def init_params(net, weight_path=None, is_train=False):
'''Init parameters randomly or from checkpoint file.
Args:
net, a constructed neural net
weight_path, checkpoint file path
is_train, if false, then a checkpoint file must be presented
'''
assert is_train is True or weight_path is not None, \
'must provide a checkpoint file for serving'
if weight_path is None:
for pname, pval in zip(net.param_names(), net.param_values()):
if 'conv' in pname and len(pval.shape) > 1:
initializer.gaussian(pval, 0, pval.shape[1])
elif 'dense' in pname:
if len(pval.shape) > 1:
initializer.gaussian(pval, 0, pval.shape[0])
else:
pval.set_value(0)
# init params from batch norm layer
elif 'mean' in pname or 'beta' in pname:
pval.set_value(0)
elif 'var' in pname:
pval.set_value(1)
elif 'gamma' in pname:
initializer.uniform(pval, 0, 1)
else:
net.load(weight_path, use_pickle=True)
def init_params(net, weight_path=None, is_train=False):
'''Init parameters randomly or from checkpoint file.
Args:
net, a constructed neural net
weight_path, checkpoint file path
is_train, if false, then a checkpoint file must be presented
'''
assert is_train is True or weight_path is not None, \
'must provide a checkpoint file for serving'
if weight_path is None:
for pname, pval in zip(net.param_names(), net.param_values()):
if 'conv' in pname and len(pval.shape) > 1:
initializer.gaussian(pval, 0, pval.shape[1])
elif 'dense' in pname:
if len(pval.shape) > 1:
initializer.gaussian(pval, 0, pval.shape[0])
else:
pval.set_value(0)
# init params from batch norm layer
elif 'mean' in pname or 'beta' in pname:
pval.set_value(0)
elif 'var' in pname:
pval.set_value(1)
elif 'gamma' in pname:
initializer.uniform(pval, 0, 1)
else:
net.load(weight_path, use_pickle=True)
def create_net(use_cpu=False):
if use_cpu:
layer.engine = 'singacpp'
net = ffnet.FeedForwardNet(loss.SoftmaxCrossEntropy(), metric.Accuracy())
ConvBnReLU(net, 'conv1_1', 64, (3, 32, 32))
net.add(layer.Dropout('drop1', 0.3))
ConvBnReLU(net, 'conv1_2', 64)
net.add(layer.MaxPooling2D('pool1', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv2_1', 128)
net.add(layer.Dropout('drop2_1', 0.4))
ConvBnReLU(net, 'conv2_2', 128)
net.add(layer.MaxPooling2D('pool2', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv3_1', 256)
net.add(layer.Dropout('drop3_1', 0.4))
ConvBnReLU(net, 'conv3_2', 256)
net.add(layer.Dropout('drop3_2', 0.4))
ConvBnReLU(net, 'conv3_3', 256)
net.add(layer.MaxPooling2D('pool3', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv4_1', 512)
net.add(layer.Dropout('drop4_1', 0.4))
ConvBnReLU(net, 'conv4_2', 512)
net.add(layer.Dropout('drop4_2', 0.4))
ConvBnReLU(net, 'conv4_3', 512)
net.add(layer.MaxPooling2D('pool4', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv5_1', 512)
net.add(layer.Dropout('drop5_1', 0.4))
ConvBnReLU(net, 'conv5_2', 512)
net.add(layer.Dropout('drop5_2', 0.4))
ConvBnReLU(net, 'conv5_3', 512)
net.add(layer.MaxPooling2D('pool5', 2, 2, border_mode='valid'))
net.add(layer.Flatten('flat'))
net.add(layer.Dropout('drop_flat', 0.5))
net.add(layer.Dense('ip1', 512))
net.add(layer.BatchNormalization('batchnorm_ip1'))
net.add(layer.Activation('relu_ip1'))
net.add(layer.Dropout('drop_ip2', 0.5))
net.add(layer.Dense('ip2', 10))
print('Start intialization............')
