Timeline for Can you analyse a neural network to determine good states?
Current License: CC BY-SA 4.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 22, 2018 at 8:44 | vote | accept | OrangeMan | ||
| Aug 21, 2018 at 15:52 | answer | added | Dennis Soemers♦ | timeline score: 4 | |
| Aug 21, 2018 at 10:10 | comment | added | OrangeMan | @DukeZhou There are about 500 cards in the library right now. | |
| Aug 21, 2018 at 10:09 | comment | added | OrangeMan | @DennisSoemers There is 1 output for every action that could ever be played (almost all of this is playing card X from your hand), the highest output that is a valid move is selected to play. I used PPO to train the network. | |
| Aug 20, 2018 at 18:33 | comment | added | DukeZhou | This may or may not be relevant, but I'm personally interested in what type of card game it is. Deck building implies some form of post-Garfield CCG. Algorithmic deck building is an exciting prospect! Can I ask, how many cards in the card "library" which a deck can be built from? | |
| Aug 20, 2018 at 18:17 | comment | added | Dennis Soemers♦ | What kind of neural network do you have / how did you train it? What outputs does it have (did you train a network to output values with an algorithm like DQN, or did you train it to output a policy using policy gradient methods, or something else?) | |
| Aug 20, 2018 at 14:09 | history | asked | OrangeMan | CC BY-SA 4.0 |