summary_writer = tf.summary.FileWriter(logs_dir + '/valid/', max_queue=4)
# summary_dic = summary_generation(valid_file_dir)
print("Done")
# Model Save Part #
print("Setting up Saver...")
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(initial_logs_dir)
print("Done")
# Session Part #
sess_config = tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)
sess_config.gpu_options.allow_growth = True
sess = tf.Session(config=sess_config)
if mode is 'train':
train_summary_writer = tf.summary.FileWriter(logs_dir + '/train/', sess.graph, max_queue=2)
valid_summary_writer = tf.summary.FileWriter(logs_dir + '/valid/', max_queue=2)
if ckpt and ckpt.model_checkpoint_path: # model restore
print("Model restored...")
if mode is 'train':
saver.restore(sess, ckpt.model_checkpoint_path)
else:
saver.restore(sess, initial_logs_dir+ckpt_name)
# print(initial_logs_dir)
# saver.save(sess, initial_logs_dir + "/model_bDNN.ckpt", 0) # model save
print("Done")
else:
sess.run(tf.global_variables_initializer()) # if the checkpoint doesn't exist, do initialization
if mode is 'train':
train_data_set = dr.DataReader(input_dir, output_dir, norm_dir, w=w, u=u, name="train") # training data reader initialization
# train_data_set = dr.DataReader(input_dir, output_dir, norm_dir, w=w, u=u, name="train") # training data reader initialization
if mode is 'train':
for itr in range(max_epoch):
train_inputs, train_labels = train_data_set.next_batch(batch_size)
feed_dict = {m_train.inputs: train_inputs, m_train.labels: train_labels,
m_train.keep_probability: dropout_rate}
sess.run(m_train.train_op, feed_dict=feed_dict)
if itr % 10 == 0 and itr >= 0:
train_cost, logits = sess.run([m_train.cost, m_train.logits], feed_dict=feed_dict)
result = bdnn_prediction(batch_size, logits, threshold=th)
raw_indx = int(np.floor(train_labels.shape[1] / 2))
raw_labels = train_labels[:, raw_indx]
raw_labels = raw_labels.reshape((-1, 1))
train_accuracy = np.equal(result, raw_labels)
train_accuracy = train_accuracy.astype(int)
train_accuracy = np.sum(train_accuracy) / batch_size # change to mean...
print("Step: %d, train_cost: %.4f, train_accuracy=%4.4f" % (itr, train_cost, train_accuracy*100))
train_cost_summary_str = sess.run(cost_summary_op, feed_dict={summary_ph: train_cost})
train_accuracy_summary_str = sess.run(accuracy_summary_op, feed_dict={summary_ph: train_accuracy})
train_summary_writer.add_summary(train_cost_summary_str, itr) # write the train phase summary to event files
train_summary_writer.add_summary(train_accuracy_summary_str, itr)
# if train_data_set.eof_checker():
if itr % 50 == 0 and itr > 0:
saver.save(sess, logs_dir + "/model.ckpt", itr) # model save
print('validation start!')
valid_accuracy, valid_cost = \
utils.do_validation(m_valid, sess, valid_file_dir, norm_dir, type='bDNN')
print("valid_cost: %.4f, valid_accuracy=%4.4f" % (valid_cost, valid_accuracy * 100))
valid_cost_summary_str = sess.run(cost_summary_op, feed_dict={summary_ph: valid_cost})
valid_accuracy_summary_str = sess.run(accuracy_summary_op, feed_dict={summary_ph: valid_accuracy})
valid_summary_writer.add_summary(valid_cost_summary_str, itr) # write the train phase summary to event files
valid_summary_writer.add_summary(valid_accuracy_summary_str, itr)
# full_evaluation(m_valid, sess, valid_batch_size, valid_file_dir, valid_summary_writer, summary_dic, itr)
# train_data_set.reader_initialize()
# print('Train data reader was initialized!') # initialize eof flag & num_file & start index
elif mode is 'test':
# full_evaluation(m_valid, sess, valid_batch_size, test_file_dir, valid_summary_writer, summary_dic, 0)
final_softout, final_label = utils.vad_test(m_valid, sess, valid_batch_size, test_file_dir, norm_dir, data_len,
eval_type)
if data_len is None:
return final_softout, final_label
else:
return final_softout[0:data_len, :], final_label[0:data_len, :]