slavakung
commented
7 years ago Please see on stackoverflow:
in particular I received the response: "When you write gradsb = self.get_gradients(loss, [tf.Variable(a) for a in params]) you are defining a new tf.Variable for each a in params. Because the loss does not depend on these new variables, your gradients are None.
If you want to compute a second gradient you need to make sure that you're computing it with respect to Tensors that the objective does depend on."
and then I wrote:
"Apparently even replacing the current vector of parameters is not OK!! If I type this in the code:
grads = self.get_gradients(loss, params)
tempparam = [tf.Variable(a) for a in params]
params = [tf.add(a,a) for a in params]
gradsn = self.get_gradients(loss, params)
for a in gradsn:
print(a)
params = [tf.Variable(a) for a in tempparam]The result is still that None is printed!!
I know you understand what I am trying to do, at each iteration of get_updates, I would like to compute the gradients at a (slightly) different value of the parameter tensors, and use that to construct the update to the parameters for optimization and training. Is there any way to do this within the keras package?"
Perhaps it is entirely impossible to compute a gradient of the loss function with respect to its parameters at a different value than the current one, in some ways the get_updates is binded to only consider one (the current value) of the parameter. If this is not the case, please assist me in computing a second gradient tensor in my optimization code. If this is indeed the case, then I suggest this be investigated and corrected in a future release, as there are many optimization methods that do not involve just one computation of a gradient at each iteration and this effectively blocks all such methods from being tested on training the models, a serious impediment to research and practice involving keras.
bstriner
stale[bot]
RyanCV
namp
Hello,
I am a researcher in optimization and I trying to write a custom optimizer. I have come across a problem.
Take any optimizer code, say just copy SGD. In the beginning of get_updates, you see
now add the following line right after this one:
this should compute the gradients at a new tensor, with all the values the same as before
now try to see what you get:
you get a list of Nones (but if you print the list grads you see that they are still Tensors)
Why? And how to circumvent this problem? This is important as I'd like to compute the gradients at another point for my algorithm.
On a perhaps related level, when is batch_size used? get_gradients just goes to the gradients backend. So in some sense does the keras wrapper modify the loss function to just include the batch at the current iteration? Where is this done? I would be interested in adaptively modifying the batch_size in the optimizer, eventually.