def nbody_wrapper():
galaxy = nbody.random_galaxy(NUM_BODIES)
nbody.simulate(galaxy, TIMESTEPS)
r = spartan.sum(galaxy['x'] + galaxy['y'] + galaxy['z'])
r.optimized()
print 'r = ', r.glom()
def nbody_wrapper(): galaxy = nbody.random_galaxy(NUM_BODIES) nbody.simulate(galaxy, TIMESTEPS) r = spartan.sum(galaxy['x'] + galaxy['y'] + galaxy['z']) r.optimized() print 'r = ', r.glom()
def run(self):
(train_data, test_data) = load_mb_from_mat(self.data_file, self.mb_size)
#num_test_samples = test_data[0].shape[0]
#(test_samples, test_labels) = test_data
count = 1
begin = time.time()
for epoch in range(self.num_epochs):
print '---Start epoch #%d' % epoch
# train
iterations = len(train_data)
for (mb_samples, mb_labels) in train_data:
num_samples = mb_samples.shape[0]
a1 = mb_samples.T
target = mb_labels.T
# ff
#a2 = relu(np.dot(self.w1, a1) + self.b1)
#a3 = np.dot(self.w2, a2) + self.b2
a2 = relu(spartan.dot(self.w1, a1) + self.b1)
a3 = spartan.dot(self.w2, a2) + self.b2
# softmax & error
out = softmax(a3)
s3 = out - target
# bp
#s2 = np.dot(self.w2.T, s3)
s2 = spartan.dot(self.w2.T, s3)
s2 = relu_back(s2, a2)
# grad
#gw1 = np.dot(s2, a1.T) / num_samples
#gb1 = np.sum(s2, axis=1, keepdims=True) / num_samples
#gw2 = np.dot(s3, a2.T) / num_samples
#gb2 = np.sum(s3, axis=1, keepdims=True) / num_samples
gw1 = spartan.dot(s2, a1.T) / num_samples
gb1 = (spartan.sum(s2, axis=1) / num_samples).reshape(s2.shape[0], 1)
gw2 = spartan.dot(s3, a2.T) / num_samples
gb2 = (spartan.sum(s3, axis=1) / num_samples).reshape(s3.shape[0], 1)
# update
self.w1 -= self.eps_w * gw1
self.w2 -= self.eps_w * gw2
self.b1 -= self.eps_b * gb1
self.b2 -= self.eps_b * gb2
iterations -= 1
out.evaluate()
self.w1.evaluate()
self.w2.evaluate()
self.b1.evaluate()
self.b2.evaluate()
#if count % 40 == 0 or iterations == 0:
if count % 40 == 0:
#correct = np.argmax(out, axis=0) - np.argmax(target, axis=0)
#print 'Training error:', float(np.count_nonzero(correct)) / num_samples
correct = spartan.argmax(out, axis=0) - spartan.argmax(target, axis=0)
print 'Training error:', (float(spartan.count_nonzero(correct).glom())
/ num_samples)
count = count + 1
print 'spent ', (time.time() - begin)
# test
#a1 = test_samples.T
#a2 = relu(np.dot(self.w1, a1) + self.b1)
#a3 = np.dot(self.w2, a2) + self.b2
#correct = np.argmax(a3, axis=0) - np.argmax(test_labels.T, axis=0)
a1 = test_samples.T
a2 = relu(spartan.dot(self.w1, a1) + self.b1)
a3 = spartan.dot(self.w2, a2) + self.b2
correct = spartan.argmax(a3, axis=0) - spartan.argmax(test_labels.T, axis=0)