Hi,
I am trying to implement a mean covariance RBM class. I have two *shared
variables Din, genSamp* and a
*tensor variable x*I have a cost function as follows
cost, updates = mcrbm.get_cost_updates(lr = 0.01, persistent = None, k=1)
I have a theano function to iterate over 'x' as follows
train_rbm = theano.function(
[index],
cost,
updates=updates,
givens={
x: train_set_x[index * batch_size: (index + 1) * batch_size]
},
on_unused_input='warn',
name='train_rbm'
)
In the cost function, DIn undergoes some changes during negative sampling.
With this setup, I intend to update genSamp over each iteration of
train_rbm and use it for calculating the cost during RBM training
I tried the following lines of code inside the cost function to update
genSamp
a. self.genSamp.set_value(self.Din.get_value(borrow=True),
borrow=True)
b. self.genSamp = theano.shared(self.Din.get_value(borrow=True),
borrow= True)
c. updates[self.genSamp] =
theano.shared(self.Din.get_value(borrow=True), borrow= True)
However, none of these lines are affecting genSamp and it is not updating
after each iteration of the theano function (I am using a for loop, no
theano.scan)
Am I missing something? Have I gone wrong in my understanding of some of
theano's fundamental workings? If not, has this issue been raised before?
How can I update this genSamp shared variable?
--
---
You received this message because you are subscribed to the Google Groups
"theano-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
