Quinn Dougherty
f2460a5e12
It compiles
...
Value: [1e-3 to 4e-1]
2022-05-25 08:22:57 -04:00
Quinn Dougherty
3a56d6fca4
intermediate commit
...
Value: [1e-10 to 1e-5]
2022-05-25 08:22:57 -04:00
Quinn Dougherty
bafcb4f7b8
Cleaned up Scoring file: no dispatch yet
...
Value: [1e-4 to 6e-2]
2022-05-25 08:22:57 -04:00
Quinn Dougherty
9e7319ed57
More substantial CR; more named args
...
Value: [1e-6 to 1e-2]
2022-05-16 18:06:14 -04:00
Quinn Dougherty
1d2bb556de
Minor CR comments
...
Value: [1e-6 to 1e-3]
2022-05-16 15:39:40 -04:00
Quinn Dougherty
bdbb86aa9e
logScore on records now interprets almost every which way we're
...
interested in
Value: [1e-3 to 9e-1]
2022-05-13 16:15:04 -04:00
Quinn Dougherty
b4a1137019
Implemented correct math underlying logScoreWithPointResolution
...
Value: [1e-2 to 7e-1]
Realized that I need to switch argument order, put `prior` last maybe.
2022-05-13 15:43:59 -04:00
Quinn Dougherty
2ab395b4e5
Some minor CR
...
Value: [1e-10 to 1e-4]
2022-05-12 16:03:29 -04:00
Quinn Dougherty
51310819a1
logScore now in interface.
...
Value: [1e-4 to 1e-1]
2022-05-12 14:16:52 -04:00
Quinn Dougherty
978e149913
Initialized logScore and logScoreAgainstImproperPrior
...
Value: [1e-5 to 6e-3]
2022-05-12 13:11:51 -04:00
Quinn Dougherty
f5e3701a79
Clean up for CR
...
Value: [1e-7 to 43-4]
2022-05-12 09:51:20 -04:00
599c14b32c
tweak: Pass off to Quinn
2022-05-11 12:39:03 -04:00
4df4597ed3
fix: "Enrich" (add more x points) when integrating
...
in order to get more numerical precision.
Note: not complete yet.
Value: [1e-3 to 3e-1]
2022-05-11 12:23:36 -04:00
Quinn Dougherty
0b8da034c6
klDivergence on mixed distributions works for one test case
...
Value: [1e-4 to 5e-2]
2022-05-10 19:59:50 -04:00
Quinn Dougherty
f7690c33e0
Some cleanup
...
Value: [1e-4 to 1e-2]
2022-05-10 11:56:13 -04:00
Quinn Dougherty
15f1ebb429
KLDivergence on discretes is passing
...
Value: [1e-3 to 2e-1]
2022-05-10 11:27:59 -04:00
Quinn Dougherty
ccd55ef8f1
good evening, not working yet, but out of time for the night
...
Value: [1e-6 to 1e-4]
2022-05-09 19:17:27 -04:00
Quinn Dougherty
b2d80eef86
Removed negative infinity error handling
...
Value: [1e-5 to 1e-3]
2022-05-09 18:28:35 -04:00
Quinn Dougherty
06352357a2
response to CR
...
Value: [1e-5 to 5e-3]
2022-05-09 11:14:33 -04:00
Quinn Dougherty
28cb6b9c88
Final cleanup, but typescript is still borked
...
Value: [1e-8 to 1e-6]
2022-05-06 15:09:00 -04:00
Quinn Dougherty
722bfc6366
All three tests pass
...
- `uniform` `toPointSet` method has been changed for numerical
stability.
Value: [1e-1 to 1.75e0]
2022-05-06 13:58:15 -04:00
d9a40c973a
feat: Get KL divergence working except in case of numerical errors ()
...
- Quinn was of great help here.
- I also left some dead code, which still has to be cleaned up
- There are still very annoying numerical errors, so I left one test
failing. These are due to how the interpolation is done
- Quinn to pick up from here
Value: [0.6 to 2]
2022-05-06 12:26:51 -04:00
5dd272fb0c
tweak: Add tests for combineAlongSupportOfSecondArgument
2022-05-06 11:45:11 -04:00
87d6f8d4f6
fix: change integrand check order in KL divergence code
2022-05-06 09:49:42 -04:00
Quinn Dougherty
dcf56d7bc6
combineAlongSupportOfSecondArgument implemented, tests still failing
...