for (p, name) in zip(net.param_values(), net.param_names()):
print(name, p.shape)
if 'mean' in name or 'beta' in name:
p.set_value(0.0)
elif 'var' in name:
p.set_value(1.0)
elif 'gamma' in name:
initializer.uniform(p, 0, 1)
elif len(p.shape) > 1:
if 'conv' in name:
initializer.gaussian(p, 0, 3 * 3 * p.shape[0])
else:
p.gaussian(0, 0.02)
else:
p.set_value(0)
print(name, p.l1())
return net
def create_net(use_cpu=False):
if use_cpu:
layer.engine = 'singacpp'
net = ffnet.FeedForwardNet(loss.SoftmaxCrossEntropy(), metric.Accuracy())
ConvBnReLU(net, 'conv1_1', 64, (3, 32, 32))
net.add(layer.Dropout('drop1', 0.3))
ConvBnReLU(net, 'conv1_2', 64)
net.add(layer.MaxPooling2D('pool1', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv2_1', 128)
net.add(layer.Dropout('drop2_1', 0.4))
ConvBnReLU(net, 'conv2_2', 128)
net.add(layer.MaxPooling2D('pool2', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv3_1', 256)
net.add(layer.Dropout('drop3_1', 0.4))
ConvBnReLU(net, 'conv3_2', 256)
net.add(layer.Dropout('drop3_2', 0.4))
ConvBnReLU(net, 'conv3_3', 256)
net.add(layer.MaxPooling2D('pool3', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv4_1', 512)
net.add(layer.Dropout('drop4_1', 0.4))
ConvBnReLU(net, 'conv4_2', 512)
net.add(layer.Dropout('drop4_2', 0.4))
ConvBnReLU(net, 'conv4_3', 512)
net.add(layer.MaxPooling2D('pool4', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv5_1', 512)
net.add(layer.Dropout('drop5_1', 0.4))
ConvBnReLU(net, 'conv5_2', 512)
net.add(layer.Dropout('drop5_2', 0.4))
ConvBnReLU(net, 'conv5_3', 512)
net.add(layer.MaxPooling2D('pool5', 2, 2, border_mode='valid'))
net.add(layer.Flatten('flat'))
net.add(layer.Dropout('drop_flat', 0.5))
net.add(layer.Dense('ip1', 512))
net.add(layer.BatchNormalization('batchnorm_ip1'))
net.add(layer.Activation('relu_ip1'))
net.add(layer.Dropout('drop_ip2', 0.5))
net.add(layer.Dense('ip2', 10))
print 'Start intialization............'
for (p, name) in zip(net.param_values(), net.param_names()):
print name, p.shape
if 'mean' in name or 'beta' in name:
p.set_value(0.0)
elif 'var' in name:
p.set_value(1.0)
elif 'gamma' in name:
initializer.uniform(p, 0, 1)
elif len(p.shape) > 1:
if 'conv' in name:
initializer.gaussian(p, 0, 3 * 3 * p.shape[0])
else:
p.gaussian(0, 0.02)
else:
p.set_value(0)
print name, p.l1()
return net
def create_net(input_shape, use_cpu=False):
if use_cpu:
layer.engine = 'singacpp'
net = ffnet.FeedForwardNet(loss.SoftmaxCrossEntropy(), metric.Accuracy())
net.add(
layer.Conv2D('conv1',
nb_kernels=32,
kernel=7,
stride=3,
pad=1,
input_sample_shape=input_shape))
net.add(layer.Activation('relu1'))
net.add(layer.MaxPooling2D('pool1', 2, 2, border_mode='valid'))
net.add(layer.Conv2D('conv2', nb_kernels=64, kernel=5, stride=3))
net.add(layer.Activation('relu2'))
net.add(layer.MaxPooling2D('pool2', 2, 2, border_mode='valid'))
net.add(layer.Conv2D('conv3', nb_kernels=128, kernel=3, stride=1, pad=2))
net.add(layer.Activation('relu3'))
net.add(layer.MaxPooling2D('pool3', 2, 2, border_mode='valid'))
net.add(layer.Conv2D('conv4', nb_kernels=256, kernel=3, stride=1))
net.add(layer.Activation('relu4'))
net.add(layer.MaxPooling2D('pool4', 2, 2, border_mode='valid'))
net.add(layer.Flatten('flat'))
net.add(layer.Dense('ip5', 256))
net.add(layer.Activation('relu5'))
net.add(layer.Dense('ip6', 16))
net.add(layer.Activation('relu6'))
net.add(layer.Dense('ip7', 2))
print 'Parameter intialization............'
