imagining syntax

← schedopt_v4

free-91df844a

done · peak unseen_mismatch 93.7

α schedule

α 2.535 ×152 α 1.42 ×148 α 1.42 ×100

direction
descending
α range
1.42–2.535
stages
3

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.42

seen_* probed at reference α = 1.4 · unseen_* = held-out pairings, uniform (α-independent) · 1000 pairs per condition

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 54.3 ± 5.550.8 ± 2.852.0 ± 4.749.7 ± 2.4
110 56.5 ± 7.751.7 ± 4.553.8 ± 8.049.9 ± 3.7
120 60.0 ± 10.454.0 ± 6.056.3 ± 10.251.2 ± 4.9
130 64.7 ± 13.658.2 ± 9.260.7 ± 13.354.6 ± 7.9
140 70.3 ± 15.163.7 ± 11.464.7 ± 15.358.3 ± 10.7
150 74.8 ± 16.668.8 ± 14.068.9 ± 17.062.6 ± 13.1
160 79.1 ± 15.672.6 ± 14.073.2 ± 17.066.8 ± 13.7
170 84.1 ± 16.177.3 ± 14.579.9 ± 17.472.2 ± 14.7
180 87.3 ± 15.280.5 ± 14.584.1 ± 16.876.1 ± 14.9
190 88.8 ± 14.981.1 ± 14.785.8 ± 17.076.9 ± 15.5
200 90.3 ± 12.580.8 ± 12.687.1 ± 15.376.5 ± 13.5
210 93.0 ± 10.081.8 ± 11.790.5 ± 12.377.6 ± 12.9
220 96.2 ± 6.484.7 ± 12.794.1 ± 8.880.9 ± 12.4
230 98.4 ± 3.584.6 ± 13.396.9 ± 5.081.5 ± 12.9
240 99.1 ± 2.084.6 ± 15.398.3 ± 3.382.5 ± 14.5
250 99.8 ± 0.484.5 ± 16.599.5 ± 0.981.9 ± 15.5
260 99.8 ± 0.583.8 ± 15.899.6 ± 1.080.8 ± 16.3
270 100.0 ± 0.069.4 ± 13.0100.0 ± 0.067.1 ± 10.8
280 100.0 ± 0.075.0 ± 19.3100.0 ± 0.072.5 ± 17.6
290 100.0 ± 0.068.7 ± 24.4100.0 ± 0.066.8 ± 23.9
300 100.0 ± 0.072.4 ± 21.3100.0 ± 0.072.8 ± 20.2
310 100.0 ± 0.054.1 ± 0.0100.0 ± 0.055.7 ± 0.0
320 100.0 ± 0.069.6 ± 0.0100.0 ± 0.069.9 ± 0.0
330 100.0 ± 0.053.1 ± 0.0100.0 ± 0.054.9 ± 0.0
340 100.0 ± 0.071.3 ± 0.0100.0 ± 0.071.9 ± 0.0
350 100.0 ± 0.074.5 ± 0.0100.0 ± 0.073.5 ± 0.0
360 100.0 ± 0.093.8 ± 0.0100.0 ± 0.093.7 ± 0.0

Downloads