Particle filter proposal implementation #1080
Merged
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
A-acuto
requested review from
nperree-dstl and
spike-dstl
and removed request for
a team
sdhiscocks
approved these changes
Sep 30, 2024
stonesoup/proposal/simple.py
Outdated
Comment on lines
124
to
127
| post_log_weights = np.array([mvn.logpdf(sample, | ||
| np.array(update.mean).reshape(-1), | ||
| update.covar) | ||
| for sample, update in zip(samples, updates)]) |
There was a problem hiding this comment.
This would be better to handle custom types, as mvn converts to float arrays before taking difference.
Suggested change
| post_log_weights = np.array([mvn.logpdf(sample, | |
| np.array(update.mean).reshape(-1), | |
| update.covar) | |
| for sample, update in zip(samples, updates)]) | |
| post_log_weights = np.array([mvn.logpdf(sample - update.mean.reshape(-1), cov=update.covar) | |
| for sample, update in zip(samples, updates)]) |
Also, here you use update.mean but above update.state_vector. Should these not be both the same? (In theory they should be anyway I guess? but...)
There was a problem hiding this comment.
I see your point, mean and state_vector are equal in this case. I changed it
| def test_prior_proposal(): | ||
| # test that the proposal as prior and basic PF implementation | ||
| # yield same results, since they are driven by the transition model | ||
| np.random.seed(16549) |
There was a problem hiding this comment.
See #1064 which aims to remove any setting of global random seeds in tests.
May need to add random state option to Proposal to support this, like models have.
jswright-dstl
approved these changes
Oct 1, 2024
…the state_vectr instead of mean and handling of arrays
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 PR updates the particle filter implementation with proposals, in particular prior as proposal and (local) kalman filter.
This is developed by @sglvladi and myself.
Now the particle filter uses prior as proposal as default, otherwise a Kalman filter proposal can be used.
Since in the prediction step we need to perform an update, we have modified as well the distance hypothesiser to include the measurements (or detections) to perform correctly the predict-update step.
We have implemented the local Lalman Filter proposal which propagates each particle using the kalman filter predict and update and then samples once for each particle the new state (before doing the particle update).
There is another way, commonly known as global Kalman Filter, where the particle distribution is approximated by a gaussian, which is propagated using the Kalman Filter and then a new set of particles is drawn from the new state. However, we encountered a number of issues in the making it working accordingly, in particular with problems with weights normalisation during the data-association step. I will open an issue to try to get to the bottom of this.
Summary paper that highlights the two implementations
We have not modified the probability hypothesier (yet) because there is the need of a fix proposed in #802 (PDA and JPDA to work with particle state) and we don't want to duplicate the work.
A couple of stone soup examples are :
single target no data association case
data association case