for (p, name) in zip(net.param_values(), net.param_names()):
print name, p.shape
if 'mean' in name or 'beta' in name:
p.set_value(0.0)
elif 'var' in name:
p.set_value(1.0)
elif 'gamma' in name:
initializer.uniform(p, 0, 1)
elif len(p.shape) > 1:
if 'conv' in name:
initializer.gaussian(p, 0, p.size())
else:
p.gaussian(0, 0.02)
else:
p.set_value(0)
print name, p.l1()
return net
def create_net(input_shape, use_cpu=False):
if use_cpu:
layer.engine = 'singacpp'
net = ffnet.FeedForwardNet(loss.SoftmaxCrossEntropy(), metric.Accuracy())
ConvBnReLUPool(net, 'conv1', 32, input_shape)
ConvBnReLUPool(net, 'conv2', 64)
ConvBnReLUPool(net, 'conv3', 128)
ConvBnReLUPool(net, 'conv4', 128)
ConvBnReLUPool(net, 'conv5', 256)
ConvBnReLUPool(net, 'conv6', 256)
ConvBnReLUPool(net, 'conv7', 512)
ConvBnReLUPool(net, 'conv8', 512)
net.add(layer.Flatten('flat'))
net.add(layer.Dense('ip1', 256))
net.add(layer.BatchNormalization('bn1'))
net.add(layer.Activation('relu1'))
net.add(layer.Dropout('dropout1', 0.2))
net.add(layer.Dense('ip2', 16))
net.add(layer.BatchNormalization('bn2'))
net.add(layer.Activation('relu2'))
net.add(layer.Dropout('dropout2', 0.2))
net.add(layer.Dense('ip3', 2))
print 'Parameter intialization............'
for (p, name) in zip(net.param_values(), net.param_names()):
print name, p.shape
if 'mean' in name or 'beta' in name:
p.set_value(0.0)
elif 'var' in name:
p.set_value(1.0)
elif 'gamma' in name:
initializer.uniform(p, 0, 1)
elif len(p.shape) > 1:
if 'conv' in name:
initializer.gaussian(p, 0, p.size())
else:
p.gaussian(0, 0.02)
else:
p.set_value(0)
print name, p.l1()
return net
def create_net(use_cpu=False):
if use_cpu:
layer.engine = 'singacpp'
net = ffnet.FeedForwardNet(loss.SoftmaxCrossEntropy(), metric.Accuracy())
net.add(
layer.Conv2D("conv1", 16, 3, 1, pad=1, input_sample_shape=(3, 32, 32)))
net.add(layer.BatchNormalization("bn1"))
net.add(layer.Activation("relu1"))
Block(net, "2a", 16, 1)
Block(net, "2b", 16, 1)
Block(net, "2c", 16, 1)
Block(net, "3a", 32, 2)
Block(net, "3b", 32, 1)
Block(net, "3c", 32, 1)
Block(net, "4a", 64, 2)
Block(net, "4b", 64, 1)
Block(net, "4c", 64, 1)
net.add(layer.AvgPooling2D("pool4", 8, 8, border_mode='valid'))
net.add(layer.Flatten('flat'))
net.add(layer.Dense('ip5', 10))
print 'Start intialization............'
for (p, name) in zip(net.param_values(), net.param_names()):
# print name, p.shape
if 'mean' in name or 'beta' in name:
p.set_value(0.0)
elif 'var' in name:
p.set_value(1.0)
elif 'gamma' in name:
initializer.uniform(p, 0, 1)
elif len(p.shape) > 1:
if 'conv' in name:
# initializer.gaussian(p, 0, math.sqrt(2.0/p.shape[1]))
initializer.gaussian(p, 0, 9.0 * p.shape[0])
else:
initializer.uniform(p, p.shape[0], p.shape[1])
else:
p.set_value(0)
# print name, p.l1()
return net
def create_net(use_cpu=False):
if use_cpu:
layer.engine = 'singacpp'
net = ffnet.FeedForwardNet(loss.SoftmaxCrossEntropy(), metric.Accuracy())
net.add(layer.Conv2D("conv1", 16, 3, 1, pad=1, input_sample_shape=(3, 32, 32)))
net.add(layer.BatchNormalization("bn1"))
net.add(layer.Activation("relu1"))
Block(net, "2a", 16, 1)
Block(net, "2b", 16, 1)
Block(net, "2c", 16, 1)
Block(net, "3a", 32, 2)
Block(net, "3b", 32, 1)
Block(net, "3c", 32, 1)
Block(net, "4a", 64, 2)
Block(net, "4b", 64, 1)
Block(net, "4c", 64, 1)
net.add(layer.AvgPooling2D("pool4", 8, 8, border_mode='valid'))
net.add(layer.Flatten('flat'))
net.add(layer.Dense('ip5', 10))
print('Start intialization............')
