batched SSM compatibility#685
Draft
Dekermanjian wants to merge 10 commits into
Draft
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #685 +/- ##
===========================================
+ Coverage 51.60% 91.24% +39.63%
===========================================
Files 73 79 +6
Lines 8003 8097 +94
===========================================
+ Hits 4130 7388 +3258
+ Misses 3873 709 -3164
🚀 New features to boost your workflow:
|
…, SARIMAX, ETS so that positional arguments don't clash with the new parameter, updated error in control flow for data registry
jessegrabowski
requested changes
May 22, 2026
jessegrabowski
left a comment
jessegrabowski
left a comment
Member
There was a problem hiding this comment.
This is a really great start, i'll fire up the notebook and play it. Here are some thoughts off the cuff.
@ricardoV94 your voice would be very welcome here :)
| data_dims = None | ||
|
|
||
| if self.batch_size: | ||
| data_dims = (BATCH_DIM, TIME_DIM, OBS_STATE_DIM) |
Member
There was a problem hiding this comment.
What if there are more than one batch dim?
| shock_trajectory = shock_trajectory.swapaxes(0, 1) | ||
|
|
||
| def bmv(A, x): | ||
| return pt.matmul(A, x[..., None])[..., 0] |
Member
There was a problem hiding this comment.
I thought we automatically do something like this already
|
|
||
| pm.Deterministic("irf", irf, dims=[TIME_DIM, ALL_STATE_DIM]) | ||
| if self.batch_size: | ||
| irf = irf.swapaxes(0, 1) |
Member
There was a problem hiding this comment.
This si the right pattern but you need to reason about the case where there are multiple batches
…tch_tools utilities file, use slicing logic instead of branching for forecast indexing, make batch_size a tuple for destructuring instead of branching
…_sizes, added a utility to infer the batch dimension
… vectorize certain matrices 2. Updated observation covariance computation in sample_conditional to support batched dimensions 3. added tests for batched dimension SSMs
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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 following is a WIP implementation of allowing the SSM module to work with multiple time-series via batched inputs.
This first draft allows batched sampling and supports the following internals:
sample_unconditional_priorsample_conditional_priorsample_unconditional_posteriorsample_conditional_posteriorAdditional testing is still required and redundancies in code need to be cleaned-up.
Update 1:
I updated the way batching is defined from implicitly defined to explicitly defined by a
batch_sizeparameter and I fixed an error that I introduced in the data registry control flow. AllStateSpacetests now are passing locally. This just means that adding the batching implementation did not break any existing logic.@jessegrabowski, Here are the next steps before this will be ready:
StateSpacemethods support until all methods are supported