def train_network(model, num_epochs=100, minibatch_size=256, lr=0.01, mom=0.75, wd=5e-4):
# load data
(train_data, test_data) = mnist_io.load_mb_from_mat('mnist_all.mat', minibatch_size / len(gpu))
num_test_samples = test_data[0].shape[0]
test_samples = owl.from_numpy(test_data[0]).reshape([28, 28, 1, num_test_samples])
test_labels = owl.from_numpy(test_data[1])
for i in xrange(num_epochs):
print "---Epoch #", i
last = time.time()
count = 0
weightgrads = [None] * len(gpu)
biasgrads = [None] * len(gpu)
for (mb_samples, mb_labels) in train_data:
count += 1
current_gpu = count % len(gpu)
owl.set_device(gpu[current_gpu])
num_samples = mb_samples.shape[0]
data = owl.from_numpy(mb_samples).reshape([28, 28, 1, num_samples])
label = owl.from_numpy(mb_labels)
out, weightgrads[current_gpu], biasgrads[current_gpu] = bpprop(model, data, label)
out.start_eval()
if current_gpu == 0:
for k in range(len(model.weights)):
model.weightdelta[k] = mom * model.weightdelta[k] - lr / num_samples / len(gpu) * multi_gpu_merge(weightgrads, 0, k) - lr * wd * model.weights[k]
model.biasdelta[k] = mom * model.biasdelta[k] - lr / num_samples / len(gpu) * multi_gpu_merge(biasgrads, 0, k)
model.weights[k] += model.weightdelta[k]
model.bias[k] += model.biasdelta[k]
if count % (len(gpu) * lazy_cycle) == 0:
print_training_accuracy(out, label, num_samples, 'Training')
print '---End of Epoch #', i, 'time:', time.time() - last
# do test
out, _, _ = bpprop(model, test_samples, test_labels)
print_training_accuracy(out, test_labels, num_test_samples, 'Testing')
def train_network(model, num_epochs=100, minibatch_size=256, lr=0.01, mom=0.75, wd=5e-4):
# load data
(train_data, test_data) = mnist_io.load_mb_from_mat('mnist_all.mat', minibatch_size / len(gpu))
num_test_samples = test_data[0].shape[0]
test_samples = owl.from_numpy(test_data[0]).reshape([28, 28, 1, num_test_samples])
test_labels = owl.from_numpy(test_data[1])
for i in xrange(num_epochs):
print "---Epoch #", i
last = time.time()
count = 0
weightgrads = [None] * len(gpu)
biasgrads = [None] * len(gpu)
for (mb_samples, mb_labels) in train_data:
count += 1
current_gpu = count % len(gpu)
owl.set_device(gpu[current_gpu])
num_samples = mb_samples.shape[0]
data = owl.from_numpy(mb_samples).reshape([28, 28, 1, num_samples])
label = owl.from_numpy(mb_labels)
out, weightgrads[current_gpu], biasgrads[current_gpu] = bpprop(model, data, label)
if current_gpu == 0:
for k in range(len(model.weights)):
model.weightdelta[k] = mom * model.weightdelta[k] - lr / num_samples / len(gpu) * multi_gpu_merge(weightgrads, 0, k) - lr * wd * model.weights[k]
model.biasdelta[k] = mom * model.biasdelta[k] - lr / num_samples / len(gpu) * multi_gpu_merge(biasgrads, 0, k)
model.weights[k] += model.weightdelta[k]
model.bias[k] += model.biasdelta[k]
if count % (len(gpu) * lazy_cycle) == 0:
print_training_accuracy(out, label, num_samples, 'Training')
print '---End of Epoch #', i, 'time:', time.time() - last
# do test
out, _, _ = bpprop(model, test_samples, test_labels)
print_training_accuracy(out, test_labels, num_test_samples, 'Testing')
def run(self):
(train_data,
test_data) = mnist_io.load_mb_from_mat(self.data_file, self.mb_size)
np.set_printoptions(linewidth=200)
num_test_samples = test_data[0].shape[0]
(test_samples,
test_labels) = map(lambda npdata: owl.from_numpy(npdata), test_data)
count = 1
owl.set_device(self.gpu)
for epoch in range(self.num_epochs):
print '---Start epoch #%d' % epoch
# train
for (mb_samples, mb_labels) in train_data:
num_samples = mb_samples.shape[0]
a1 = owl.from_numpy(mb_samples)
target = owl.from_numpy(mb_labels)
# ff
a2 = ele.relu(self.w1 * a1 + self.b1)
a3 = self.w2 * a2 + self.b2
# softmax & error
out = co.softmax(a3)
s3 = out - target
# bp
s2 = self.w2.trans() * s3
s2 = ele.relu_back(s2, a2)
# grad
gw1 = s2 * a1.trans() / num_samples
gb1 = s2.sum(1) / num_samples
gw2 = s3 * a2.trans() / num_samples
gb2 = s3.sum(1) / num_samples
# 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
if (count % 40 == 0):
correct = out.argmax(0) - target.argmax(0)
val = correct.to_numpy()
print 'Training error:', float(
np.count_nonzero(val)) / num_samples
count = count + 1
# test
a1 = test_samples
a2 = ele.relu(self.w1 * a1 + self.b1)
a3 = self.w2 * a2 + self.b2
correct = a3.argmax(0) - test_labels.argmax(0)
val = correct.to_numpy()
#print val
print 'Testing error:', float(
np.count_nonzero(val)) / num_test_samples
print '---Finish epoch #%d' % epoch
def run(self):
(train_data,
test_data) = mnist_io.load_mb_from_mat(self.data_file, self.mb_size)
np.set_printoptions(linewidth=200)
num_test_samples = test_data[0].shape[0]
(test_samples, test_labels) = test_data
count = 1
for epoch in range(self.num_epochs):
print '---Start epoch #%d' % epoch
# train
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
# softmax & error
out = softmax(a3)
s3 = out - target
