Create sampler module and gaussian mixture sampling function #793
Conversation
…an alternative method if not available
timothy-glover
requested review from
sdhiscocks and
nperree-dstl
and removed request for
a team
stonesoup/functions/__init__.py
Outdated
|
|
||
| Returns | ||
| ------- | ||
| : :class:`np.ndarray` of shape (:attr:`size`, num_dims)""" |
There was a problem hiding this comment.
Should return a StateVectors type?
There was a problem hiding this comment.
I was torn with this when I wrote the function but I think you are right. Changed in latest commit.
stonesoup/sampler/base.py
Outdated
|
|
||
|
|
||
| class Sampler(Base): | ||
| """Sampler base class""" |
There was a problem hiding this comment.
Should give a description of what a general sampler does.
There was a problem hiding this comment.
Description added
| @@ -509,6 +509,47 @@ def rotz(theta): | |||
| [zero, zero, one]]) | |||
|
|
|||
|
|
|||
| def gm_sample(means, covars, size, weights=None): | |||
| """Sample from a mixture of multi-variate Gaussians | |||
|
|
|||
There was a problem hiding this comment.
You might wish to explain how this is done, as for non-statisticians it might not be obvious. (First do a multinomial sample from the weight distribution to get the number of times you should sample from each component, then sample from those components that many times.. I think.)
Modified Sampler base class description
Just typos
There was a problem hiding this comment.
Looks good to me. As discussed it probably makes sense to have your DetectionSampler and GaussianDetectionParticleSampler both inheriting from a "DetectionSampler" class. Happy for that to be amended in a future PR but it may be worth renaming your DetectionSampler here if you do plan on restructuring slightly.
…oduce base DetectionSampler class
Codecov ReportPatch coverage:
Additional details and impacted files@@ Coverage Diff @@
## main #793 +/- ##
==========================================
+ Coverage 94.90% 94.92% +0.02%
==========================================
Files 176 183 +7
Lines 9757 9919 +162
Branches 1938 1970 +32
==========================================
+ Hits 9260 9416 +156
- Misses 353 355 +2
- Partials 144 148 +4
Flags with carried forward coverage won't be shown. Click here to find out more.
☔ View full report in Codecov by Sentry. |
Introduced the sampler module for producing samples from distributions.
ParticleSampleris the main class contributed for this and it accepts a callable sampling function such asnumpy.random.multivariate_normalorscipy.stats.uniform.rvs, and a dictionary of parameters to be used as kwargs for the callable and returns aParticleStatecontaining the samples.GaussianDetectionParticleSampleris a child class ofParticleSamplerthat specifically samples fromDetectionor sets ofDetection.DetectionSampleris a generalDetectionsampler that is able to switch between a detection based sampler object such asGaussianDetectionParticleSamplerand a non-detection backup sampler such asParticleSamplerthat is used if no detections are available.No single callable has been found for sampling from Gaussian mixture distributions which lead to the introduction of
gm_samplemethod instonesoup.functions. This method can be used as a callable forParticleSamplerand is the sampler used byGaussianDetectionParticleSampler.