Ecosyste.ms: Issues

An open API service for providing issue and pull request metadata for open source projects.

GitHub / danijar/mindpark issues and pull requests

#30 - Update tf and gym API calls

Pull Request - State: closed - Opened by fapaul almost 8 years ago

#29 - Seed policy and env with a hashed tuple of user seed and episode number

Issue - State: open - Opened by danijar over 8 years ago
Labels: enhancement

#28 - Run learners in processes instead of threads

Issue - State: open - Opened by danijar over 8 years ago
Labels: enhancement

#27 - Make Metrics thread-safe for use in A3C

Issue - State: open - Opened by danijar over 8 years ago - 1 comment
Labels: bug

#26 - Register functions and algorithms from outside the package for use in definitions

Issue - State: open - Opened by danijar over 8 years ago
Labels: enhancement

#25 - Package project and set up platform testing with tox

Issue - State: open - Opened by danijar over 8 years ago
Labels: feature

#24 - Experiment crashes when plotting stats from the same database

Issue - State: open - Opened by danijar over 8 years ago
Labels: bug

#23 - Improve plotting performance

Issue - State: open - Opened by danijar over 8 years ago
Labels: enhancement

#22 - Add Schedule classes

Issue - State: open - Opened by danijar over 8 years ago
Labels: enhancement

#21 - Infer experiment name from definition file name instead

Issue - State: closed - Opened by danijar over 8 years ago

#20 - Use sum to aggregate costs rather than mean

Issue - State: closed - Opened by danijar over 8 years ago
Labels: bug

#19 - Consider letting policies return probability distributions

Issue - State: closed - Opened by danijar over 8 years ago
Labels: enhancement

#18 - Show previous observation to policy to simulate env and policy in parallel

Issue - State: open - Opened by danijar over 8 years ago
Labels: feature

#17 - Switch to YAML 1.2 to support scientific notation floats natively

Issue - State: closed - Opened by danijar over 8 years ago
Labels: enhancement

#16 - Make sure env.close() is always called

Issue - State: closed - Opened by danijar over 8 years ago
Labels: bug

#15 - Scatter plot scores from all training episodes per epoch

Issue - State: open - Opened by danijar over 8 years ago
Labels: feature

#14 - Make preprocessing part of the agent rather than the environment

Issue - State: closed - Opened by danijar over 8 years ago
Labels: cleanup

#13 - Add logging mechanism to agents

Issue - State: open - Opened by danijar over 8 years ago
Labels: feature

#12 - Fix benchmarking Atari domains in parallel

Issue - State: open - Opened by danijar over 8 years ago
Labels: bug

#11 - Use tuple states for LSTM cells

Issue - State: open - Opened by danijar over 8 years ago
Labels: cleanup

#10 - Count timesteps of the underlying environment rather than the preprocessed one

Issue - State: closed - Opened by danijar over 8 years ago - 1 comment
Labels: bug, enhancement

#9 - Make control utilities thread safe

Issue - State: open - Opened by danijar over 8 years ago
Labels: bug

#8 - Only require matplotlib for vizbot.plot but not for vizbot.train

Issue - State: open - Opened by danijar over 8 years ago
Labels: enhancement

#7 - Train agents in parallel and interleaved

Issue - State: closed - Opened by danijar over 8 years ago
Labels: feature

#6 - Plot average reward per episode instead of cumulative reward

Issue - State: closed - Opened by danijar over 8 years ago
Labels: enhancement

#5 - Simplify DQN agent code

Issue - State: closed - Opened by danijar over 8 years ago
Labels: cleanup

#4 - Implement A3C

Issue - State: closed - Opened by danijar over 8 years ago
Labels: algorithm

#3 - Profile and improve performance

Issue - State: closed - Opened by danijar over 8 years ago - 3 comments
Labels: enhancement

#2 - Add target network for DQN

Issue - State: closed - Opened by danijar over 8 years ago
Labels: algorithm

#1 - Train and compare the same agent with different settings

Issue - State: closed - Opened by danijar over 8 years ago
Labels: feature