Shared tests for various distributions #35
Conversation
| @@ -47,8 +47,8 @@ | |||
| ([] (uniform-distribution 0.0 1.0)) | |||
| ([lo hi] (->Uniform (rng) lo hi))) | |||
|
|
|||
| (defn gaussian-distribution | |||
There was a problem hiding this comment.
I changed this to match the other conventions / Gen.jl's convention.
Codecov Report
@@ Coverage Diff @@
## main #35 +/- ##
==========================================
+ Coverage 62.50% 66.51% +4.01%
==========================================
Files 14 16 +2
Lines 616 666 +50
Branches 21 21
==========================================
+ Hits 385 443 +58
+ Misses 210 202 -8
Partials 21 21
|
| @@ -85,7 +85,7 @@ | |||
| {:pre [(pos? alpha) (pos? beta)]} | |||
| (if (< 0 v 1) | |||
| (- (+ (* (- alpha 1) (Math/log v)) | |||
| (* (- beta alpha) (Math/log (- 1 v)))) | |||
|
|
||
| The implementation follows the algorithm described on the Cauchy | ||
| distribution's [Wikipedia | ||
| page](https://en.wikipedia.org/wiki/Cauchy_distribution#Probability_density_function_(PDF))." | ||
| [scale location v] |
There was a problem hiding this comment.
Changing these to match Gen.jl's arg order.
sritchie
force-pushed
the
sritchie/dist_tests
branch
from
fb8dee0
to
57aa1ca
Compare
|
@sritchie Tagging @Schaechtle in on this one! |
There was a problem hiding this comment.
this looks mostly good: I left minor inline comments. However, this line indicates an error due to name clashing of gen.trace and gen/trace. If this is fixed by a different PR, feel free to update/ignore
| (checking "Delta properties" | ||
| [center (gen-double -100 100) | ||
| v (gen-double -100 100)] | ||
| (if (= center v) |
There was a problem hiding this comment.
I am confused: Unless I am misreading what gen/double* does, this will never hold. Or is there some shared random seeding that I am missing?
There was a problem hiding this comment.
This is actually calling
(defn gen-double
"Returns a generator that produces numerical doubles between `min` and
`max` (inclusive)."
[min max]
(gen/double*
{:min min :max max :infinite? false :NaN? false}))and with shrinking, center and v will both match probably at least once on each test run (as the generator tries the min, max, 0, etc)
But let me know if I am missing something!
test/gen/distribution_test.cljc
Outdated
| (is (= -1.8862943611198906 (dist/logpdf (->laplace 0 2) 1))) | ||
| (is (= 4.214608098422191 (dist/logpdf (->laplace 0 0.001) 0.002))))) | ||
|
|
||
| (defn gaussian-tests [->gaussian] |
There was a problem hiding this comment.
I guess this should be renamed to normal, too.
|
@Schaechtle yep, that was a problem for cljs, fixed in #41 . Thank you! |
sritchie
force-pushed
the
sritchie/dist_tests
branch
2 times, most recently
from
d078f60
to
fe597aa
Compare
sritchie
force-pushed
the
sritchie/dist_tests
branch
from
fe597aa
to
70df6e0
Compare
@zane , let me know what you think of this approach. This gives at least a baseline of tests for each of the logpdf implementations (and writing these exposed a bug I had in my beta log-likelihood!)
I'll add docs to each of the distribution namespaces before we merge.