imagining syntax

← schedopt_v4

free-5411cbfe

done · peak unseen_mismatch 99.0

α schedule

α 4.761 ×160 α 0.8109 ×140 α 0.8109 ×100

direction
descending
α range
0.8109–4.761
stages
3

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.8109

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

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 61.9 ± 9.157.9 ± 7.254.8 ± 7.051.7 ± 4.4
110 65.8 ± 9.961.6 ± 9.057.2 ± 8.153.6 ± 5.9
120 70.6 ± 9.466.6 ± 9.059.7 ± 8.256.3 ± 6.4
130 74.9 ± 8.371.3 ± 7.962.2 ± 9.058.6 ± 7.1
140 77.8 ± 6.974.4 ± 6.663.9 ± 8.660.6 ± 7.2
150 80.0 ± 5.377.0 ± 5.365.1 ± 8.362.2 ± 7.0
160 80.6 ± 4.477.9 ± 4.665.3 ± 7.362.6 ± 6.2
170 84.4 ± 5.281.7 ± 5.171.0 ± 8.367.6 ± 7.5
180 91.9 ± 5.389.3 ± 5.684.6 ± 9.080.5 ± 9.0
190 97.3 ± 2.895.1 ± 3.994.4 ± 5.691.1 ± 6.5
200 98.5 ± 2.495.7 ± 3.997.1 ± 3.992.6 ± 5.1
210 99.6 ± 0.694.6 ± 4.598.9 ± 1.292.8 ± 3.9
220 100.0 ± 0.077.8 ± 15.2100.0 ± 0.176.1 ± 12.0
230 100.0 ± 0.084.3 ± 12.299.5 ± 0.584.8 ± 10.8
240 100.0 ± 0.051.9 ± 0.0100.0 ± 0.050.9 ± 0.0
250 100.0 ± 0.050.7 ± 0.0100.0 ± 0.050.2 ± 0.0
260 100.0 ± 0.049.4 ± 0.0100.0 ± 0.049.0 ± 0.0
270 100.0 ± 0.054.9 ± 0.0100.0 ± 0.053.7 ± 0.0
280 100.0 ± 0.070.7 ± 0.0100.0 ± 0.071.8 ± 0.0
290 100.0 ± 0.060.6 ± 0.0100.0 ± 0.059.2 ± 0.0
300 100.0 ± 0.070.8 ± 0.0100.0 ± 0.070.4 ± 0.0
310 100.0 ± 0.082.9 ± 0.0100.0 ± 0.081.8 ± 0.0
320 100.0 ± 0.062.1 ± 0.0100.0 ± 0.061.3 ± 0.0
330 100.0 ± 0.057.0 ± 0.0100.0 ± 0.055.3 ± 0.0
340 100.0 ± 0.091.0 ± 0.0100.0 ± 0.092.0 ± 0.0
350 100.0 ± 0.099.7 ± 0.0100.0 ± 0.099.0 ± 0.0

Downloads