SwamyDev / udacity-deep-rl-navigation / 62

Builds Branch Commit Type Ran Committer Via Coverage
62 master Merge pull request #5 from SwamyDev/dependabot/pip/resources/unity-mlagent/tensorflow-1.15.4 Bump tensorflow from 1.15.2 to 1.15.4 in /resources/unity-mlagent push web-flow travis-ci-com pending completion
58 priority-sampling Extract the replay buffer logic This should make it easier to integrate the priority replay logic. push SwamyDev travis-ci-com pending completion
57 priority-sampling Implement SumTree data structure This data structure will be used by the priority sampling algorithm. It provides an efficient means of keeping track of experience weights, to be used in sampling. push SwamyDev travis-ci-com pending completion
56 master Merge pull request #4 from SwamyDev/p3 Integrate final multi-agent project push web-flow travis-ci-com pending completion
53 p3 Update report and readme draft Improved wording and corrected some spelling mistakes. Also corrected for the score issue, where I've used the mean of both agent rewards before. Now I use the correct max score. push SwamyDev travis-ci-com pending completion
54 p3 Update report and readme draft Improved wording and corrected some spelling mistakes. Also corrected for the score issue, where I've used the mean of both agent rewards before. Now I use the correct max score. Pull #4 web-flow travis-ci-com pending completion
51 p3 Add P3 report first draft The report is necessary to finish the final project. push SwamyDev travis-ci-com pending completion
50 p3 Rename MADDPG to NDDPG As the current implementation does not really follow the MADDPG algorithm as described in the paper (link at the bottom), but rather just trains two independent DDPG algorithms, I decided to rename it, to avoid confusion. ... push SwamyDev travis-ci-com pending completion
49 p3 Fix NDDPG tests and coverage report With the new agent setup the basic learning tests failed, because it now required different observation, action and reward shapes. Also the multiple inheritance approach didn't work that well, because of the or... push SwamyDev travis-ci-com pending completion
47 p3 Implement keyboard interrupt and snapshots I want to keep the state when of the agent when I interrupt training. Often because it already reached a very high value. Also I wanted a way of investigating agent behaviour during training to get a bet... push SwamyDev travis-ci-com pending completion
46 p3 Remove MultiAgentWrapper It is good practise to remove code that is not used anymore. This particular implementation has been removed, because I decided to treat multi agents as their own algorithms. Mostly because it is not so easy to keep agent... push SwamyDev travis-ci-com pending completion
45 p3 Improve exploration logic of MADDPG agent By allowing preheat steps which are completely random the agent can observe a more diverse set of states. Otherwise it wouldn't see that much because it tends to crash early on into a local minimum. With ... push SwamyDev travis-ci-com pending completion
43 p3 Implement multi agent wrapper This allows me to use the CLI interface to start and train a multi agent setup. By wrapping existing agents, I make minimal changes to the code and keep the architecture flexible. push SwamyDev travis-ci-com pending completion
42 p3 Add multi agent handling to unity env adapter To solve the third project the unity environment gym adapter needs to handle multi agent environments properly, hence the adapter has been extended. Also another test mode for the new environment has ... push SwamyDev travis-ci-com pending completion
41 master Add link to spinning up RL push web-flow travis-ci-com pending completion