imagining syntax

← schedopt_v4

free-9723128c

done · peak unseen_mismatch 89.1

α schedule

α 3.286 ×17 α 4.929 ×59 α 0.05794 ×40 α 2.26 ×61 α 4.754 ×11 α 3.407 ×13 α 3.382 ×14 α 1.659 ×14 α 1.292 ×17 α 2.521 ×54 α 2.521 ×100

direction
mixed
α range
0.05794–4.929
stages
11

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.521

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 55.5 ± 5.050.5 ± 2.953.1 ± 4.349.6 ± 2.1
110 57.5 ± 9.249.6 ± 3.755.8 ± 8.848.9 ± 3.3
120 62.5 ± 14.149.5 ± 5.560.8 ± 13.848.6 ± 5.1
130 69.1 ± 17.751.4 ± 7.667.4 ± 18.049.6 ± 7.3
140 75.5 ± 18.655.6 ± 11.173.6 ± 19.753.8 ± 10.7
150 81.4 ± 18.560.0 ± 13.579.9 ± 19.757.9 ± 13.0
160 86.3 ± 17.163.9 ± 15.584.7 ± 18.362.0 ± 15.2
170 89.5 ± 16.169.1 ± 15.888.1 ± 16.966.6 ± 15.6
180 92.2 ± 13.575.8 ± 14.590.7 ± 14.773.2 ± 14.4
190 93.5 ± 10.982.3 ± 12.991.7 ± 12.478.8 ± 13.0
200 94.6 ± 9.484.9 ± 12.292.8 ± 10.581.6 ± 11.9
210 95.0 ± 7.788.0 ± 10.292.5 ± 9.484.1 ± 10.6
220 94.8 ± 8.590.7 ± 7.291.9 ± 9.686.0 ± 8.1
230 95.7 ± 9.492.9 ± 8.693.6 ± 10.289.1 ± 9.4
240 95.1 ± 12.290.5 ± 12.193.2 ± 13.387.1 ± 13.6
250 87.4 ± 19.274.6 ± 18.485.8 ± 19.072.0 ± 16.6
260 88.0 ± 17.785.2 ± 15.684.2 ± 18.479.2 ± 16.2
270 87.8 ± 11.885.3 ± 12.782.7 ± 12.878.9 ± 12.7
280 91.3 ± 8.089.1 ± 8.684.9 ± 11.180.7 ± 12.0
290 88.7 ± 6.286.3 ± 7.181.2 ± 6.876.3 ± 6.7
300 92.2 ± 4.787.9 ± 5.182.6 ± 4.677.7 ± 3.3
310 89.9 ± 0.486.0 ± 2.080.1 ± 3.574.7 ± 1.6
320 92.1 ± 1.889.0 ± 3.082.8 ± 0.977.2 ± 0.8
330 93.0 ± 1.688.8 ± 3.682.8 ± 1.677.3 ± 0.2
340 93.8 ± 1.290.5 ± 1.686.7 ± 1.382.5 ± 0.7
350 95.6 ± 1.491.8 ± 0.286.8 ± 7.379.5 ± 4.5
360 95.9 ± 2.191.2 ± 0.788.3 ± 7.880.4 ± 3.5
370 95.7 ± 1.790.7 ± 0.188.0 ± 8.079.8 ± 3.4
380 96.3 ± 1.293.5 ± 0.790.5 ± 5.284.7 ± 4.7
390 95.2 ± 0.592.3 ± 0.787.5 ± 4.883.1 ± 3.4
400 95.9 ± 0.994.3 ± 1.890.8 ± 5.086.7 ± 4.7

Downloads