def get_loss(self, predictions, answer):
#predictions: [candidate_answer_num, 2* hidden_dim]
with tf.variable_scope('projection_layer',reuse=True):
softmax_w=tf.get_variable('softmax_w')
softmax_b=tf.get_variable('softmax_b')
scores=tf.matmul(predictions, softmax_w)+softmax_b
scores=tf.squeeze(scores) #[candidate_answer_num]
scores=tf.expand_dims(scores, 0) #[1, candidate_answer_num]
truth=tf.onehot(answer, tf.gather(tf.shape(predictions),0))
truth=tf.expand_dims(truth, 0)
cross_entropy=tf.nn.softmax_cross_entropy_with_logits(logits=scores, labels=truth)
cross_entropy=tf.squeeze(cross_entropy)
return cross_entropy
def load_omniglot_data(data_dir, is_training):
"""
loads training date or validation data
"""
omniglot_data = np.load(data_dir + '/data.npy')
omniglot_data = np.reshape(omniglot_data, newshape=(1622, 20, 28, 28, 1))
omniglot_label = np.arange(1622)
omniglot_label = tf.onehot(omniglot_label, 1622)
if is_training:
tra_data = omniglot_data[:1200]
tra_label = omniglot_label[:1200]
return tra_data, tra_label
else:
val_data = omniglot_data[1200:1411]
val_label = omniglot_label[1200:1411]
return val_data, val_label
def __init__(self, is_training=True):
self.graph = Graph()
with self.graph as_default():
if is_training:
self.x, self.y = get_batch_data()
else:
self.x = tf.placeholder(tf.int32, shape=(None, pre.maxlen))
self.y = tf.placeholder(tf.int32, shape=(None, pre.maxlen))
self.decoder_input = tf.concat((tf.ones_like(self.y[:,:1])*2,self.y[:,:-1]),axis=-1)
en2id, id2en = load_vocab("./data/en_words.txt")
ch2id, id2ch = load_vocab("./data/ch_words.txt")
with tf.variable_scope("encoder"):
self.enc = embedding(self.x, vocab_size=len(en2id), num_units=pre.embedding)
key_masks = tf.expand_dims(tf.sign(tf.reduce_sum(tf.abs(self.enc),axis=-1)), -1)
self.enc = self.enc + position_embedding(self.x, embedding_dim=pre.embedding)
self.enc = self.enc*key_masks
self.enc = tf.layers.dropout(self.enc, rate=0.1, training=tf.convert_to_tensor(is_training))
for i in range(6):
self.enc = multihead_attention(queries=self.enc,
keys=self.enc,
embedding_dim=pre.embedding,
num_head=8,
dropout_rate=0.1,
is_training=is_training,
future_blind=False)
self.enc = feedforward(inputs=self.enc)
with tf.variable_scope("decode"):
self.dec = embedding(inputs=self.y, vocab_size=len(ch2id), embedding_dim=pre.embedding)
key_masks = tf.expand_dims(tf.sign(tf.reduce_sum(tf.abs(self.dec), axis=-1)),-1)
self.dec = self.dec + position_embedding(self.y, embedding_dim=pre.embedding)
self.dec = self.dec*key_masks
self.dec = tf.layers.dropout(self.dec, rate=0.1, training=tf.convert_to_tensor(is_training))
for i in range(6):
self.dec = multihead_attention(queries=self.dec,
keys=self.dec,
embedding_dim=pre.embedding,
num_head=8,
dropout_rate=0.1,
is_training=is_training,
future_blind=True)
self.dec = multihead_attention(queries=self.dec,
keys=self.enc,
embedding_dim=pre.embedding,
num_head=8,
dropout_rate=0.1,
is_training=is_training,
future_blind=False)
self.dec = feedforward(self.dec)
self.logits = tf.layers.dense(self.dec, len(ch2id))
self.preds = tf.to_int32(tf.argmax(self.logits,axis=-1))
self.istarget = tf.to_float(tf.not_equal(self.y, 0))
self.acc = tf.reduce_sum(tf.to_float(tf.equal(self.preds,self.y))*self.istarget)/(tf.reduce_sum(self.istarget))
if is_training:
self.y_smoothed = label_smoothing(tf.onehot(self.y, depth=len(ch2id)))
self.loss = tf.nn.softmax_cross_entropy_with_logits(labels=self.y, logits=self.logits)
self.mean_loss = tf.reduce_sum(self.loss*self.istarget)/tf.reduce_sum(self.istarget)
self.optimizer = tf.train.AdamOptimizer(learning_rate=0.0001,
beta1=0.9,
beta2=0.98,
epsilon=1e-8)
self.opt = self.optimizer.minimize(self.mean_loss)
tf.set_random_seed(777)
xy = np.loadtxt('data-04-zoo.csv', delimiter = ',',dtype=np.float32)
X_data = xy[:,0:-1]
N = X_data.shape[0]
Y_data = xy[:,[-1]]
print("y has one of the following values")
nb_classes = 7
X = tf.placeholder(tf.float32, [None, 16])
Y = tf.placeholder(tf.int32,[None, 1])
target = tf.onehot(y, nb_classes)
target = tf.reshape(target, [-1,nb_classes])
target = tf.cast(target, tf.float32)
W = tf.Variable(tf.random_normal([16, nb_classes]),name='weight')
b = tf.Variable(tf.random_normal([nb_classes]),name='bias')
def sigma(x):
return 1./(1.+ tf.exp(-x))
def sigma_prime(x):
return sigma(x) * (1. - sigma(x))
layer_1 = tf.matmul(X,W)+b
y_pred = sigma(layer_1)
def __init__(self, learning_rate, num_bandit, num_arm):
#현재 환경 입력(슬롯머신 index)
self.state_in = tf.placeholder(tf.int32, [1])
#환경을 one_hot encoding으로 표현
state_in_onehot = tf.onehot(self.state_in, num_bandit)
crp=resize(crp,((scale, scale)))
return crp
#cover.save('test-image-cover.jpeg', image.format)
#print(type(np.asarray(img)))
#def load_img(path):
# image = cv2.imread(path)
X = np.asmatrix(list2)
X = np.reshape(x,(len(list2),784))
'''
convert y to logits
'''
y_ = tf.onehot(indices=y_,depth=4)
x = tf.placeholder(shape=[None,a])
y = tf.placeholder(shape=[None,4])
w1 = tf.Variable(shape = [a,400])
b1 = tf.Variable(shape = [400,1])
w2 = tf.Variable(shape = [400,4])
b2 = tf.Variable(shape = [4,1])
a1 = tf.matmul(x,w1)+b1
h1 = tf.sigmoid(a1)