Implement mass matrix adaptation #2327
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
twiecki
reviewed
Jun 17, 2017
pymc3/sampling.py
Outdated
|
Should we actually use this for tuning (unrelated to init) in NUTS? |
aseyboldt
force-pushed
the
nuts-adapt
branch
5 times, most recently
from
dffba11 to
fec319e
Compare
|
I think this is ready to be merged now. |
junpenglao
reviewed
Jul 16, 2017
pymc3/sampling.py
Outdated
| return {k: np.asarray(v) for k, v in ppc.items()} | ||
|
|
||
|
|
||
| def init_nuts(init='ADVI', njobs=1, n_init=500000, model=None, |
There was a problem hiding this comment.
Shouldnt init='auto' as default?
There was a problem hiding this comment.
yes, fixed now.
It was the default in pm.sample.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
This implements adaptation for diagonal mass matrices during tuning.
samplegets a new option forinit, namelyadvi+adapt_diag, which first runs advi to get an initial estimate for the mass matrix and then further adapts it using the covariance of the tuning samples.The approach for computing adapting the mass matrix differs a bit from that in stan. Stan keeps the mass matrix constant within a window and then uses the variances in the latest window as mass matrix for the next window.
In this PR I use an online algorithm for computing the variance and change it continuously. It starts with the estimate from advi with a weight of 30 and updates the variance for each new sample. After 200 samples it starts computing new variances from ground up, but still uses the first variance for sampling. after 400 samples it switches to the second set of variances. This should prevent bad initialization from advi to ruin the whole trace. (see
quadpotential.QuadpotentialDiagAdapt.adapt)Still needs a lot of unit tests/doc/tinkering and we should compare it to the stan approach carefully.
It would probably also be an option to change weights over time and therefore achieve a similar effect as using windows.