Commit b1d3274
committed
data(aroma): broaden the twelve memorizing heads with diverse positives (#256)
Each of these heads was drawing its positives solely from the weak-labelled HSDB odor corpus,
with no entry in CURATED at all — which is why they were thin and narrow enough to memorize.
Adds well-established character-impact chemistry for each, chosen for structural spread rather
than volume, the same way #247 moved tingling off zero: pinenes/camphene/carene/bornyl esters
for pine, cineole/terpinen-4-ol/aromadendrene for eucalyptus, phthalides for celery,
indole/cresol/benzyl esters/cinnamyl alcohol for narcissus.
Honest result: five moved off zero at the 0.5 threshold — narcissus 0 to 4 novel discoveries,
frankincense and costus to 2, fennel and allspice to 1. Another five (pine, eucalyptus,
rosemary, freesia, elemi) turned out to be under-confident rather than memorizing once probed
at 0.35, so the extra positives sharpened heads that had already learned their class. Turmeric
and celery did not move at any threshold and are tracked separately; a single-scaffold class
that thin is a model problem, not a data problem.
AUROC fell for several heads (pine 0.965 to 0.905, frankincense 0.916 to 0.829). That is the
right direction — a broader class is genuinely harder to fit, and a score dropping while
discovery rises is what learning instead of memorizing looks like.
Corpus now trains 167 aroma heads, up from 164.
Signed-off-by: Austin L. <86896075+rvnminers-A-and-N@users.noreply.github.com>1 parent f7c4512 commit b1d3274
1 file changed
Lines changed: 29 additions & 0 deletions
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
82 | 82 | | |
83 | 83 | | |
84 | 84 | | |
| 85 | + | |
| 86 | + | |
| 87 | + | |
| 88 | + | |
| 89 | + | |
| 90 | + | |
| 91 | + | |
| 92 | + | |
| 93 | + | |
| 94 | + | |
| 95 | + | |
| 96 | + | |
| 97 | + | |
| 98 | + | |
| 99 | + | |
| 100 | + | |
| 101 | + | |
| 102 | + | |
| 103 | + | |
| 104 | + | |
| 105 | + | |
| 106 | + | |
| 107 | + | |
| 108 | + | |
| 109 | + | |
| 110 | + | |
| 111 | + | |
| 112 | + | |
| 113 | + | |
85 | 114 | | |
86 | 115 | | |
87 | 116 | | |
| |||
0 commit comments