for (p, name) in zip(net.param_values(), net.param_names()):
# print name, p.shape
if 'mean' in name or 'beta' in name:
p.set_value(0.0)
elif 'var' in name:
p.set_value(1.0)
elif 'gamma' in name:
initializer.uniform(p, 0, 1)
elif len(p.shape) > 1:
if 'conv' in name:
# initializer.gaussian(p, 0, math.sqrt(2.0/p.shape[1]))
initializer.gaussian(p, 0, 9.0 * p.shape[0])
else:
initializer.uniform(p, p.shape[0], p.shape[1])
else:
p.set_value(0)
# print name, p.l1()
return net
def init_params(net, weight_path=None):
if weight_path == None:
for pname, pval in zip(net.param_names(), net.param_values()):
print(pname, pval.shape)
if 'conv' in pname and len(pval.shape) > 1:
initializer.gaussian(pval, 0, pval.shape[1])
elif 'dense' in pname:
if len(pval.shape) > 1:
initializer.gaussian(pval, 0, pval.shape[0])
else:
pval.set_value(0)
# init params from batch norm layer
elif 'mean' in pname or 'beta' in pname:
pval.set_value(0)
elif 'var' in pname:
pval.set_value(1)
elif 'gamma' in pname:
initializer.uniform(pval, 0, 1)
else:
net.load(weight_path, use_pickle='pickle' in weight_path)
def init_params(net, weight_path=None):
if weight_path == None:
for pname, pval in zip(net.param_names(), net.param_values()):
print(pname, pval.shape)
if 'conv' in pname and len(pval.shape) > 1:
initializer.gaussian(pval, 0, pval.shape[1])
elif 'dense' in pname:
if len(pval.shape) > 1:
initializer.gaussian(pval, 0, pval.shape[0])
else:
pval.set_value(0)
# init params from batch norm layer
elif 'mean' in pname or 'beta' in pname:
pval.set_value(0)
elif 'var' in pname:
pval.set_value(1)
elif 'gamma' in pname:
initializer.uniform(pval, 0, 1)
else:
net.load(weight_path, use_pickle = 'pickle' in weight_path)
def train(data_file, use_gpu, num_epoch=10, batch_size=100):
print 'Start intialization............'
lr = 0.1 # Learning rate
weight_decay = 0.0002
hdim = 1000
vdim = 784
opt = optimizer.SGD(momentum=0.8, weight_decay=weight_decay)
tweight = tensor.Tensor((vdim, hdim))
tweight.gaussian(0.0, 0.1)
tvbias = tensor.from_numpy(np.zeros(vdim, dtype = np.float32))
thbias = tensor.from_numpy(np.zeros(hdim, dtype = np.float32))
opt = optimizer.SGD(momentum=0.5, weight_decay=weight_decay)
print 'Loading data ..................'
train_x, valid_x = load_train_data(data_file)
if use_gpu:
dev = device.create_cuda_gpu()
else:
dev = device.get_default_device()
for t in [tweight, tvbias, thbias]:
t.to_device(dev)
num_train_batch = train_x.shape[0] / batch_size
print "num_train_batch = %d " % (num_train_batch)
for epoch in range(num_epoch):
trainerrorsum = 0.0
print 'Epoch %d' % epoch
for b in range(num_train_batch):
# positive phase
tdata = tensor.from_numpy(
train_x[(b * batch_size):((b + 1) * batch_size), : ])
tdata.to_device(dev)
tposhidprob = tensor.mult(tdata, tweight)
tposhidprob.add_row(thbias)
tposhidprob = tensor.sigmoid(tposhidprob)
tposhidrandom = tensor.Tensor(tposhidprob.shape, dev)
tposhidrandom.uniform(0.0, 1.0)
tposhidsample = tensor.gt(tposhidprob, tposhidrandom)
# negative phase
tnegdata = tensor.mult(tposhidsample, tweight.T())
tnegdata.add_row(tvbias)
tnegdata = tensor.sigmoid(tnegdata)
tneghidprob = tensor.mult(tnegdata, tweight)
tneghidprob.add_row(thbias)
tneghidprob = tensor.sigmoid(tneghidprob)
error = tensor.sum(tensor.square((tdata - tnegdata)))
trainerrorsum = error + trainerrorsum
tgweight = tensor.mult(tnegdata.T(), tneghidprob) -\
tensor.mult(tdata.T(), tposhidprob)
tgvbias = tensor.sum(tnegdata, 0) - tensor.sum(tdata, 0)
tghbias = tensor.sum(tneghidprob, 0) - tensor.sum(tposhidprob, 0)
opt.apply_with_lr(epoch, lr / batch_size, tgweight, tweight, 'w')
opt.apply_with_lr(epoch, lr / batch_size, tgvbias, tvbias, 'vb')
opt.apply_with_lr(epoch, lr / batch_size, tghbias, thbias, 'hb')
print 'training errorsum = %f' % (trainerrorsum)
tvaliddata = tensor.from_numpy(valid_x)
tvaliddata.to_device(dev)
tvalidposhidprob = tensor.mult(tvaliddata, tweight)
tvalidposhidprob.add_row(thbias)
tvalidposhidprob = tensor.sigmoid(tvalidposhidprob)
tvalidposhidrandom = tensor.Tensor(tvalidposhidprob.shape, dev)
initializer.uniform(tvalidposhidrandom, 0.0, 1.0)
tvalidposhidsample = tensor.gt(tvalidposhidprob, tvalidposhidrandom)
tvalidnegdata = tensor.mult(tvalidposhidsample, tweight.T())
tvalidnegdata.add_row(tvbias)
tvalidnegdata = tensor.sigmoid(tvalidnegdata)
validerrorsum = tensor.sum(tensor.square((tvaliddata - tvalidnegdata)))
print 'valid errorsum = %f' % (validerrorsum)
def train(data_file, use_gpu, num_epoch=10, batch_size=100):
print('Start intialization............')