# bp
s2 = np.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
# 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
if (count % 40 == 0):
correct = np.max_index(out, axis=0) - np.max_index(target,
axis=0)
print 'Training error:', float(
np.count_nonzero(correct)) / num_samples
count = count + 1
# test
a1 = test_samples.T
a2 = relu(np.dot(self.w1, a1) + self.b1)
a3 = np.dot(self.w2, a2) + self.b2
correct = np.max_index(a3, axis=0) - np.max_index(test_labels.T,
axis=0)
#print correct
print 'Testing error:', float(
np.count_nonzero(correct)) / num_test_samples
print '---Finish epoch #%d' % epoch
def run(self):
(train_data, test_data) = mnist_io.load_mb_from_mat(self.data_file, self.mb_size)
np.set_printoptions(linewidth=200)
num_test_samples = test_data[0].shape[0]
(test_samples, test_labels) = map(lambda npdata : owl.from_numpy(npdata), test_data)
count = 1
owl.set_device(self.gpu)
for epoch in range(self.num_epochs):
print '---Start epoch #%d' % epoch
# train
for (mb_samples, mb_labels) in train_data:
num_samples = mb_samples.shape[0]
a1 = owl.from_numpy(mb_samples)
target = owl.from_numpy(mb_labels)
# ff
a2 = ele.relu(self.w1 * a1 + self.b1)
a3 = self.w2 * a2 + self.b2
# softmax & error
out = co.softmax(a3)
s3 = out - target
# bp
s2 = self.w2.trans() * s3
s2 = ele.relu_back(s2, a2)
# grad
gw1 = s2 * a1.trans() / num_samples
gb1 = s2.sum(1) / num_samples
gw2 = s3 * a2.trans() / num_samples
gb2 = s3.sum(1) / num_samples
# 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
if (count % 40 == 0):
correct = out.max_index(0) - target.max_index(0)
val = correct.to_numpy()
print 'Training error:', float(np.count_nonzero(val)) / num_samples
count = count + 1
# test
a1 = test_samples
a2 = ele.relu(self.w1 * a1 + self.b1)
a3 = self.w2 * a2 + self.b2
correct = a3.max_index(0) - test_labels.max_index(0)
val = correct.to_numpy()
#print val
print 'Testing error:', float(np.count_nonzero(val)) / num_test_samples
print '---Finish epoch #%d' % epoch
def run(self):
(train_data, test_data) = mnist_io.load_mb_from_mat(self.data_file, self.mb_size)
np.set_printoptions(linewidth=200)
num_test_samples = test_data[0].shape[0]
(test_samples, test_labels) = test_data
count = 1
for epoch in range(self.num_epochs):
print '---Start epoch #%d' % epoch
# train
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
# softmax & error
out = softmax(a3)
s3 = out - target
# bp
s2 = np.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
# 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
if (count % 40 == 0):
correct = np.max_index(out, axis=0) - np.max_index(target, axis=0)
print 'Training error:', float(np.count_nonzero(correct)) / num_samples
count = count + 1
# test
a1 = test_samples.T
a2 = relu(np.dot(self.w1, a1) + self.b1)
a3 = np.dot(self.w2, a2) + self.b2
correct = np.max_index(a3, axis=0) - np.max_index(test_labels.T, axis=0)
#print correct
print 'Testing error:', float(np.count_nonzero(correct)) / num_test_samples
print '---Finish epoch #%d' % epoch
def train_network(filename, model, num_epochs=5, minibatch_size=256, lr=0.1, lr_decay= 0.95, mom=0.9, wd=5e-4):
# load data
(train_data, test_data) = mnist_io.load_mb_from_mat(filename, minibatch_size / len(devs))
num_test_samples = test_data[0].shape[0]
test_samples = owl.from_numpy(test_data[0]).reshape([28, 28, 1, num_test_samples])
test_labels = owl.from_numpy(test_data[1])
for i in xrange(num_epochs):
print "---Epoch #", i
last = time.time()
count = 0
weightgrads = [None] * len(devs)
biasgrads = [None] * len(devs)
for (mb_samples, mb_labels) in train_data:
count += 1
current_dev = count % len(devs)
owl.set_device(devs[current_dev])
num_samples = mb_samples.shape[0]
data = owl.from_numpy(mb_samples).reshape([28, 28, 1, num_samples])
label = owl.from_numpy(mb_labels)
#print "\t[{}]Train Data imported to minerva format".format(count)
out, weightgrads[current_dev], biasgrads[current_dev] = bpprop(model, data, label)
#print "\t[{}]Backprop complete".format(count)
# print "dev {}".format(current_dev)
if current_dev == 0:
# print "pre-merge"
for k in range(len(model.weights)):
model.weightdelta[k] = mom * model.weightdelta[k] - lr / num_samples / len(devs) * multi_dev_merge(weightgrads, 0, k) - lr * wd * model.weights[k]
# print "\t weight merge"
model.biasdelta[k] = mom * model.biasdelta[k] - lr / num_samples / len(devs) * multi_dev_merge(biasgrads, 0, k)
# print "\t bias merge"
model.weights[k] += model.weightdelta[k]
model.bias[k] += model.biasdelta[k]
# print "post-merge"
if count % (len(devs) * lazy_cycle) == 0:
print_training_accuracy(out, label, num_samples, 'Training ' + str(count))
owl.print_profiler_result()
print '---End of Epoch #', i, 'time:', time.time() - last
lr = lr*lr_decay
# do test
out, _, _ = bpprop(model, test_samples, test_labels)
print_training_accuracy(out, test_labels, num_test_samples, 'Testing')