imagining syntax

← schedopt_v4

free-61f53dcb

done · peak unseen_mismatch 97.0

α schedule

α 4.898 ×13 α 4.498 ×25 α 4.634 ×18 α 2.667 ×22 α 2.878 ×55 α 2.445 ×26 α 1.997 ×27 α 0.2762 ×95 α 3.231 ×19 α 3.231 ×100

direction
mixed
α range
0.2762–4.898
stages
10

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.231

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 59.2 ± 9.155.6 ± 6.554.4 ± 7.551.5 ± 4.1
110 63.4 ± 10.359.1 ± 8.956.9 ± 8.853.4 ± 5.9
120 68.8 ± 12.464.0 ± 10.960.5 ± 10.956.1 ± 7.8
130 73.9 ± 13.068.9 ± 11.864.2 ± 12.759.4 ± 9.3
140 77.8 ± 12.773.4 ± 11.967.3 ± 13.263.0 ± 10.3
150 81.6 ± 11.477.2 ± 11.770.2 ± 12.965.8 ± 10.5
160 84.5 ± 9.581.0 ± 10.073.1 ± 11.568.6 ± 9.9
170 87.9 ± 8.784.5 ± 9.077.0 ± 10.872.2 ± 9.7
180 90.4 ± 6.887.6 ± 7.480.0 ± 9.675.4 ± 8.9
190 92.6 ± 6.189.5 ± 7.683.6 ± 8.678.7 ± 8.6
200 97.0 ± 5.194.1 ± 8.294.3 ± 7.289.7 ± 10.0
210 98.3 ± 3.288.0 ± 17.797.4 ± 4.685.8 ± 17.8
220 99.8 ± 0.567.0 ± 22.999.4 ± 0.966.0 ± 22.7
230 99.9 ± 0.257.5 ± 21.799.7 ± 0.956.9 ± 21.3
240 100.0 ± 0.055.5 ± 28.9100.0 ± 0.054.9 ± 28.4
250 100.0 ± 0.057.1 ± 27.4100.0 ± 0.056.2 ± 26.7
260 100.0 ± 0.047.6 ± 20.5100.0 ± 0.047.6 ± 20.4
270 100.0 ± 0.050.0 ± 20.0100.0 ± 0.049.4 ± 19.9
280 100.0 ± 0.051.9 ± 16.6100.0 ± 0.052.5 ± 14.8
290 100.0 ± 0.063.1 ± 27.2100.0 ± 0.063.8 ± 27.2
300 100.0 ± 0.074.1 ± 20.4100.0 ± 0.074.2 ± 19.8
310 100.0 ± 0.074.7 ± 20.9100.0 ± 0.075.7 ± 19.8
320 100.0 ± 0.078.0 ± 17.1100.0 ± 0.078.2 ± 17.6
330 100.0 ± 0.073.9 ± 0.0100.0 ± 0.074.2 ± 0.0
340 100.0 ± 0.075.5 ± 0.0100.0 ± 0.077.6 ± 0.0
350 100.0 ± 0.095.5 ± 0.0100.0 ± 0.097.0 ± 0.0

Downloads