lr = 0.1 # Learning rate
weight_decay = 0.0002
hdim = 1000
vdim = 784
tweight = tensor.Tensor((vdim, hdim))
tweight.gaussian(0.0, 0.1)
tvbias = tensor.from_numpy(np.zeros(vdim, dtype=np.float32))
thbias = tensor.from_numpy(np.zeros(hdim, dtype=np.float32))
opt = optimizer.SGD(momentum=0.5, weight_decay=weight_decay)
print('Loading data ..................')
train_x, valid_x = load_train_data(data_file)
if use_gpu:
dev = device.create_cuda_gpu()
else:
dev = device.get_default_device()
for t in [tweight, tvbias, thbias]:
t.to_device(dev)
num_train_batch = train_x.shape[0] // batch_size
print("num_train_batch = %d " % (num_train_batch))
for epoch in range(num_epoch):
trainerrorsum = 0.0
print('Epoch %d' % epoch)
for b in range(num_train_batch):
# positive phase
tdata = tensor.from_numpy(
train_x[(b * batch_size):((b + 1) * batch_size), :])
tdata.to_device(dev)
tposhidprob = tensor.mult(tdata, tweight)
tposhidprob = tposhidprob + thbias
tposhidprob = tensor.sigmoid(tposhidprob)
tposhidrandom = tensor.Tensor(tposhidprob.shape, dev)
tposhidrandom.uniform(0.0, 1.0)
tposhidsample = tensor.gt(tposhidprob, tposhidrandom)
# negative phase
tnegdata = tensor.mult(tposhidsample, tweight.T())
tnegdata = tnegdata + tvbias
tnegdata = tensor.sigmoid(tnegdata)
tneghidprob = tensor.mult(tnegdata, tweight)
tneghidprob = tneghidprob + thbias
tneghidprob = tensor.sigmoid(tneghidprob)
error = tensor.sum(tensor.square((tdata - tnegdata)))
trainerrorsum = error + trainerrorsum
tgweight = tensor.mult(tnegdata.T(), tneghidprob) \
- tensor.mult(tdata.T(), tposhidprob)
tgvbias = tensor.sum(tnegdata, 0) - tensor.sum(tdata, 0)
tghbias = tensor.sum(tneghidprob, 0) - tensor.sum(tposhidprob, 0)
opt.apply_with_lr(epoch, lr / batch_size, tgweight, tweight, 'w')
opt.apply_with_lr(epoch, lr / batch_size, tgvbias, tvbias, 'vb')
opt.apply_with_lr(epoch, lr / batch_size, tghbias, thbias, 'hb')
print('training erroraverage = %f' %
(tensor.to_numpy(trainerrorsum) / train_x.shape[0]))
tvaliddata = tensor.from_numpy(valid_x)
tvaliddata.to_device(dev)
tvalidposhidprob = tensor.mult(tvaliddata, tweight)
tvalidposhidprob = tvalidposhidprob + thbias
tvalidposhidprob = tensor.sigmoid(tvalidposhidprob)
tvalidposhidrandom = tensor.Tensor(tvalidposhidprob.shape, dev)
initializer.uniform(tvalidposhidrandom, 0.0, 1.0)
tvalidposhidsample = tensor.gt(tvalidposhidprob, tvalidposhidrandom)
tvalidnegdata = tensor.mult(tvalidposhidsample, tweight.T())
tvalidnegdata = tvalidnegdata + tvbias
tvalidnegdata = tensor.sigmoid(tvalidnegdata)
validerrorsum = tensor.sum(tensor.square((tvaliddata - tvalidnegdata)))
print('valid erroraverage = %f' %
(tensor.to_numpy(validerrorsum) / valid_x.shape[0]))
def train(data,
max_epoch,
hidden_size=100,
seq_length=100,
batch_size=16,
num_stacks=1,
dropout=0.5,
model_path='model'):
# SGD with L2 gradient normalization
opt = optimizer.RMSProp(constraint=optimizer.L2Constraint(5))
cuda = device.create_cuda_gpu()
rnn = layer.LSTM(name='lstm',
hidden_size=hidden_size,
num_stacks=num_stacks,
dropout=dropout,
input_sample_shape=(data.vocab_size, ))
rnn.to_device(cuda)
print 'created rnn'
rnn_w = rnn.param_values()[0]
rnn_w.uniform(-0.08, 0.08) # init all rnn parameters
print 'rnn weight l1 = %f' % (rnn_w.l1())
dense = layer.Dense('dense',
