Bugfix: ManagedOptimisation creates op at every iteration for ScipyOptimizer #8
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The
gpflow.train.ScipyOptimizer
doesn't come with an operation that increments theglobal_step
by default. The solution in here is to create the operation ourselves and then run it in the callback that is passed to it.Unfortunately the current code both creates and runs the operation inside the callback.
Minimum failing example:
Output without a fix: the optimisation is slow and gets slower at every iteration.
(lines manually aligned)
(etc.)
Output with a fix: the optimiser is fast.
A possible good practice for the future would be to call
tf.graph.finalize()
before running the optimization. Unfortunately, some new operations are created at the beginning of optimisation in the normal course of GPflow usage. For example,gpflow.core.Node.initialize
callsmisc.initialize_variables
which creates variable initializer operations, causing an error in a finalised graph.