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

move losses into Model and refactor methods

Browse source
This commit is contained in:
Connor Olding 2017-09-16 16:59:36 +00:00
parent 910facf98d
commit e22316a4c9
3 changed files with 60 additions and 40 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

34
onn.py
View file

@ -696,8 +696,8 @@ class StochMRitual(Ritual):
# this probably doesn't make sense for regression problems,
# let alone small models, but here it is anyway!
def __init__(self, learner=None, loss=None, mloss=None, gamma=0.5):
super().__init__(learner, loss, mloss)
def __init__(self, learner=None, gamma=0.5):
super().__init__(learner)
self.gamma = _f(gamma)
def prepare(self, model):
@ -726,12 +726,12 @@ class StochMRitual(Ritual):
# np.clip(layer.W, -1, 1, out=layer.W)
class NoisyRitual(Ritual):
def __init__(self, learner=None, loss=None, mloss=None,
def __init__(self, learner=None,
input_noise=0, output_noise=0, gradient_noise=0):
self.input_noise = _f(input_noise)
self.output_noise = _f(output_noise)
self.gradient_noise = _f(gradient_noise)
super().__init__(learner, loss, mloss)
super().__init__(learner)
def learn(self, inputs, outputs):
# this is pretty crude
@ -1077,13 +1077,13 @@ def lookup_loss(maybe_name):
return SomethingElse()
raise Exception('unknown objective', maybe_name)
def ritual_from_config(config, learner, loss, mloss):
def ritual_from_config(config, learner):
if config.ritual == 'default':
ritual = Ritual(learner=learner, loss=loss, mloss=mloss)
ritual = Ritual(learner=learner)
elif config.ritual == 'stochm':
ritual = StochMRitual(learner=learner, loss=loss, mloss=mloss)
ritual = StochMRitual(learner=learner)
elif config.ritual == 'noisy':
ritual = NoisyRitual(learner=learner, loss=loss, mloss=mloss,
ritual = NoisyRitual(learner=learner,
input_noise=1e-1, output_noise=1e-2,
gradient_noise=2e-7)
else:
@ -1105,7 +1105,10 @@ def model_from_config(config, input_features, output_features, callbacks=None):
if y.output_shape[0] != output_features:
y = y.feed(Dense(output_features, init))
model = Model(x, y, unsafe=config.unsafe)
loss = lookup_loss(config.loss)
mloss = lookup_loss(config.mloss) if config.mloss else loss
model = Model(x, y, loss=loss, mloss=mloss, unsafe=config.unsafe)
if config.fn_load is not None:
log('loading weights', config.fn_load)
@ -1122,10 +1125,7 @@ def model_from_config(config, input_features, output_features, callbacks=None):
learner = learner_from_config(config, optim, rscb)
loss = lookup_loss(config.loss)
mloss = lookup_loss(config.mloss) if config.mloss else loss
ritual = ritual_from_config(config, learner, loss, mloss)
ritual = ritual_from_config(config, learner)
return model, learner, ritual
@ -1234,8 +1234,8 @@ def run(program, args=None):
def measure_error():
def print_error(name, inputs, outputs, comparison=None):
predicted = model.forward(inputs)
err = ritual.mloss.forward(predicted, outputs)
predicted = model.evaluate(inputs)
err = model.mloss.forward(predicted, outputs)
if config.log10_loss:
print(name, "{:12.6e}".format(err))
if comparison:
@ -1272,7 +1272,7 @@ def run(program, args=None):
# use plain SGD in warmup to prevent (or possibly cause?) numeric issues
temp_optim = learner.optim
temp_loss = ritual.loss
temp_loss = model.loss
learner.optim = Optimizer(lr=0.001)
ritual.loss = Absolute() # less likely to blow up; more general
@ -1292,7 +1292,7 @@ def run(program, args=None):
ritual.reset()
learner.optim = temp_optim
ritual.loss = temp_loss
model.loss = temp_loss
if training:
measure_error()

