Optimization Target must be a link

Question

I have an autoencoder model of 4 linear layers written using chainer.Chain. Running the optimizer.setup line in Trainer section give me the following error:

TypeError                                 Traceback (most recent call 
last)
<ipython-input-9-a2aabc58d467> in <module>()
      8 
      9 optimizer = optimizers.AdaDelta()
---> 10 optimizer.setup(sda)
     11 
     12 train_iter = iterators.SerialIterator(train_data,batchsize)

/usr/local/lib/python3.6/dist-packages/chainer/optimizer.py in setup(self, 
link)
    415         """
    416         if not isinstance(link, link_module.Link):
--> 417             raise TypeError('optimization target must be a link')
    418         self.target = link
    419         self.t = 0

TypeError: optimization target must be a link

The link to class StackedAutoEncoder is as follows: StackAutoEncoder link

The link to class NNBase which is used to write class AutoEncoder is as follows: NNBase link

model = chainer.Chain(
    enc1=L.Linear(1764, 200),
    enc2=L.Linear(200, 30),
    dec2=L.Linear(30, 200),
    dec1=L.Linear(200, 1764)
)


sda = StackedAutoEncoder(model, gpu=0)
sda.set_order(('enc1', 'enc2'), ('dec2', 'dec1'))
sda.set_optimizer(Opt.AdaDelta)
sda.set_encode(encode)
sda.set_decode(decode)

from chainer import iterators, training, optimizers
from chainer import Link, Chain, ChainList

optimizer = optimizers.AdaDelta()
optimizer.setup(sda)

train_iter = iterators.SerialIterator(train_data,batchsize)
valid_iter = iterators.SerialIterator(test_data,batchsize)

updater = training.StandardUpdater(train_iter,optimizer)
trainer = training.Trainer(updater,(epoch,"epoch"),out="result")

from chainer.training import extensions
trainer.extend(extensions.Evaluator(valid_iter, sda, device=gpu))

Chain is made of Links. I want to understand why the optimizer is not recognizing the sda which is StackedAutoencoder(model)?

I have edited the question and added the links to StackedAutoEncoder and NNBase class...Also any additional pointers for the program is appreciated. — TulakHord, Apr 30 '19 at 12:09

score 1 · Accepted Answer · answered Apr 30 '19 at 23:29

StackedAutoencoder inherits NNBase class, which inherits object class, so they are not chainer.Chain class.

You can refer official example for how to define your own network. For example, MNIST example defines MLP as follows:

class MLP(chainer.Chain):

    def __init__(self, n_units, n_out):
        super(MLP, self).__init__()
        with self.init_scope():
            # the size of the inputs to each layer will be inferred
            self.l1 = L.Linear(None, n_units)  # n_in -> n_units
            self.l2 = L.Linear(None, n_units)  # n_units -> n_units
            self.l3 = L.Linear(None, n_out)  # n_units -> n_out

    def forward(self, x):
        h1 = F.relu(self.l1(x))
        h2 = F.relu(self.l2(h1))
        return self.l3(h2)

So is there any way to make the optimizer recognize the optimization target as link even though StackedAutoEncoder class is not made up of links? Do I have to write StackedAutoEncoder class in ChainList to make it work or is there a workaround? — TulakHord, May 02 '19 at 05:11
You can pass only Link to optimizer's target. For example `optimizer.setup(sda.model)` if sda itself is not link but it holds `model` Link. — corochann, May 03 '19 at 14:41

Optimization Target must be a link

1 Answers1