def sgd_lr_4sps(data,label,sample,param): start = time.time() L = len(sample) num_samples = L * (L-1) / 2 num_features = data.shape[1] * 2 print "#Samples:%d, #Features:%d"%(num_samples,num_features) weights = np.random.random(num_features) alpha = param['alpha'] iters = param['iters'] gamma = param['gamma'] for it in xrange(iters): print "iter%d "%it ic = 0 for pi,py in pair_dataset_sto_sparse_generator(data,label,sample): output = sigmoid(np.sum(weights[pi.indices])) err = output - py weights = weights - alpha * (pi.toarray().reshape(num_features) * err + gamma * weights) util.view_bar(ic,num_samples) ic += 1 print "train complete! took %.2fs"%(time.time()-start) return weights
def sgd_lr(data,label,param): start = time.time() num_samples = data.shape[0] * (data.shape[0]-1) / 2 num_features = data.shape[1] * 2 print "#Samples:%d, #Features:%d"%(num_samples,num_features) weights = np.random.random(num_features) alpha = param['alpha'] iters = param['iters'] gamma = param['gamma'] for it in xrange(iters): print "iter%d "%it ic = 0 for pi,py in pair_dataset_sto_generator(data,label): output = sigmoid(np.dot(pi, weights)) err = output - py weights = weights - alpha * (pi * err + gamma * weights) # sys.stdout.write(str(ic*100.0/num_samples)+'%\r') util.view_bar(ic,num_samples) ic += 1 print "train complete! took %.2fs"%(time.time()-start) return weights
def sgd_lr_4sps(data, label, sample, param):
start = time.time()
L = len(sample)
num_samples = L * (L - 1) / 2
num_features = data.shape[1] * 2
print "#Samples:%d, #Features:%d" % (num_samples, num_features)
weights = np.random.random(num_features)
alpha = param['alpha']
iters = param['iters']
gamma = param['gamma']
for it in xrange(iters):
print "iter%d " % it
ic = 0
for pi, py in pair_dataset_sto_sparse_generator(data, label, sample):
output = sigmoid(np.sum(weights[pi.indices]))
err = output - py
weights = weights - alpha * (
pi.toarray().reshape(num_features) * err + gamma * weights)
util.view_bar(ic, num_samples)
ic += 1
print "train complete! took %.2fs" % (time.time() - start)
return weights
def pwlr_predict4sps_offline(train_data, train_label, sample, weights):
start = time.time()
print "Prediction Begin!"
L = len(sample)
test_inx = list(
set([i for i in xrange(train_data.shape[0])]) - set(sample))
result = np.zeros(len(test_inx))
num_test = len(test_inx)
ic = 0
for i in xrange(num_test):
test_instance = train_data.getrow(test_inx[i]).toarray().tolist()
for j in xrange(L):
train_instance = train_data.getrow(sample[j]).toarray().tolist()
sample_instance = sp.sparse.csr_matrix(test_instance +
train_instance)
result[i] += (1 +
be_positive(np.sum(weights[sample_instance.indices]))
) * train_label[sample[j]]
# sys.stdout.write(str((i+1)*(j+1)*100.0/(num_test*L))+'%\r')
util.view_bar(ic, num_test * L)
ic += 1
print "prediction complete! took %.2fs" % (time.time() - start)
return result / L
def sgd_lr(data, label, param):
start = time.time()
num_samples = data.shape[0] * (data.shape[0] - 1) / 2
num_features = data.shape[1] * 2
print "#Samples:%d, #Features:%d" % (num_samples, num_features)
weights = np.random.random(num_features)
alpha = param['alpha']
iters = param['iters']
gamma = param['gamma']
for it in xrange(iters):
print "iter%d " % it
ic = 0
for pi, py in pair_dataset_sto_generator(data, label):
output = sigmoid(np.dot(pi, weights))
err = output - py
weights = weights - alpha * (pi * err + gamma * weights)
# sys.stdout.write(str(ic*100.0/num_samples)+'%\r')
util.view_bar(ic, num_samples)
