notwa/optim
1
0
Fork 0
Code Issues Pull requests Releases Wiki Activity

update mnist example with new features

Browse source
This commit is contained in:
Connor Olding 2017-04-11 04:48:53 +00:00
parent d08b5b91a1
commit 87ffa014ca
1 changed files with 58 additions and 14 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

72
optim_nn_mnist.py
View file

@ -17,33 +17,48 @@ if use_emnist:
sgdr = True
restart_decay = 0.5
CLR = SineCLR
n_dense = 0
n_denses = 2
new_dims = (28, 28)
activation = GeluApprox
reg = None
final_reg = None
actreg_lamb = None
load_fn = None
save_fn = 'emnist.h5'
log_fn = 'emnist_losses.npz'
fn = 'emnist-balanced.npz'
mnist_dim = 28
mnist_classes = 47
else:
lr = 0.0032
epochs = 125
starts = 5
epochs = 60
starts = 3
bs = 200
activation = Relu
sgdr = False
sgdr = True
restart_decay = 0.5
CLR = SineCLR
n_dense = 1
n_dense = 2
n_denses = 1
new_dims = (4, 12)
activation = Relu
reg = L1L2(3.2e-5, 3.2e-4)
final_reg = L1L2(3.2e-5, 1e-3)
actreg_lamb = None # 1e-3
load_fn = None
save_fn = 'mnist.h5'
log_fn = 'mnist_losses.npz'
fn = 'mnist.npz'
mnist_dim = 28
mnist_classes = 10
@ -74,23 +89,39 @@ def get_mnist(fn='mnist.npz'):
inputs, outputs, valid_inputs, valid_outputs = get_mnist(fn)
def actreg(y):
if not actreg_lamb:
return y
lamb = actreg_lamb # * np.prod(y.output_shape)
reg = SaturateRelu(lamb)
act = ActivityRegularizer(reg)
reg.lamb_orig = reg.lamb # HACK
return y.feed(act)
x = Input(shape=inputs.shape[1:])
y = x
y = y.feed(Reshape(new_shape=(mnist_dim, mnist_dim)))
for i in range(n_denses):
if i > 0:
y = actreg(y)
y = y.feed(activation())
y = y.feed(Denses(new_dims[0], axis=0, init=init_he_normal))
y = y.feed(Denses(new_dims[1], axis=1, init=init_he_normal))
y = y.feed(Denses(new_dims[0], axis=0, init=init_he_normal,
reg_w=reg, reg_b=reg))
y = y.feed(Denses(new_dims[1], axis=1, init=init_he_normal,
reg_w=reg, reg_b=reg))
y = y.feed(Flatten())
for i in range(n_dense):
if i > 0:
y = actreg(y)
y = y.feed(activation())
y = y.feed(Dense(y.output_shape[0], init=init_he_normal))
y = y.feed(Dense(y.output_shape[0], init=init_he_normal,
reg_w=reg, reg_b=reg))
y = actreg(y)
y = y.feed(activation())
y = y.feed(Dense(mnist_classes, init=init_glorot_uniform))
y = y.feed(Dense(mnist_classes, init=init_glorot_uniform,
reg_w=final_reg, reg_b=final_reg))
y = y.feed(Softmax())
model = Model(x, y, unsafe=True)
@ -101,10 +132,8 @@ if sgdr:
restarts=starts-1, restart_decay=restart_decay,
expando=lambda i:0)
else:
# learner = TriangularCLR(optim, epochs=epochs, lower_rate=0, upper_rate=lr,
# frequency=epochs//starts)
learner = SineCLR(optim, epochs=epochs, lower_rate=0, upper_rate=lr,
frequency=epochs//starts)
learner = CLR(optim, epochs=epochs, lower_rate=0, upper_rate=lr,
frequency=epochs//starts)
loss = CategoricalCrossentropy()
mloss = Accuracy()
@ -113,6 +142,11 @@ ritual = Ritual(learner=learner, loss=loss, mloss=mloss)
#ritual = NoisyRitual(learner=learner, loss=loss, mloss=mloss,
# input_noise=1e-1, output_noise=3.2e-2, gradient_noise=1e-1)
for node in model.ordered_nodes:
children = [str(n) for n in node.children]
if children:
sep = '->'
print(str(node) + sep + ('\n' + str(node) + sep).join(children))
log('parameters', model.param_count)
ritual.prepare(model)
@ -151,6 +185,12 @@ def measure_error(quiet=False):
measure_error()
while learner.next():
act_t = (learner.epoch - 1) / (learner.epochs - 1)
if actreg_lamb:
for node in model.ordered_nodes:
if isinstance(node, ActivityRegularizer):
node.reg.lamb = act_t * node.reg.lamb_orig # HACK
indices = np.arange(inputs.shape[0])
np.random.shuffle(indices)
shuffled_inputs = inputs[indices]
@ -174,6 +214,10 @@ while learner.next():
if not measure_every_epoch:
measure_error()
if save_fn is not None:
log('saving weights', save_fn)
model.save_weights(save_fn, overwrite=True)
if log_fn:
log('saving losses', log_fn)
np.savez_compressed(log_fn,