Reinforcement learning with a neuronal network as Q-function in a maze (toy example) https://windfis.ch/git/maze-rlnn/atom?h=master 2016-02-06T16:42:53+00:00 2016-02-06T16:42:53+00:00 Florian Jung flo@windfisch.org 2016-02-06T16:42:53+00:00 urn:sha1:03de67dc103c2557265d42f6e5b5d5b94375dd61 2016-02-05T00:17:03+00:00 Florian Jung flo@windfisch.org 2016-02-05T00:17:03+00:00 urn:sha1:330f513d059bdff3ff9ff96895a434332eda17f3 2016-02-03T17:43:29+00:00 Florian Jung flo@windfisch.org 2016-02-03T17:43:29+00:00 urn:sha1:f352a2cdc621b78b2771748c5bc60b212480d08b 2016-02-03T17:09:19+00:00 Florian Jung flo@windfisch.org 2016-02-03T17:09:19+00:00 urn:sha1:f0ebbd21a6ab27e4b97d03a19e06b01eb9a2114a 2016-01-08T20:01:22+00:00 Florian Jung flo@windfisch.org 2016-01-08T20:01:22+00:00 urn:sha1:d8b98f347d0858128074054dea583772692365d0 2016-01-08T20:00:41+00:00 Florian Jung flo@windfisch.org 2016-01-08T20:00:41+00:00 urn:sha1:92f25c46b2d3e7e22b62faa4ca450b29757e977a 2016-01-08T17:09:36+00:00 Florian Jung flo@windfisch.org 2016-01-08T17:09:36+00:00 urn:sha1:c77a636c82fcf5211beadab90371719f175ef954 2016-01-08T17:07:19+00:00 Florian Jung flo@windfisch.org 2016-01-08T17:07:19+00:00 urn:sha1:2483f393d9a3740b35606ca6acb6cb2df8ffdcd2 - make input values fit into 0..1 range (not 0..5/0..7) - reward of 0.5 instead of 10 - randomize initial weights much more to avoid degeneration 2016-01-08T17:07:01+00:00 Florian Jung flo@windfisch.org 2016-01-08T17:07:01+00:00 urn:sha1:1d8d0cc92e55b8bd51afbbdab14dade5c646e39d 2016-01-08T17:06:40+00:00 Florian Jung flo@windfisch.org 2016-01-08T17:06:40+00:00 urn:sha1:63a2121abe15a5d7e08d7e7a8f92fdbfbe3b5987