def load_dataset(path, config):
print('Loading data: ' + path)
train, valid, test = read_data.load_data(path, n_words=config.vocab_size, \
valid_portion=0.15, maxlen=config.maxlen)
train = read_data.prepare_data(train[0], train[1], maxlen=config.maxlen)
valid = read_data.prepare_data(valid[0], valid[1], maxlen=config.maxlen)
test = read_data.prepare_data(test[0], test[1], maxlen=config.maxlen)
return (train, valid, test)
import numpy as np
from read_data import read_messages, read_classes, prepare_data
#messages = read_messages()
#classes = read_classes()
data = prepare_data()
a = np.array(data)
np.savetxt('data_mapbook.csv', a, fmt='%s', delimiter=',')
config.layer = int(sys.argv[1])
config.step = int(sys.argv[2])
print("dataset: " + sys.argv[3])
print("iteration: " + str(config.layer))
print("step: " + str(config.step))
print("model: " + str(sys.argv[4]))
#word2vec
f = open(vector_path, 'rb')
matrix = np.array(pickle.load(f))
config.vocab_size = matrix.shape[0]
#load datasets
train_dataset, valid_dataset, test_dataset = read_data.load_data(\
path=path,n_words=config.vocab_size)
config.num_label = len(set(train_dataset[1]))
print("number label: " + str(config.num_label))
train_dataset = read_data.prepare_data(train_dataset[0], train_dataset[1])
valid_dataset = read_data.prepare_data(valid_dataset[0], valid_dataset[1])
test_dataset = read_data.prepare_data(test_dataset[0], test_dataset[1])
with tf.Graph().as_default(), tf.Session() as session:
initializer = tf.random_normal_initializer(0, 0.05)
classifier = Classifer(config=config, session=session)
total = 0
#print trainable variables
for v in tf.trainable_variables():
print(v.name)
shape = v.get_shape()
try:
size = shape[0].value * shape[1].value
domain_size, domain_list = get_domains()
#load dataset
train_datasets, valid_datasets, test_datasets = [], [], []
for domain in domain_list:
train, valid, test = read_data.load_data(path='dataset'+config.dataset+'/'+domain+'/dataset',n_words=config.vocab_size, \
valid_portion=config.valid_portion, maxlen=config.maxlen)
train_datasets.append(train)
valid_datasets.append(valid)
test_datasets.append(test)
#transform dataset to matrix
for index in range(domain_size):
train = read_data.prepare_data(train_datasets[index][0],
train_datasets[index][1],
maxlen=config.maxlen,
traindata=True,
index=index)
valid = read_data.prepare_data(valid_datasets[index][0],
valid_datasets[index][1],
maxlen=config.maxlen,
traindata=False,
index=index)
test = read_data.prepare_data(test_datasets[index][0],
test_datasets[index][1],
maxlen=config.maxlen,
traindata=False,
index=index)
train_datasets[index] = train
valid_datasets[index] = valid
test_datasets[index] = test
hyperparameters = Hyperparameters()
with tf.Graph().as_default():
# =========================================================================================================
# BUILD MODEL
# =========================================================================================================
train_operation = model.model_architecture(hyperparameters)
# =========================================================================================================
# LOAD DATA
# =========================================================================================================
input_train, train_label, input_test, test_label = read_data.load_data(
hyperparameters.num_points)
scaled_laplacian_train, scaled_laplacian_test = read_data.prepare_data(
input_train, input_test, hyperparameters.num_neighhbors,
hyperparameters.num_points)
# =========================================================================================================
# TRAIN MODEL
# =========================================================================================================
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
saver = tf.train.Saver()
