Commit Graph

77 Commits

Author SHA1 Message Date
f7d6cda8b6 Report AUCs in Delphi2M format 2026-07-25 11:15:06 +08:00
8d0d71292e Refactor TrajMixer to single residual 2026-07-24 15:25:02 +08:00
7b48cb8425 Fix RBF time-bias initialization 2026-07-24 14:39:15 +08:00
20c99484f3 Add two-stage TrajMixer mixing 2026-07-24 10:51:41 +08:00
85352dae0f Revert "Implement shared event-trajectory reasoning backbone"
This reverts commit 06f29c0f0a.
2026-07-23 16:05:00 +08:00
22faee7c51 Revert "Add nano and tiny model presets"
This reverts commit 6f7b5be405.
2026-07-23 16:03:29 +08:00
6f7b5be405 Add nano and tiny model presets 2026-07-23 15:01:33 +08:00
06f29c0f0a Implement shared event-trajectory reasoning backbone 2026-07-23 14:23:34 +08:00
68a6a3df88 Scale TrajMixer hidden width with head count 2026-07-22 16:14:52 +08:00
978c88a4ed Log and save model parameter counts 2026-07-22 14:44:31 +08:00
db0947ce9d Implement TrajMixer block 2026-07-22 11:52:44 +08:00
f6bde7e167 Add t-query logits and hidden export script 2026-07-08 17:29:25 +08:00
0ff0b6d861 Add script to compute landmark future death and incident disease risks 2026-07-01 17:45:17 +08:00
bf31aa0432 Avoid multiprocessing data loader for shape export 2026-07-01 15:37:36 +08:00
d08e5b34f4 Add Weibull shape export scripts 2026-07-01 15:31:31 +08:00
f417a91a74 Parallelize extra-info summary reduction 2026-07-01 13:21:08 +08:00
e3c9ecd19f Batch extra-info disease attribution stats 2026-07-01 12:47:30 +08:00
d3e68ac09b Avoid overflow in attribution variance 2026-07-01 10:58:09 +08:00
16df74c04d Skip attribution for next-token runs 2026-07-01 10:47:09 +08:00
d9dc115e09 Report disease parameter attribution 2026-07-01 10:16:51 +08:00
3cf756ccb0 Add landmark attribution dependencies 2026-07-01 09:55:38 +08:00
93450ab06b Update missing evaluation runner 2026-07-01 09:19:42 +08:00
84ae07bf48 Add missing training runs script 2026-06-29 15:40:55 +08:00
ca294193f4 Clarify empty extra-info logging 2026-06-29 11:26:48 +08:00
d4a29063d6 Add no extra-info type list 2026-06-29 11:14:50 +08:00
00ccedf82f Report death distribution parameters for disease attribution 2026-06-29 10:42:32 +08:00
fe826c4052 Defer attribution output until after GPU pass 2026-06-27 16:11:49 +08:00
980dd7c282 Align disease attribution batching with survival evaluator 2026-06-27 16:01:35 +08:00
517d573d3b Reduce attribution GPU synchronization 2026-06-27 15:56:12 +08:00
20686c128f Vectorize per-disease attribution batches 2026-06-27 15:51:12 +08:00
8928688aca Optimize per-disease attribution scanning 2026-06-27 15:43:27 +08:00
8ef88ff325 Pad attribution batches before concatenation 2026-06-27 15:38:32 +08:00
a7e969ae51 Summarize disease mortality attribution by age and sex 2026-06-27 15:33:50 +08:00
34353d33d5 Scan all diseases for mortality attribution 2026-06-27 15:30:38 +08:00
128a8ce655 Add single-disease mortality attribution script 2026-06-27 15:24:15 +08:00
01629cec56 Fix mortality attribution active rows 2026-06-27 14:29:32 +08:00
2ae92da0ec Stream mortality attribution batches 2026-06-27 14:08:02 +08:00
12265bd248 Batch mortality attribution queries 2026-06-27 13:57:16 +08:00
0b7c866292 Add mortality attribution evaluation 2026-06-27 13:34:30 +08:00
00de2ff4a9 Fix event-free survival inference tensor use 2026-06-27 12:08:19 +08:00
307704f74d Simplify event-free survival data loading 2026-06-27 12:02:48 +08:00
3cd1109249 Add event-free survival evaluation 2026-06-27 11:48:47 +08:00
df6b45b95a Clarify all-future model interpretation 2026-06-27 10:53:55 +08:00
6df0435f65 Refactor DeepHealth indices around disease expression 2026-06-26 16:25:36 +08:00
9ec3921bed Add burden trajectory plotting script 2026-06-26 12:49:21 +08:00
fb873326ee Keep burden landmark query times consistent 2026-06-26 12:24:51 +08:00
7ee29329f5 Write burden CSV shards in workers 2026-06-26 12:06:57 +08:00
263267f583 Use DataLoader for burden readouts 2026-06-26 11:50:12 +08:00
2b0b20d231 Use fork for Linux burden workers 2026-06-26 11:36:52 +08:00
c5ecbb79f2 Keep burden reduction on GPU 2026-06-26 11:27:44 +08:00