Implement (Smoothed) Finite Difference Approximation of Influence Function #482
Conversation
This file contains 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
* initial test against analytic fisher vp (pair coded w/ sam) * linting * added check against analytic ate * added vmap and grad smoke tests * added missing init * linting and consolidated fisher tests to one file * fixed types * fixing linting errors * trying to fix type error for python 3.8 * fixing test errors * added patch to test to prevent from failing when denom is small * composition issue * removed missing import * fixed failing test with seeding * addressing Eli's comments
* upper bound on cg_iters * address comment
* initial test against analytic fisher vp (pair coded w/ sam) * linting * added check against analytic ate * added vmap and grad smoke tests * added missing init * linting and consolidated fisher tests to one file * fixed types * fixing linting errors * trying to fix type error for python 3.8 * fixing test errors * added patch to test to prevent from failing when denom is small * composition issue * seeded NMC implementation * linting * removed missing import * changed to eli's seedmessenger suggestion * added failing edge case * explicitly add max plate argument * added warning message * fixed linting error and test failure case from too many cg iters * eli's contextlib seeding strategy * removed seedmessenger from test * randomness should be shared across calls * switched back to different
…er_vp` (#430) * hessian vector product formulation for fisher * ignoring small type error * fixed linting error
* initial test against analytic fisher vp (pair coded w/ sam) * linting * added check against analytic ate * added vmap and grad smoke tests * added missing init * linting and consolidated fisher tests to one file * fixed types * fixing linting errors * trying to fix type error for python 3.8 * fixing test errors * added patch to test to prevent from failing when denom is small * composition issue * seeded NMC implementation * linting * removed missing import * changed to eli's seedmessenger suggestion * added failing edge case * explicitly add max plate argument * added warning message * fixed linting error and test failure case from too many cg iters * eli's contextlib seeding strategy * removed seedmessenger from test * randomness should be shared across calls * uncomitted change before branch switch * switched back to different * added revised simple model and guide * added multiple link functions in test * linting
* one step correction * increased tolerance * fixing lint issue
…PredictiveLikelihood` (#473) * sketch batched nmc lpd * nits * fix type * format * comment * comment * comment * typo * typo * add condition to help guarantee idempotence * simplify edge case * simplify plate_name * simplify batchedobservation logic * factorize * simplify batched * reorder * comment * remove plate_names * types * formatting and type * move unbind to utils * remove max_plate_nesting arg from get_traces * comment * nit * move get_importance_traces to utils * fix types * generic obs type * lint * format * handle observe in batchedobservations * event dim * move batching handlers to utils * replace 2/3 vmaps, tests pass * remove dead code * format * name args * lint * shuffle code * try an extra optimization in batchedlatents * add another optimization * undo changes to test * remove inplace adds * add performance test showing speedup * document internal helpers * batch latents test * move batch handlers to predictive * add bind_leftmost_dim, document PredictiveFunctional and PredictiveModel * use bind_leftmost_dim in log prob
* documentation * documentation clean up w/ eli * fix lint issue
…that properly composes with other Predictive models, some vectorization pending.
azane
changed the title
|
Needs merge conflict resolution following #398 |
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.
*closed in favor of #501
WIP pending full vectorization and integration test of approximator.
TODO description
blocked by: #478
addresses: #469