Michael Pilosov
cd9da60893
10 epochs
2026-05-25 22:51:43 +00:00
Michael Pilosov
e9d3d72499
Improve training signal and add honest eval metrics
...
- prepare_training_data: bag_size 5→8, street signs fill slots first so
every sample contains the most geographically discriminative texts
- train: HuberLoss replaces MSE (robust to outlier intersections),
ReduceLROnPlateau scheduler added, split logic extracted to data_utils
- eval: reproduce train/val split to report honest per-bag and
per-intersection-aggregated metrics separately for train and val sets
- data_utils: shared split_indices() so train and eval use identical splits
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-05-25 22:41:14 +00:00
Michael Pilosov
b72cd1b917
device fix (cuda)
2026-05-25 22:10:06 +00:00
Michael Pilosov
40967a303c
drop pena
2026-05-25 22:09:57 +00:00
Michael Pilosov
8af279771a
frozen layers more epochs
2026-05-25 21:45:10 +00:00
Michael Pilosov
8f4d4c1057
optimizations?
2026-05-25 21:45:08 +00:00
Michael Pilosov
44c7753856
cuda
2026-05-25 21:25:24 +00:00
Michael Pilosov
b5810dd282
more training data + frozen layers options
2026-05-25 15:16:19 -06:00
Michael Pilosov
bfad4547ce
no mps
2026-05-25 14:26:37 -06:00
Michael Pilosov
e8419354f4
first train/eval
2026-05-25 14:11:05 -06:00
6c5d71e2d9
fix bug in city lookups
2023-05-05 21:48:16 +00:00
ab26735c82
fix combinatorial problem, cleanup
2023-05-05 16:02:34 +00:00
9c439bb6c8
readme update
2023-05-05 07:13:41 +00:00
5c33b5135f
rename data file
2023-05-05 07:12:45 +00:00
294d4bb1cd
shuffle option
2023-05-05 06:42:32 +00:00
b8ac59d942
progress bar
2023-05-05 06:37:35 +00:00
4500c1b483
chunking
2023-05-05 06:17:16 +00:00
c111678bb8
install efficiency
2023-05-05 05:50:51 +00:00
4c36e07085
bugfix
2023-05-05 05:45:01 +00:00
4e2e160072
linting
2023-05-05 05:43:15 +00:00
9083e9d6e1
chunk data gen
2023-05-05 05:41:52 +00:00
1e38ce04c9
linting
2023-05-05 03:41:59 +00:00
96f0c3a313
add argparse
2023-05-05 03:41:51 +00:00
57ef4c06df
major bugfix
2023-05-05 03:34:56 +00:00
8bd3e6f55e
update plots to reflect epochs used
2023-05-05 02:08:20 +00:00
03313d3904
bigger batch size
2023-05-05 01:45:50 +00:00
948c337ec2
batchsize
2023-05-05 00:54:43 +00:00
f75c99cb4e
less wordy
2023-05-05 00:45:22 +00:00
6b0f413eab
but why section
2023-05-05 00:44:30 +00:00
336531421d
prettier readme + full eval + images
2023-05-04 23:08:01 +00:00
313c850631
lock / reorder
2023-05-04 22:28:21 +00:00
c0fcf3eb32
requirements file
2023-05-04 22:25:46 +00:00
5f3174e6a3
print
2023-05-04 22:00:59 +00:00
502bbdba5e
better targets formatting
2023-05-04 19:10:33 +00:00
f193018ac2
full training process on US cities
2023-05-04 19:05:46 +00:00
282c0466d8
add check for git status
2023-05-04 19:05:33 +00:00
e9adbed41a
details in readme
2023-05-04 19:05:12 +00:00
fab8952d59
full training
2023-05-04 10:17:41 +00:00
c7b294c557
readme
2023-05-04 10:13:12 +00:00
b14a33c984
initial commit, working code
2023-05-04 10:03:15 +00:00
