def train_model(self, trainset, testset, **kwargs):
chain, model = self.setup_chain_model(**kwargs)
nepochs = int(kwargs.get("nepochs", 2))
ent_T = kwargs.get("ent_T", None)
name = self.get_name(**kwargs)
reports = [
'epoch', 'validation/main/branch0accuracy',
'validation/main/branch1accuracy',
'validation/main/branch2accuracy',
'validation/main/branch3accuracy',
'validation/main/branch4accuracy',
'validation/main/branch5accuracy',
'validation/main/branch6accuracy',
'validation/main/branch7accuracy',
'validation/main/branch8accuracy', 'validation/main/accuracy',
'validation/main/communication0', 'validation/main/communication1',
'validation/main/memory', 'validation/main/ent_T',
'validation/main/branch0exit', 'validation/main/branch1exit',
'validation/main/branch2exit'
]
trainer = Trainer('{}/{}'.format(self.folder, name),
chain,
trainset,
testset,
batchsize=self.batchsize,
nepoch=nepochs,
resume=self.resume,
reports=reports)
trainer.run()
return trainer, model, chain
def train_model(self, trainset, testset, **kwargs):
chain, model = self.setup_chain_model(**kwargs)
nepochs = int(kwargs.get("nepochs", 2))
name = self.get_name(**kwargs)
reports = [
'validation/main/branch0accuracy',
'validation/main/branch1accuracy',
'validation/main/branch2accuracy',
'validation/main/branch3accuracy',
'validation/main/branch4accuracy',
'validation/main/branch5accuracy',
'validation/main/branch6accuracy',
'validation/main/branch7accuracy'
]
trainer = Trainer('{}/{}'.format(self.folder, name),
chain,
trainset,
testset,
batchsize=self.batchsize,
nepoch=nepochs,
resume=self.resume,
reports=reports)
trainer.run()
return trainer, model
def train_model(self, trainset, testset, **kwargs):
chain, model = self.setup_chain_model(**kwargs)
nepochs = int(kwargs.get("nepochs", 2))
name = self.get_name(**kwargs)
trainer = Trainer('{}/{}'.format(self.folder,name), chain, trainset,
testset, nepoch=nepochs, resume=True)
trainer.run()
return trainer, model, chain
def train_model(self, trainset, testset, **kwargs):
chain, model = self.setup_chain_model(**kwargs)
nepochs = int(kwargs.get("nepochs", 2))
ent_T = kwargs.get("ent_T", None)
name = self.get_name(**kwargs)
reports = [
'epoch', 'main/loss', 'main/accuracy', 'validation/main/accuracy',
'validation/main/memory'
]
trainer = Trainer('{}/{}'.format(self.folder, name),
chain,
trainset,
testset,
batchsize=self.batchsize,
nepoch=nepochs,
resume=True,
reports=reports)
trainer.run()
return trainer, model, chain
def train_model(self, trainset, testset, **kwargs):
chain, model = self.setup_chain_model(**kwargs)
nepochs = int(kwargs.get("nepochs", 2))
pretrain_nepochs = int(kwargs.get("pretrain_nepochs", 20))
name = self.get_name(**kwargs)
# Train stage 0
trainer = Trainer('{}/{}'.format(self.folder,"pretrained_" + name), chain, trainset,
testset, nepoch=pretrain_nepochs, resume=True)
model.set_current_stage(0)
trainer.run()
# Load from stage 0 iinstead since optimize might keep some stages
# Train stage 1
chain.cleargrads()
lr = kwargs.get("lr", 0.001)
chain.setup_optimizers('adam', lr)
trainer = Trainer('{}/{}'.format(self.folder,name), chain, trainset,
testset, nepoch=nepochs, resume=True)
model.set_current_stage(1)
trainer.run()
return trainer, model
chain = Chain(branchweight =branchweight )
chain.add_sequence(model)
chain.setup_optimizers('adam', lr)
#for i in xrange(100):
# y = chain(x)
# loss = F.mean_squared_error(x, y)
# chain.backprop(loss)
# print float(loss.data)
#
#chain.save("model")
trainset, testset = chainer.datasets.get_mnist(ndim=3)
trainer = Trainer('{}/{}'.format(folder,name), chain, trainset, testset, nepoch=nepochs, resume=True)
trainer.run()
#y = Variable(chain)
#y = chain(x)
#g = c.build_computational_graph(chain)
#with open('graph.dot', 'w') as o:
# o.write(g.dump())
#with open('graph.dot', 'w') as o:
# g = computational_graph.build_computational_graph((y, ))
# o.write(g.dump())