def get(self):
result = generate_data.gen(6)
#result = metric_writer.write('https://shining-fire-2617.firebaseio.com/data/cloud-androidwear/testmetric.json?auth=wGLkuGRzoPqkBFvICYIhnI8jj3rUUQ9E1jZxTtoy', m.timestamp, m.value)
#metric_list = generate_data.gen_multiple_metrics()
#result = metric_writer.write('https://shining-fire-2617.firebaseio.com/data/cloud-androidwear/', '.json?auth=wGLkuGRzoPqkBFvICYIhnI8jj3rUUQ9E1jZxTtoy', metric_list)
self.response.headers['Content-Type'] = 'text/plain'
self.response.write('ScheduledMetric: ' + str(result))
def gen_body(n = 3):
metric_list = generate_data.gen(n)
rows = []
body = {"rows":rows}
for i in range (0, n):
item = {"json": {"time": metric_list[i]['timestamp'], "value": metric_list[i]['value'],}}
body['rows'].append(item)
return body
def main():
graph = tf.Graph()
with graph.as_default():
x = tf.placeholder(tf.float32, [None, max_seq_len, 2],
name='x') # input : position with noise
y = tf.placeholder(tf.float32, [None, 2],
name='y') # label : ground truth position
l = tf.placeholder(tf.int32, [None], name='l') # seq. length
pred = net(x, l)
cost = tf.nn.l2_loss(y - pred)
tf.summary.scalar('cost', cost)
opt = tf.train.AdamOptimizer(
learning_rate=learning_rate).minimize(cost)
init = tf.global_variables_initializer()
sess = tf.Session()
g = gen(max_seq_len=max_seq_len, min_seq_len=max_seq_len, noise=5e-2)
with tf.Session(graph=graph) as sess:
sess.run(init)
merged = tf.summary.merge_all()
writer = tf.summary.FileWriter('/tmp/localize_logs', sess.graph)
for step, label, data, len in g.get(batch_size, training_iters):
feed_dict = {x: data, y: label, l: len}
# label = [batch, length, channel]
summary, _ = sess.run([merged, opt], feed_dict=feed_dict)
writer.add_summary(summary, step)
if step % 100 == 0:
prediction, loss = sess.run([pred, cost], feed_dict=feed_dict)
print('[Step %02d] loss : %f]' % (step, loss))
print('real :', label[0])
print('pred :', prediction[0])
output_graph_def = graph_util.convert_variables_to_constants(
sess, graph.as_graph_def(), [FINAL_TENSOR_NAME])
with gfile.FastGFile('output_graph.pb', 'wb') as f:
f.write(output_graph_def.SerializeToString())
def main():
graph = tf.Graph()
with graph.as_default():
x = tf.placeholder(tf.float32, [None, 1, 2],
name='x') # input : position with noise
y = tf.placeholder(tf.float32, [None, 1, 2],
name='y') # label : ground truth position
with tf.variable_scope(MODEL_SCOPE, reuse=None):
train = {}
train['pred'], train['reset'], train['update'] = net(
x, batch_size, None)
with tf.variable_scope(MODEL_SCOPE, reuse=True):
test = {}
test['pred'], test['reset'], test['update'] = net(x, 1, True)
cost = tf.nn.l2_loss(y - train['pred'])
tf.summary.scalar('cost', cost)
opt = tf.train.AdamOptimizer(
learning_rate=learning_rate).minimize(cost)
init = tf.global_variables_initializer()
for v in tf.trainable_variables():
print v.name, v.shape
sess = tf.Session()
g = gen(seq_len=seq_len, noise=4e-2)
with tf.Session(graph=graph) as sess:
saver = tf.train.Saver()
sess.run(init)
if load_ckpt:
saver.restore(sess, ckpt_path)
## TRAIN ##
if do_train:
## LOGGING ##
if not os.path.exists(log_root):
os.makedirs(log_root)
run_id = len(os.walk('/tmp/localize_logs/').next()[1])
writer = tf.summary.FileWriter(
os.path.join('/tmp/localize_logs/', ('run_%02d' % run_id)),
graph)
merged = tf.summary.merge_all()
#############
for step, label, data in g.get(batch_size, training_iters):
_ = sess.run(train['reset'])
for x_in, y_in in zip(data, label):
feed_dict = {
x: x_in[:, np.newaxis, :],
y: y_in[:, np.newaxis, :]
}
# label = [batch, length, channel]
summary, _, _ = sess.run([merged, opt, train['update']],
feed_dict=feed_dict)
writer.add_summary(summary, step)
if step % 100 == 0:
prediction, loss = sess.run([train['pred'], cost],
feed_dict=feed_dict)
print('[Step %02d] loss : %f]' % (step, loss))
#print('real :' , label[-1])
#print('pred :' , prediction[-1])
saver.save(sess, ckpt_path)
############
## SAVE ##
output_graph_def = graph_util.convert_variables_to_constants(
sess, graph.as_graph_def(),
['model_1/pred', 'model_1/update', 'model_1/reset'])
with gfile.FastGFile('output_graph.pb', 'wb') as f:
f.write(output_graph_def.SerializeToString())
############
## TESTING ##
if do_test:
w, h = 512, 512
frame = np.zeros((h, w, 3), dtype=np.uint8)
for step, label, data in g.get(1, testing_iters):
frame.fill(0)
_ = sess.run(test['reset'])
prvx = None
prvy = None
prvp = None
err_m = 0
err_p = 0
for x_in, y_in in zip(data, label):
feed_dict = {
x: x_in[:, np.newaxis, :],
y: y_in[:, np.newaxis, :]
}
pred, _ = sess.run([test['pred'], test['update']],
feed_dict=feed_dict)
x_in, y_in, pred = [(np.squeeze(e) * [h / 2, w / 2] +
[h / 2, w / 2]).astype(np.int32)
for e in (x_in, y_in, pred)]
y_pt = (y_in[1], y_in[0])
x_pt = (x_in[1], x_in[0])
p_pt = (pred[1], pred[0])
err_m += np.linalg.norm(y_in - x_in)
err_p += np.linalg.norm(y_in - pred)
cv2.circle(frame, y_pt, 5, (255, 0, 0),
thickness=-1) # --> true pos, blue
cv2.circle(frame, x_pt, 4, (0, 255, 0),
thickness=1) # --> measured pos, green
cv2.circle(frame, p_pt, 4, (0, 0, 255),
thickness=1) # --> predicted pos, reg
if prvx is not None:
cv2.line(frame, prvy, y_pt, (255, 0, 0), 1)
cv2.line(frame, prvx, x_pt, (0, 255, 0), 1)
cv2.line(frame, prvp, p_pt, (0, 0, 255), 1)
prvy = y_pt
prvx = x_pt
prvp = p_pt
cv2.imshow('frame', frame)
if cv2.waitKey(20) == 27:
break
print('err_m : %.2f ; err_p : %.2f' % (err_m, err_p))
if cv2.waitKey(0) == 27:
break
import numpy as np
from scipy import signal
import mnist
import matplotlib.pyplot as plt
import generate_data
np.random.seed(5)
X_train, y_train, X_test, y_test = generate_data.gen()
def tanh(x):
return np.tanh(x)
def d_tanh(x):
return 1 - np.tanh(x)**2
def log(x):
return 1 / (1 + np.exp(-1 * x))
def d_log(x):
return log(x) * (1 - log(x))
def relu(a):
a[a < 0] = 0.
return a