- 07 Dec, 2019 4 commits
-
-
Vivien Kraus authored
This lets me use commas as separators.
-
Vivien Kraus authored
-
Vivien Kraus authored
-
Vivien Kraus authored
I often get times where all the predictions are set to 0. This is because in the beginning where there is a lot of exploration, the regression is simply based on the bid and whether we are the taker. So, most parts of the network are "discarded", which is to say the weights optimize so that the hidden neuron is never activated. However, at the end of the exploration, we need to exploit these neurons.
-
- 02 Dec, 2019 2 commits
-
-
Vivien Kraus authored
I don't have that much time
-
Vivien Kraus authored
This kills exploration a little, but it lets us learn from 100% of the games. Otherwise, 4vs1 and 3vs1 games would be ignored most of the time.
-
- 29 Nov, 2019 15 commits
-
-
Vivien Kraus authored
-
Vivien Kraus authored
-
Vivien Kraus authored
This reverts commit d5318d1d.
-
Vivien Kraus authored
If someone bids a straight keep, there is a 1/2 chance that the next player would bid a double keep. This is not acceptable. This patch reverts this.
-
Vivien Kraus authored
-
Vivien Kraus authored
-
Vivien Kraus authored
If I were to use a random perceptron, random bid would be either pass or double keep (because the only feature that would vary would be the strength strategy, which is linear).
-
Vivien Kraus authored
-
Vivien Kraus authored
-
Vivien Kraus authored
If I keep the weak initialization, then we won't be able to use 2/3 of the hidden neurons, because the initial weights are cyclic.
-
Vivien Kraus authored
The initial weights are set up so as to predict something between 0 and 1. The score is likely to be between -500 and 500.
-
Vivien Kraus authored
-
Vivien Kraus authored
-
Vivien Kraus authored
-
Vivien Kraus authored
-