def __init__(self):

super(Application, self).__init__()

self.config = config.Config()

self.model = model.Model(**self.config.__dict__)

print(self.config.__dict__)

def train(self):

def preprocess_fn(inputs):

for name in self.config.embedding_config:

value_size = self.config.embedding_config[name][0]

if inputs[name].dtype.is_integer:

inputs[name] = tf.mod(inputs[name], value_size)

else:

inputs[name] = tf.string_to_hash_bucket_fast(inputs[name], value_size)

inputs['context'] = tf.reshape(tf.to_float(inputs['context']), [-1, 1])

inputs['position'] = tf.reshape(tf.to_float(inputs['position']), [-1, 1])

inputs['label'] = tf.reshape(tf.to_float(inputs['label']), [-1, 1])

return inputs

with tf.Graph().as_default():

print("Build Graph")

data_sources = data.DataSourceMananger([self.config.train_data_source, self.config.test_data_source])

inputs = data_sources.get_features(self.config.batch_size, preprocess_fn)

scores, labels, context, global_step, losses, learning_rate, train_op = self.model.build_graph(inputs)

config = tf.ConfigProto(allow_soft_placement=True, gpu_options=tf.GPUOptions(allow_growth=True))

with tf.Session(config=config) as sess:

print("Init")

saver = tf.train.Saver()

summary_writer = tf.summary.FileWriterCache.get(self.config.task_dir)

sess.run(tf.global_variables_initializer())

test_auc, test_logloss, test_scores, test_labels = collections.defaultdict(list), collections.defaultdict(list), collections.defaultdict(list), collections.defaultdict(list)

# Generators

def train_generator():

data_sources.set_source(sess, self.config.train_data_source)

try:

while True:

_, step_value, learning_rate_value, losses_value = sess.run([train_op, global_step, learning_rate, losses])

yield step_value, learning_rate_value, losses_value

for _ in range(999):

sess.run(train_op)

except tf.errors.OutOfRangeError:

pass

def test_generator():

data_sources.set_source(sess, self.config.test_data_source)

try:

while True:

context_arr, scores_arr, labels_arr = sess.run((context, scores, labels))

yield context_arr, scores_arr, labels_arr

except tf.errors.OutOfRangeError:

pass

# Flow

step_per_epoch = self.config.step_per_epoch

while True:

# Train

print("[Train] Start")

for step_value, learning_rate_value, losses_value in train_generator():

epoch_value = 0 if step_per_epoch is None else int(step_value / step_per_epoch)

step_value = step_value if step_per_epoch is None else int(step_value % step_per_epoch)

losses_keys_str, losses_values_str = '+'.join(map(str, losses_value.keys())), '+'.join(map(str, losses_value.values()))

print("[Train] Epoch {epoch_value} Batch {step_value}/{step_per_epoch} Learning Rate {learning_rate_value:.5f} Loss({losses_keys_str}) {losses_values_str}".format(**locals()))

# Test

print("[Test] Start")

for idx, (context_arr, scores_arr, labels_arr) in enumerate(test_generator()):

print("[Test] Step %d" % idx) if idx % 100 == 0 else None

for _context, _score, _label in zip(context_arr, scores_arr, labels_arr):

test_scores[_context].append(_score)

test_labels[_context].append(_label)

names = sorted(test_scores)

print("[Test] Metrics Names%s" % names)

records = []

for name in names:

test_auc[name].append(roc_auc_score(test_labels[name], test_scores[name]))

test_logloss[name].append(log_loss(test_labels[name], test_scores[name]))

print("Context %s, AUC%s, LogLoss%s" % (name, test_auc[name], test_logloss[name]))

records.extend([test_auc[name][-1], test_logloss[name][-1]])

del test_scores[name][:], test_labels[name][:]

print("[Test] Early Stopping Check")

if len(test_auc[names[0]]) >= 2 and test_auc[names[0]][-1] < test_auc[names[0]][-2]:

break

else:

print('\t'.join(map(str, records)))

print(self.config.__dict__)