ic += 1
print "train complete! took %.2fs" % (time.time() - start)
return weights
def pwlr_predict4sps_online(train_data,train_label,sample,test_data,weights): start = time.time() print "Prediction Begin!" L = len(sample) result = np.zeros(test_data.shape[0]) num_test = test_data.shape[0] ic = 0 for i in xrange(num_test): test_instance = test_data.getrow(i).toarray().tolist() for j in xrange(L): train_instance = train_data.getrow(sample[j]).toarray().tolist() sample_instance = sp.sparse.csr_matrix(test_instance + train_instance) result[i] += (1+be_positive(np.sum(weights[sample_instance.indices]))) * train_label[sample[j]] # sys.stdout.write(str((i+1)*(j+1)*100.0/(num_test*L))+'%\r') util.view_bar(ic,num_test*L) ic += 1 print "prediction complete! took %.2fs"%(time.time()-start) return result/L
def pwlr_predict(train_data,train_label,test_data,weights): start = time.time() print "Prediction Begin!" L = len(train_data) result = np.zeros(len(test_data)) num_test = len(test_data) ic = 0 for i in xrange(len(test_data)): test_instance = test_data[i] for j in xrange(L): train_instance = train_data[j] sample_instance = test_instance + train_instance result[i] += (1+be_positive(np.dot(sample_instance,weights))) * train_label[j] # sys.stdout.write(str((i+1)*(j+1)*100.0/(num_test*L))+'%\r') util.view_bar(ic,num_test*L) ic += 1 print "prediction complete! took %.2fs"%(time.time()-start) return result/L
def pwlr_predict(train_data, train_label, test_data, weights):
start = time.time()
print "Prediction Begin!"
L = len(train_data)
result = np.zeros(len(test_data))
num_test = len(test_data)
ic = 0
for i in xrange(len(test_data)):
test_instance = test_data[i]
for j in xrange(L):
train_instance = train_data[j]
sample_instance = test_instance + train_instance
result[i] += (1 + be_positive(np.dot(sample_instance,
weights))) * train_label[j]
# sys.stdout.write(str((i+1)*(j+1)*100.0/(num_test*L))+'%\r')
util.view_bar(ic, num_test * L)
ic += 1
print "prediction complete! took %.2fs" % (time.time() - start)
return result / L
def train(length, width, image, check):
x = tf.placeholder(tf.float32, [1, width, length, NUM_CHANNELS], name='x-input')
y_ = tf.placeholder(tf.float32, [None, 4], name='y-input')
regularizer = tf.contrib.layers.l2_regularizer(REGULARIZATRION_RATE)
y = process_image(x, False, regularizer)
global_step = tf.Variable(0, trainable=False)
variable_averages = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY, global_step)
variable_averages_op = variable_averages.apply(tf.trainable_variables())
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1))
cross_entropy_mean = tf.reduce_mean(cross_entropy)
loss = cross_entropy_mean + tf.add_n(tf.get_collection('losses'))
learning_rate = tf.train.exponential_decay(LEARNING_RATE_BASE, global_step, TRAINING_STEPS, LEARNING_RATE_DECAY, staircase=True)
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step=global_step)
with tf.control_dependencies([train_step, variable_averages_op]):
train_op = tf.no_op(name='train')
saver = tf.train.Saver()
with tf.Session() as sess:
if os.path.exists('path/'):
saver.restore(sess, 'path/train.ckpt')
else:
tf.global_variables_initializer().run()
for step in range(TRAINING_STEPS):
test, loss_value, step = sess.run([train_op, loss, global_step], feed_dict={x: image, y_: check})
util.view_bar("processing step of " , step, 100)
saver.save(sess, 'path/train.ckpt')
return sess.run(y, feed_dict={x: image})