learning_rate = hyperparameters.learning_rate
save_model_path = '../model/'
weight_dict = utils.weight_dict_fc(train_label, hyperparameters)
test_label_whole = []
def main(_):
# Neural net definition
x = tf.placeholder(tf.float32, shape=[None, 2352])
y_ = tf.placeholder(tf.float32, shape=[None, 2])
initializer = tf.contrib.layers.xavier_initializer()
x_reshaped = tf.reshape(x, [-1, 28, 28, 3])
conv_1 = tf.layers.conv2d(
inputs=x_reshaped,
filters=32,
kernel_size=5,
padding='same',
activation=tf.nn.relu,
kernel_initializer=initializer
)
pool_1 = tf.layers.max_pooling2d(inputs=conv_1, pool_size=[2,2], strides=2)
conv_2 = tf.layers.conv2d(
inputs=pool_1,
filters=64,
kernel_size=5,
padding='same',
activation=tf.nn.relu,
kernel_initializer=initializer
)
pool_2 = tf.layers.max_pooling2d(inputs=conv_2, pool_size=[2,2], strides=2)
pool_2_flat = tf.reshape(pool_2, [-1, 7*7*64])
dense_1 = tf.layers.dense(inputs=pool_2_flat, units=1024, activation=tf.nn.relu, kernel_initializer=initializer)
dropout = tf.layers.dropout(inputs=dense_1, rate=0.4)
y_conv = tf.layers.dense(inputs=dropout, units=2, kernel_initializer=initializer)
# Train step data
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits_v2(labels=y_, logits=y_conv))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
saver = tf.train.Saver()
path = "./hotdog-model/"
if not os.path.exists(path):
os.makedirs(path)
with tf.Session() as sess:
# sess.run(tf.global_variables_initializer())
saver.restore(sess, path + 'test-model')
file_list, y_image_label = prepare_data(FLAGS.image_dir)
le = preprocessing.LabelEncoder()
labels = ['hotdog', 'not_a_hotdog']
y_one_hot = tf.one_hot(le.fit_transform(labels),depth=2)
if FLAGS.train:
x_feed = sess.run(read_image_array(file_list))
y_feed = sess.run(y_one_hot)
for i in range(80):
if i % 10 == 0:
train_accuracy = accuracy.eval(feed_dict={
x:x_feed , y_: y_feed})
print('step %d, training accuracy %g' % (i, train_accuracy))
train_step.run(feed_dict={x:x_feed , y_: y_feed})
save_path = saver.save(sess, path + 'test-model')
elif FLAGS.predict_image != "":
predicted = tf.argmax(y_conv, 1)
x_single_img = sess.run(read_single_image(FLAGS.predict_image))
pixels = mpimg.imread(FLAGS.predict_image)
plt.imshow(pixels)
result = le.inverse_transform(sess.run(predicted, feed_dict={x: x_single_img}))[0]
if result == 'hotdog':
text = 'This is a hotdog, right?'
else:
text = 'This isn\'t a hotdog, right?'
plt.title(text)
plt.show()
if __name__ == '__main__':
# fix random seed for reproducibility
np.random.seed(7)
# Load Data
data_2_05 = read_proteins("../Data/Protein Data/astral-scope-95-2.05.fa")
data_2_06 = read_proteins("../Data/Protein Data/astral-scope-95-2.06.fa")
data_2_06 = {
'A': data_2_06['A'].difference(data_2_05['A']),
'B': data_2_06['B'].difference(data_2_05['B']),
'C': data_2_06['C'].difference(data_2_05['C']),
'D': data_2_06['D'].difference(data_2_05['D']),
}
X_train, y_train, vocab = data_formatting(prepare_data(data_2_05))
X_test, y_test, vocab = data_formatting(prepare_data(data_2_06), vocab)
# truncate and pad input sequences
max_seq_length = 200
X_train = sequence.pad_sequences(X_train, maxlen=max_seq_length)
X_test = sequence.pad_sequences(X_test, maxlen=max_seq_length)
# Find number of protein lexemes
print "Protein Lexemes Count =", len(vocab.keys()) + 1
# create the model
embedding_vecor_length = 6
model = Sequential()
model.add(
Embedding(len(vocab) + 1,
test_acc, _ = run_epoch(session, config, model, test_dataset,
tf.no_op(), 1, False)
print("Eval Accuracy = %.2f time: %.3f\n" %
(100 * test_acc, time.time() - start_time))
def word_to_vec(matrix, session, config, *args):
for model in args:
session.run(tf.assign(model.embedding, matrix))
if __name__ == "__main__":
config = Config()
train_dataset, test_dataset = read_data.load_data(path=config.data_path +
"parsed_data/")
