The time-tested way of getting better performance from deep learning — use a bigger model:
1. Double model size 2. Initialize from old parameters 3. 10 additional days of training Result: 80% win rate versus model that played on-stage at The International vs paiN Gaming.

Sep 6, 2018 · 6:16 PM UTC

16
19
2
149
Replying to @gdb
Any additional compute penalty for the larger model?
Replying to @gdb
Unless you over fit with your bigger model!
Replying to @gdb
From a students perspective: when getting more performance from a model increase like this one usually worries about overfitting. Is that a problem for you? How do you manage it?
1
3
Replying to @gdb
Love it! May I enquire the current MMR?
Replying to @gdb @HaydnBelfield
Just like the Hollywood technique for sequals.
Replying to @gdb
How did you go about expanding it?
Replying to @gdb
yes, boring
Replying to @gdb
Would be nice to see a showmatch again.
Replying to @gdb
So just download more RAM?
Replying to @gdb
Need real testing, The back part of the game is terrible