51
onn_core.py
View file

@ -884,13 +884,17 @@ class Dense(Layer):
# Models {{{1
class Model:
def __init__(self, nodes_in, nodes_out, unsafe=False):
def __init__(self, nodes_in, nodes_out, loss=None, mloss=None, unsafe=False):
self.loss = loss if loss is not None else SquaredHalved()
self.mloss = mloss if mloss is not None else loss
nodes_in = [nodes_in] if isinstance(nodes_in, Layer) else nodes_in
nodes_out = [nodes_out] if isinstance(nodes_out, Layer) else nodes_out
assert type(nodes_in) == list, type(nodes_in)
assert type(nodes_out) == list, type(nodes_out)
self.nodes_in = nodes_in
self.nodes_out = nodes_out
self.nodes = traverse_all(self.nodes_in, self.nodes_out)
self.make_weights()
for node in self.nodes:
@ -928,22 +932,40 @@ class Model:
assert inner_offset >= node.size, "Layer {} allocated less weights than it said it would".format(node)
offset += node.size
def forward(self, X, deterministic=False):
def evaluate(self, inputs, deterministic=True):
values = dict()
input_node = self.nodes[0]
output_node = self.nodes[-1]
values[input_node] = input_node._propagate(np.expand_dims(X, 0), deterministic)
values[input_node] = input_node._propagate(np.expand_dims(inputs, 0), deterministic)
for node in self.nodes[1:]:
values[node] = node.propagate(values, deterministic)
return values[output_node]
def backward(self, error):
def apply(self, error): # TODO: better name?
values = dict()
input_node = self.nodes[0]
output_node = self.nodes[-1]
values[output_node] = output_node._backpropagate(np.expand_dims(error, 0))
for node in reversed(self.nodes[:-1]):
values[node] = node.backpropagate(values)
return self.dW
return values[input_node]
def forward(self, inputs, outputs, measure=False, deterministic=False):
predicted = self.evaluate(inputs, deterministic=deterministic)
if measure:
error = self.mloss.forward(predicted, outputs)
else:
error = self.loss.forward(predicted, outputs)
return error, predicted
def backward(self, predicted, outputs, measure=False):
if measure:
error = self.mloss.backward(predicted, outputs)
else:
error = self.loss.backward(predicted, outputs)
# input_delta is rarely useful; it's just to match the forward pass.
input_delta = self.apply(error)
return self.dW, input_delta
def clear_grad(self):
for node in self.nodes:
@ -1028,11 +1050,8 @@ class Model:
# Rituals {{{1
class Ritual: # i'm just making up names at this point.
def __init__(self, learner=None, loss=None, mloss=None):
# TODO: store loss and mloss in Model instead of here.
def __init__(self, learner=None):
self.learner = learner if learner is not None else Learner(Optimizer())
self.loss = loss if loss is not None else Squared()
self.mloss = mloss if mloss is not None else loss
self.model = None
def reset(self):
@ -1041,10 +1060,10 @@ class Ritual: # i'm just making up names at this point.
self.bn = 0
def learn(self, inputs, outputs):
predicted = self.model.forward(inputs)
self.model.backward(self.loss.backward(predicted, outputs))
error, predicted = self.model.forward(inputs, outputs)
self.model.backward(predicted, outputs)
self.model.regulate()
return predicted
return error, predicted
def update(self):
optim = self.learner.optim
@ -1062,14 +1081,14 @@ class Ritual: # i'm just making up names at this point.
self.learner.batch(b / batch_count)
if test_only:
predicted = self.model.forward(batch_inputs, deterministic=True)
predicted = self.model.evaluate(batch_inputs, deterministic=True)
else:
predicted = self.learn(batch_inputs, batch_outputs)
error, predicted = self.learn(batch_inputs, batch_outputs)
self.model.regulate_forward()
self.update()
if loss_logging:
batch_loss = self.loss.forward(predicted, batch_outputs)
batch_loss = self.model.loss.forward(predicted, batch_outputs)
if np.isnan(batch_loss):
raise Exception("nan")
self.losses.append(batch_loss)
@ -1077,7 +1096,7 @@ class Ritual: # i'm just making up names at this point.
if mloss_logging:
# NOTE: this can use the non-deterministic predictions. fixme?
batch_mloss = self.mloss.forward(predicted, batch_outputs)
batch_mloss = self.model.mloss.forward(predicted, batch_outputs)
if np.isnan(batch_mloss):
raise Exception("nan")
self.mlosses.append(batch_mloss)

15
onn_mnist.py
View file

@ -145,7 +145,13 @@ y = y.feed(Dense(mnist_classes, init=init_glorot_uniform,
reg_w=final_reg, reg_b=final_reg))
y = y.feed(output_activation())
model = Model(x, y, unsafe=True)
if output_activation in (Softmax, Sigmoid):
loss = CategoricalCrossentropy()
else:
loss = SquaredHalved()
mloss = Accuracy()
model = Model(x, y, loss=loss, mloss=mloss, unsafe=True)
def rscb(restart):
log("restarting", restart)
@ -176,12 +182,7 @@ else:
lament('WARNING: no learning rate schedule selected.')
learner = Learner(optim, epochs=epochs)
loss = CategoricalCrossentropy() if output_activation == Softmax else SquaredHalved()
mloss = Accuracy()
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)
ritual = Ritual(learner=learner)
model.print_graph()
log('parameters', model.param_count)