data.vocab_size,
input_sample_shape=(hidden_size, ))
dense.to_device(cuda)
dense_w = dense.param_values()[0]
dense_b = dense.param_values()[1]
print 'dense w ', dense_w.shape
print 'dense b ', dense_b.shape
initializer.uniform(dense_w, dense_w.shape[0], 0)
print 'dense weight l1 = %f' % (dense_w.l1())
dense_b.set_value(0)
print 'dense b l1 = %f' % (dense_b.l1())
g_dense_w = tensor.Tensor(dense_w.shape, cuda)
g_dense_b = tensor.Tensor(dense_b.shape, cuda)
lossfun = loss.SoftmaxCrossEntropy()
for epoch in range(max_epoch):
train_loss = 0
for b in range(data.num_train_batch):
batch = data.train_dat[b * batch_size:(b + 1) * batch_size]
inputs, labels = convert(batch, batch_size, seq_length,
data.vocab_size, cuda)
inputs.append(tensor.Tensor())
inputs.append(tensor.Tensor())
outputs = rnn.forward(model_pb2.kTrain, inputs)[0:-2]
grads = []
batch_loss = 0
g_dense_w.set_value(0.0)
g_dense_b.set_value(0.0)
for output, label in zip(outputs, labels):
act = dense.forward(model_pb2.kTrain, output)
lvalue = lossfun.forward(model_pb2.kTrain, act, label)
batch_loss += lvalue.l1()
grad = lossfun.backward()
grad /= batch_size
grad, gwb = dense.backward(model_pb2.kTrain, grad)
grads.append(grad)
g_dense_w += gwb[0]
g_dense_b += gwb[1]
# print output.l1(), act.l1()
utils.update_progress(
b * 1.0 / data.num_train_batch,
'training loss = %f' % (batch_loss / seq_length))
train_loss += batch_loss
grads.append(tensor.Tensor())
grads.append(tensor.Tensor())
g_rnn_w = rnn.backward(model_pb2.kTrain, grads)[1][0]
dense_w, dense_b = dense.param_values()
opt.apply_with_lr(epoch, get_lr(epoch), g_rnn_w, rnn_w, 'rnnw')
opt.apply_with_lr(epoch, get_lr(epoch), g_dense_w, dense_w,
'dense_w')
opt.apply_with_lr(epoch, get_lr(epoch), g_dense_b, dense_b,
'dense_b')
print '\nEpoch %d, train loss is %f' % \
(epoch, train_loss / data.num_train_batch / seq_length)
eval_loss = 0
for b in range(data.num_test_batch):
batch = data.val_dat[b * batch_size:(b + 1) * batch_size]
inputs, labels = convert(batch, batch_size, seq_length,
data.vocab_size, cuda)
inputs.append(tensor.Tensor())
inputs.append(tensor.Tensor())
outputs = rnn.forward(model_pb2.kEval, inputs)[0:-2]
for output, label in zip(outputs, labels):
output = dense.forward(model_pb2.kEval, output)
eval_loss += lossfun.forward(model_pb2.kEval, output,
label).l1()
print 'Epoch %d, evaluation loss is %f' % \
(epoch, eval_loss / data.num_test_batch / seq_length)
if (epoch + 1) % 30 == 0:
# checkpoint the file model
with open('%s_%d.bin' % (model_path, epoch), 'wb') as fd:
print 'saving model to %s' % model_path
d = {}
for name, w in zip(['rnn_w', 'dense_w', 'dense_b'],
[rnn_w, dense_w, dense_b]):
w.to_host()
d[name] = tensor.to_numpy(w)
w.to_device(cuda)
d['idx_to_char'] = data.idx_to_char
d['char_to_idx'] = data.char_to_idx
d['hidden_size'] = hidden_size
d['num_stacks'] = num_stacks
d['dropout'] = dropout
pickle.dump(d, fd)
def train(data, max_epoch, hidden_size=100, seq_length=100, batch_size=16,
num_stacks=1, dropout=0.5, model_path='model'):
# SGD with L2 gradient normalization
opt = optimizer.RMSProp(constraint=optimizer.L2Constraint(5))
cuda = device.create_cuda_gpu()
rnn = layer.LSTM(
name='lstm',
hidden_size=hidden_size,
num_stacks=num_stacks,
dropout=dropout,
input_sample_shape=(
data.vocab_size,
