Closed mathematicalmichael closed 5 years ago
Merging #340 into master will increase coverage by
1.94%
. The diff coverage is94.12%
.
@@ Coverage Diff @@
## master #340 +/- ##
=========================================
+ Coverage 77.46% 79.4% +1.94%
=========================================
Files 22 23 +1
Lines 3953 4418 +465
=========================================
+ Hits 3062 3508 +446
- Misses 891 910 +19
Impacted Files | Coverage Δ | |
---|---|---|
bet/postProcess/__init__.py | 100% <100%> (ø) |
:arrow_up: |
bet/sample.py | 80.58% <68.18%> (+0.39%) |
:arrow_up: |
bet/postProcess/compareP.py | 95.37% <95.37%> (ø) |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update 3e78e4a...8ec9820. Read the comment docs.
@smattis all the documentation has been added, but please let me know if I need to elaborate on any of them.
Have to re-build documentation, but either way the gh-pages
branch is behind, so... doesn't seem like a priority.
@eecsu all checks pass
This PR implements a new sub-module that allows for the comparison of probability measures defined on different sample sets.
A sample set object is used as a reference "mesh" for evaluating various measures of similarity (metrics, distances, KL, etc), and a query is performed against two other sample sets so that the densities are evaluated at the same values in space.
Methods exist for setting the two measures (denoted
left
andright
for the two arguments into a metric/distance), evaluating the densities on the reference mesh (including the "expensive" step of building pointers with nearest-neighbor), and some commonly-used functions are implemented as string-arguments.All sorts of care is taken for user-induced error, avoiding mistaken computations due to differing domains, changing probabilities by re-solving a problem, etc. Since everything is pass-by-reference, in theory a solution can be updated (either by changing the observed or changing how a predicted distribution is computed), and the distance would reflect the change accordingly once the probabilities are updated.