Value: [1e-4 to 4e-2]
2022-05-05 20:02:12 -04:00
Quinn Dougherty
b49865d3aa
progress on klDivergence (still working)
...
Value: [1e-5 to 1e-2]
2022-05-05 15:37:28 -04:00
Quinn Dougherty
6211d3cfb0
Some sorta klDivergence conclusion
...
Value: [1e-6 to 1e-3]
2022-05-04 14:55:49 -04:00
Quinn Dougherty
898547f3a3
klDivergence is now LogarithmWithThreshold
2022-05-04 13:53:32 -04:00
Quinn Dougherty
c95c56cfb8
Added logScaleWithThreshold(eps) and completed renaming to
...
`klDivergence`
Value: [1e-5 to 1e-3]
2022-05-04 13:02:58 -04:00
Quinn Dougherty
683439c7e5
Changed Ok(0) to Error(NegativeInfinityError).
...
Value: [1e-7 to 1e-5]
2022-05-04 11:59:46 -04:00
Quinn Dougherty
32f141abdc
Renamed LogScoring to KLDivergence
...
Value: [1e-7 to 1e-4]
2022-05-04 11:54:23 -04:00
Quinn Dougherty
627616558a
Made multiplicative factor in Scoring file rather than with bind
...
calls
Value: [1e-5 to 7.5e-4]
2022-05-04 11:48:20 -04:00
Quinn Dougherty
d5c9705811
Corrected log score
...
Value: [5e-5 to 2e-2]
2022-05-03 14:00:34 -04:00
Quinn Dougherty
cec4bbd334
fixed build
...
Value: [1e-6 to 1e-4]
2022-05-03 10:20:01 -04:00
Quinn Dougherty
5d6de7b487
Response to CR
...
Value: [1e-4 to 4e-3]
2022-05-03 10:15:48 -04:00
Quinn Dougherty
ca95c7cef8
base 2 log => natural log (on the advice of Nuno)
...
Value: [1e-4 to 2e-1]
2022-05-02 17:46:43 -04:00
Quinn Dougherty
2d84805f88
Added infinity error; fixed infinities in logs
...
Value: [1e-3 to 3e-2]
2022-05-02 17:15:23 -04:00
Quinn Dougherty
db3acbf96c
Added extra multiplicative factor in logScore integrand
...
Value: [8e-2 to 7e-1]
migrated intregrand from `log(predicted / answer)` to `answer *
log(predicted / answer)`
2022-05-02 13:40:34 -04:00
Quinn Dougherty
d595285078
make lint happy
...
Value: [1e-9 to 1e-5]
2022-05-02 11:55:42 -04:00
Quinn Dougherty
e2762a0f62
Make typechecker happy
...
Value: [1e-7 to 1e-3]
2022-05-02 11:55:13 -04:00
Quinn Dougherty
58c885f963
Merge remote-tracking branch 'origin/develop' into log-score-attempt
2022-05-02 11:34:29 -04:00
Ozzie Gooen
282fa7726b
Added scalePower and scaleLog
2022-04-28 21:14:03 -04:00
Ozzie Gooen
ced3af5813
First attempt at LogScore
2022-04-28 20:24:13 -04:00
Ozzie Gooen
dde28e54f0
Restructuring of E Sorted
2022-04-28 10:17:29 -04:00
Sam Nolan
7302a3ec10
Give isNormalised lower bound
2022-04-26 13:28:08 -04:00
Sam Nolan
350e420884
Add isNormalized to Continuous
2022-04-26 13:25:45 -04:00
Sam Nolan
6bfecf2204
Rename flip argument to be more descriptive
2022-04-25 15:55:25 -04:00
Sam Nolan
62653d55b4
Fix pointwise not commuting for subtraction
2022-04-25 15:46:56 -04:00
Sam Nolan
cf44e6d3c7
Merge branch 'develop' into commutative-error
2022-04-25 15:08:57 -04:00
Sam Nolan
f9c02ba626
Refactor and rename checkOrdered
2022-04-25 14:50:45 -04:00