))
rnn.to_device(cuda)
print 'created rnn'
rnn_w = rnn.param_values()[0]
rnn_w.uniform(-0.08, 0.08) # init all rnn parameters
print 'rnn weight l1 = %f' % (rnn_w.l1())
dense = layer.Dense(
'dense',
data.vocab_size,
input_sample_shape=(
hidden_size,
))
dense.to_device(cuda)
dense_w = dense.param_values()[0]
dense_b = dense.param_values()[1]
print 'dense w ', dense_w.shape
print 'dense b ', dense_b.shape
initializer.uniform(dense_w, dense_w.shape[0], 0)
print 'dense weight l1 = %f' % (dense_w.l1())
dense_b.set_value(0)
print 'dense b l1 = %f' % (dense_b.l1())
g_dense_w = tensor.Tensor(dense_w.shape, cuda)
g_dense_b = tensor.Tensor(dense_b.shape, cuda)
lossfun = loss.SoftmaxCrossEntropy()
for epoch in range(max_epoch):
train_loss = 0
for b in range(data.num_train_batch):
batch = data.train_dat[b * batch_size: (b + 1) * batch_size]
inputs, labels = convert(batch, batch_size, seq_length,
data.vocab_size, cuda)
inputs.append(tensor.Tensor())
inputs.append(tensor.Tensor())
outputs = rnn.forward(model_pb2.kTrain, inputs)[0:-2]
grads = []
batch_loss = 0
g_dense_w.set_value(0.0)
g_dense_b.set_value(0.0)
for output, label in zip(outputs, labels):
act = dense.forward(model_pb2.kTrain, output)
lvalue = lossfun.forward(model_pb2.kTrain, act, label)
batch_loss += lvalue.l1()
grad = lossfun.backward()
grad /= batch_size
grad, gwb = dense.backward(model_pb2.kTrain, grad)
grads.append(grad)
g_dense_w += gwb[0]
g_dense_b += gwb[1]
# print output.l1(), act.l1()
utils.update_progress(
b * 1.0 / data.num_train_batch, 'training loss = %f' %
(batch_loss / seq_length))
train_loss += batch_loss
grads.append(tensor.Tensor())
grads.append(tensor.Tensor())
g_rnn_w = rnn.backward(model_pb2.kTrain, grads)[1][0]
dense_w, dense_b = dense.param_values()
opt.apply_with_lr(epoch, get_lr(epoch), g_rnn_w, rnn_w, 'rnnw')
opt.apply_with_lr(
epoch, get_lr(epoch),
g_dense_w, dense_w, 'dense_w')
opt.apply_with_lr(
epoch, get_lr(epoch),
g_dense_b, dense_b, 'dense_b')
print '\nEpoch %d, train loss is %f' % \
(epoch, train_loss / data.num_train_batch / seq_length)
eval_loss = 0
for b in range(data.num_test_batch):
batch = data.val_dat[b * batch_size: (b + 1) * batch_size]
inputs, labels = convert(batch, batch_size, seq_length,
data.vocab_size, cuda)
inputs.append(tensor.Tensor())
inputs.append(tensor.Tensor())
outputs = rnn.forward(model_pb2.kEval, inputs)[0:-2]
for output, label in zip(outputs, labels):
output = dense.forward(model_pb2.kEval, output)
eval_loss += lossfun.forward(model_pb2.kEval,
output, label).l1()
print 'Epoch %d, evaluation loss is %f' % \
(epoch, eval_loss / data.num_test_batch / seq_length)
if (epoch + 1) % 30 == 0:
# checkpoint the file model
with open('%s_%d.bin' % (model_path, epoch), 'wb') as fd:
print 'saving model to %s' % model_path
d = {}
for name, w in zip(
['rnn_w', 'dense_w', 'dense_b'],
[rnn_w, dense_w, dense_b]):
w.to_host()
d[name] = tensor.to_numpy(w)
w.to_device(cuda)
d['idx_to_char'] = data.idx_to_char
d['char_to_idx'] = data.char_to_idx
d['hidden_size'] = hidden_size
d['num_stacks'] = num_stacks
d['dropout'] = dropout
pickle.dump(d, fd)
def create_net(input_shape, use_cpu=False):
if use_cpu:
layer.engine = 'singacpp'
net = ffnet.FeedForwardNet(loss.SoftmaxCrossEntropy(), metric.Accuracy())
