imagining syntax

← schedopt_v4

free-281a55d8

done · peak unseen_mismatch 87.6

α schedule

α 3.587 ×46 α 2.394 ×24 α 2.975 ×75 α 4.536 ×12 α 2.385 ×12 α 0.8641 ×131 α 0.8641 ×100

direction
mixed
α range
0.8641–4.536
stages
7

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.8641

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 57.2 ± 6.552.6 ± 4.553.2 ± 4.850.4 ± 3.1
110 61.9 ± 9.255.4 ± 6.156.3 ± 8.451.3 ± 4.1
120 66.7 ± 11.159.7 ± 8.260.6 ± 11.454.3 ± 6.4
130 73.5 ± 12.366.0 ± 10.065.9 ± 13.558.6 ± 8.6
140 79.5 ± 12.272.2 ± 10.771.1 ± 13.963.8 ± 10.0
150 82.9 ± 12.477.3 ± 10.674.5 ± 14.168.1 ± 10.8
160 84.7 ± 10.880.3 ± 9.675.3 ± 12.870.2 ± 9.9
170 87.4 ± 9.383.8 ± 8.778.0 ± 12.172.9 ± 10.1
180 91.1 ± 9.387.5 ± 9.584.2 ± 12.279.4 ± 11.3
190 94.8 ± 9.792.0 ± 9.791.8 ± 11.587.6 ± 11.5
200 94.4 ± 11.690.1 ± 11.692.2 ± 12.786.2 ± 12.3
210 94.3 ± 12.387.0 ± 14.693.4 ± 12.484.9 ± 14.4
220 94.4 ± 12.176.6 ± 17.792.3 ± 14.476.1 ± 16.8
230 94.1 ± 11.167.3 ± 25.392.7 ± 13.665.3 ± 23.7
240 96.5 ± 7.064.0 ± 32.395.1 ± 9.762.6 ± 31.3
250 95.7 ± 7.452.3 ± 28.695.2 ± 8.249.8 ± 27.1
260 98.6 ± 2.455.4 ± 29.498.4 ± 2.853.0 ± 27.0
270 100.0 ± 0.063.2 ± 24.9100.0 ± 0.059.9 ± 25.3
280 100.0 ± 0.063.3 ± 30.7100.0 ± 0.060.8 ± 28.3
290 100.0 ± 0.061.5 ± 33.7100.0 ± 0.058.8 ± 32.4
300 100.0 ± 0.064.8 ± 28.4100.0 ± 0.062.5 ± 28.1
310 100.0 ± 0.055.0 ± 30.7100.0 ± 0.053.4 ± 30.1
320 100.0 ± 0.053.1 ± 8.9100.0 ± 0.052.0 ± 9.5
330 100.0 ± 0.048.0 ± 7.5100.0 ± 0.047.8 ± 8.1
340 100.0 ± 0.042.6 ± 29.9100.0 ± 0.045.0 ± 29.2
350 100.0 ± 0.050.2 ± 9.9100.0 ± 0.049.6 ± 11.6
360 100.0 ± 0.061.2 ± 13.7100.0 ± 0.061.4 ± 14.4
370 100.0 ± 0.072.0 ± 22.9100.0 ± 0.071.3 ± 22.6
380 100.0 ± 0.049.3 ± 0.0100.0 ± 0.048.8 ± 0.0
390 100.0 ± 0.049.3 ± 0.0100.0 ± 0.049.0 ± 0.0
400 100.0 ± 0.049.3 ± 0.0100.0 ± 0.049.0 ± 0.0

Downloads