# conver datas into matrix
train_dataset = read_data.prepare_data(train_dataset[0], train_dataset[1],
train_dataset[2])
test_dataset = read_data.prepare_data(test_dataset[0], test_dataset[1],
test_dataset[2])
with tf.Graph().as_default(), tf.Session(
config=tf.ConfigProto()) as session:
classifier = Classifer(config=config, session=session)
session.run(tf.global_variables_initializer())
saver = tf.train.Saver(max_to_keep=config.max_to_keep)
for epoch_id in range(config.max_max_epoch):
train_test_model(config, epoch_id, session, classifier,
train_dataset, test_dataset, saver)
def main(unused_args):
#configs
config = Config()
#domains to be processed
domain_list = sys.argv[1:]
domain_size = len(domain_list)
if domain_size <= 0:
print("No dataset")
exit(1)
#load dataset
train_datasets, valid_datasets, test_datasets = [], [], []
for domain in domain_list:
train, valid, test = read_data.load_data(path='dataset'+config.dataset+'/'+domain+'/dataset',n_words=config.vocab_size, \
valid_portion=config.valid_portion, maxlen=config.maxlen)
train_datasets.append(train)
valid_datasets.append(valid)
test_datasets.append(test)
#transform dataset to matrix
for index in range(domain_size):
train = read_data.prepare_data(train_datasets[index][0],
train_datasets[index][1],
maxlen=config.maxlen,
traindata=True)
valid = read_data.prepare_data(valid_datasets[index][0],
valid_datasets[index][1],
maxlen=config.maxlen,
traindata=False)
test = read_data.prepare_data(test_datasets[index][0],
test_datasets[index][1],
maxlen=config.maxlen,
traindata=False)
train_datasets[index] = train
valid_datasets[index] = valid
test_datasets[index] = test
config.num_classes = count_labels(train_datasets[0][2])
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.8)
with tf.Graph().as_default(), tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options)) as session:
initializer = tf.random_normal_initializer(0, 0.05)
#training model for shared weights
with tf.variable_scope("shared_model",
reuse=None,
initializer=initializer):
share_model_train = EmbeddingModel(is_training=True,
config=config,
session=session,
trainable=True)
#testing model for shared weights
with tf.variable_scope("shared_model",
reuse=True,
initializer=initializer):
share_model_test = EmbeddingModel(is_training=False,
config=config,
session=session,
trainable=True)
#build models
train_models = []
test_models = []
for index in range(domain_size):
with tf.variable_scope("m" + str(index),
reuse=None,
initializer=initializer):
train_model = Combine_two_model(share_model_train, config)
with tf.variable_scope("m" + str(index),
reuse=True,
initializer=initializer):
test_model = Combine_two_model(share_model_test, config)
train_models.append(train_model)
test_models.append(test_model)
init = tf.global_variables_initializer()
session.run(init)
#initialize share model's embedding with word2vec
word_to_vec(session, config, share_model_train)
#train test model
train_test_model(config, session,\
train_models,test_models,test_models,\
train_datasets,valid_datasets,test_datasets)
def process_queue_batch(args):
batch, k = args
sim = {}
for i, j in tqdm(batch, position=k):
sim[(i, j)] = similarity(train_data[i][0], train_data[j][0])
sim[(j, i)] = sim[(i, j)]
return sim
if __name__ == '__main__':
num_threads = int(sys.argv[1])
# Read and prep data
data_2_05 = read_proteins("../Data/Data/astral-scope-95-2.05.fa")
train_data = prepare_data(data_2_05)
# Select out few samples
np.random.shuffle(train_data)
train_data = train_data[:250]
# Build similarity score matrix
sim = np.zeros((len(train_data), len(train_data)))
work_queue = [(i, j) for i in range(len(train_data))
for j in range(i, len(train_data))]
work_per_thread = int(np.ceil(float(len(work_queue)) / num_threads))
# Parallely execute load
p = Pool(num_threads)
res = p.map(process_queue_batch,
[(work_queue[i * work_per_thread:(i + 1) * work_per_thread], i)