ConvBnReLU(net, 'conv1_1', 64, input_shape)
#net.add(layer.Dropout('drop1', 0.3))
net.add(layer.MaxPooling2D('pool0', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv1_2', 128)
net.add(layer.MaxPooling2D('pool1', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv2_1', 128)
net.add(layer.Dropout('drop2_1', 0.4))
ConvBnReLU(net, 'conv2_2', 128)
net.add(layer.MaxPooling2D('pool2', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv3_1', 256)
net.add(layer.Dropout('drop3_1', 0.4))
ConvBnReLU(net, 'conv3_2', 256)
net.add(layer.Dropout('drop3_2', 0.4))
ConvBnReLU(net, 'conv3_3', 256)
net.add(layer.MaxPooling2D('pool3', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv4_1', 256)
net.add(layer.Dropout('drop4_1', 0.4))
ConvBnReLU(net, 'conv4_2', 256)
net.add(layer.Dropout('drop4_2', 0.4))
ConvBnReLU(net, 'conv4_3', 256)
net.add(layer.MaxPooling2D('pool4', 2, 2, border_mode='valid'))
ConvBnReLU(net, 'conv5_1', 512)
net.add(layer.Dropout('drop5_1', 0.4))
ConvBnReLU(net, 'conv5_2', 512)
net.add(layer.Dropout('drop5_2', 0.4))
ConvBnReLU(net, 'conv5_3', 512)
net.add(layer.MaxPooling2D('pool5', 2, 2, border_mode='valid'))
#ConvBnReLU(net, 'conv6_1', 512)
#net.add(layer.Dropout('drop6_1', 0.4))
#ConvBnReLU(net, 'conv6_2', 512)
#net.add(layer.Dropout('drop6_2', 0.4))
#ConvBnReLU(net, 'conv6_3', 512)
#net.add(layer.MaxPooling2D('pool6', 2, 2, border_mode='valid'))
#ConvBnReLU(net, 'conv7_1', 512)
#net.add(layer.Dropout('drop7_1', 0.4))
#ConvBnReLU(net, 'conv7_2', 512)
#net.add(layer.Dropout('drop7_2', 0.4))
#ConvBnReLU(net, 'conv7_3', 512)
#net.add(layer.MaxPooling2D('pool7', 2, 2, border_mode='valid'))
net.add(layer.Flatten('flat'))
net.add(layer.Dense('ip1', 256))
net.add(layer.BatchNormalization('bn1'))
net.add(layer.Activation('relu1'))
net.add(layer.Dropout('dropout1', 0.2))
net.add(layer.Dense('ip2', 16))
net.add(layer.BatchNormalization('bn2'))
net.add(layer.Activation('relu2'))
net.add(layer.Dropout('dropout2', 0.2))
net.add(layer.Dense('ip3', 2))
print 'Parameter intialization............'
for (p, name) in zip(net.param_values(), net.param_names()):
print name, p.shape
if 'mean' in name or 'beta' in name:
p.set_value(0.0)
elif 'var' in name:
p.set_value(1.0)
elif 'gamma' in name:
initializer.uniform(p, 0, 1)
elif len(p.shape) > 1:
if 'conv' in name:
initializer.gaussian(p, 0, p.size())
else:
p.gaussian(0, 0.02)
else:
p.set_value(0)
print name, p.l1()
return net
hidden_size=64,
num_stacks=5,
dropout=0.5,
input_sample_shape=(vocab_size, ))
encoder.to_device(cuda)
decoder.to_device(cuda)
encoder_w = encoder.param_values()[0]
encoder_w.uniform(-0.08, 0.08)
decoder_w = decoder.param_values()[0]
decoder_w.uniform(-0.08, 0.08)
dense = layer.Dense('dense', vocab_size, input_sample_shape=(64, ))
dense.to_device(cuda)
dense_w = dense.param_values()[0]
dense_b = dense.param_values()[1]
initializer.uniform(dense_w, dense_w.shape[0], 0)
dense_b.set_value(0)
#g_encoder_w = tensor.Tensor(encoder_w.shape, cuda)
#g_encoder_w.set_value(0.0)
g_dense_w = tensor.Tensor(dense_w.shape, cuda)
g_dense_b = tensor.Tensor(dense_b.shape, cuda)
lossfun = loss.SoftmaxCrossEntropy()
batch_size = 50
maxlength = 22
num_train_batch = 5000
num_epoch = 5
metadata, idx_q, idx_a = load_data()
trainlosslist = np.zeros(num_epoch)
for epoch in range(num_epoch):