imagining syntax

← schedopt_v4

free-da985679

done · peak unseen_mismatch 98.5

α schedule

α 2.876 ×56 α 2.241 ×10 α 1.27 ×39 α 4.653 ×15 α 2.896 ×36 α 1.647 ×73 α 1.137 ×61 α 2.524 ×10 α 2.524 ×100

direction
mixed
α range
1.137–4.653
stages
9

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.524

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 ± 6.850.3 ± 2.152.8 ± 6.149.7 ± 1.9
110 56.3 ± 9.350.2 ± 2.654.7 ± 9.249.2 ± 2.4
120 60.1 ± 12.451.5 ± 5.858.2 ± 12.450.3 ± 4.8
130 64.4 ± 15.253.1 ± 7.461.8 ± 15.751.2 ± 6.1
140 68.8 ± 17.656.7 ± 9.865.9 ± 18.054.3 ± 8.6
150 73.0 ± 17.860.8 ± 11.269.5 ± 18.558.1 ± 10.6
160 77.9 ± 17.365.9 ± 13.274.1 ± 18.661.9 ± 12.7
170 83.0 ± 16.370.9 ± 15.379.4 ± 18.466.6 ± 15.1
180 87.5 ± 14.374.2 ± 16.684.3 ± 16.870.3 ± 16.7
190 90.2 ± 13.275.7 ± 15.487.7 ± 15.672.0 ± 15.8
200 93.1 ± 10.576.2 ± 16.790.4 ± 13.672.5 ± 16.4
210 95.2 ± 8.777.5 ± 17.092.7 ± 11.674.6 ± 16.1
220 96.6 ± 7.377.3 ± 18.794.7 ± 8.975.0 ± 17.2
230 98.4 ± 3.378.5 ± 17.797.3 ± 4.276.4 ± 16.5
240 99.4 ± 1.377.4 ± 19.199.0 ± 1.975.7 ± 18.2
250 99.8 ± 0.865.9 ± 17.999.7 ± 0.865.0 ± 16.8
260 100.0 ± 0.269.6 ± 18.099.9 ± 0.368.7 ± 17.0
270 100.0 ± 0.065.4 ± 17.3100.0 ± 0.065.1 ± 15.8
280 100.0 ± 0.061.9 ± 17.0100.0 ± 0.062.0 ± 16.2
290 100.0 ± 0.067.0 ± 20.4100.0 ± 0.068.1 ± 19.0
300 100.0 ± 0.066.7 ± 21.8100.0 ± 0.068.4 ± 20.2
310 100.0 ± 0.060.8 ± 20.1100.0 ± 0.163.1 ± 18.6
320 100.0 ± 0.068.1 ± 16.0100.0 ± 0.168.5 ± 15.5
330 100.0 ± 0.080.8 ± 15.499.9 ± 0.280.6 ± 15.2
340 100.0 ± 0.082.5 ± 18.0100.0 ± 0.082.6 ± 15.9
350 100.0 ± 0.086.8 ± 16.8100.0 ± 0.087.0 ± 16.9
360 100.0 ± 0.048.8 ± 0.0100.0 ± 0.052.3 ± 0.0
370 100.0 ± 0.057.9 ± 0.0100.0 ± 0.060.8 ± 0.0
380 100.0 ± 0.075.5 ± 0.0100.0 ± 0.075.6 ± 0.0
390 100.0 ± 0.090.1 ± 0.0100.0 ± 0.089.6 ± 0.0
400 100.0 ± 0.099.4 ± 0.0100.0 ± 0.098.5 ± 0.0

Downloads