Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

A2C and Synchronous extension #17

Open
wants to merge 66 commits into
base: master
Choose a base branch
from
Open

A2C and Synchronous extension #17

wants to merge 66 commits into from

Conversation

juice1000
Copy link

  • Create vanilla A2C and A2C-Sync
  • Create separate NN and partly shared NN implementation
  • Plot extension
  • Args for demo mode, training, testing, ...
  • Measure probability distribution of actions, make confidence intervall
  • Include Vizdoom environment

juice1000 and others added 30 commits March 23, 2020 21:35
@juice1000
CARTPOLE - create vanilla a2c and synchronous a2c
0ce32a4
@juice1000
CARTPOLE - test
4918b37
@juice1000
CARTPOLE - renew model and parameters, add args and plotter function,…
63b4092 
… terminate after score of 600, save and load model
@juice1000
CARTPOLE - for all: terminate when score higher than 500 within range
a459f81
@juice1000
CARTPOLE - refactor a2c to without torch.mp module
2045869
@juice1000
CARTPOLE - stabilize algorithms: true synchronous learning for a2c_sy…
490767a 
…nc, set upper limit for ep_length, gym make unwrapped for a2c
@juice1000
CARTPOLE - training duration in plotter, args and plotter in utils, s…
b7a4bf6 
…ave model if consecutive reward over 300
@juice1000
CARTPOLE - if load model, then check for 100 iterations of ep_r over 500
f9ace27
@juice1000
CARTPOLE - if load_model, then check for last 100 iterations of ep_r …
5d78d64 
…over 500
@juice1000
Merge branch 'master' of https://github.com/juice1000/pytorch-A3C
ae02294
@juice1000
CARTPOLE - initialize only one process when load_model
93e6726
@juice1000
CARTPOLE - utils gets optimizer, a2c_sync NN update after done
2317cea
@juice1000
CARTPOLE - add episode_time plotter and confidence intervall calculat…
add50a1 
…or for actions in utils, add information to plotter and merge, add functions in A2Cs and A3C, raise lr=0.005
@juice1000
CARTPOLE - create A2Cs and A3C shared NN
b9ec460
@juice1000
CARTPOLE - fix confidence interval error in Sync_A2C and A3C, action_…
e35afd8 
…queue in record function
@juice1000
CARTPOLE - refactor shared network algorithms, delete one confidence …
6d32122 
…intervall calculation
@juice1000
CARTPOLE - refactor shared NNs and lr, take out time sleep in test ph…
211ce33 
…ase for Sync-A2C
@juice1000
CARTPOLE - results and saved models
aca0f93
@juice1000
CARTPOLE - plots of results
810b960
@juice1000
VIZDOOM - create A2C-vizdoom for deadly corridor, create viz_utils
d943e7e
@juice1000
CARTPOLE - create directory
671f057
@juice1000
VIZDOOM - create directory
019f612
@juice1000
cleanup
7104cf7
@juice1000
VIZDOOM - create A3C model
b6e57e6
@juice1000
VIZDOOM - create A3C part2
d3a4033
@juice1000
VIZDOOM - create synchronous a2c
06a759f
@juice1000
VIZDOOM - add plotter and game replay in a2c
608e20a
@juice1000
VIZDOOM - plotter and stopwatch for a2c_sync and a3c
52ac70c
@juice1000
CARTPOLE - refactor synchronous coordinator in A2C_Sync
e139bbe
@juice1000
VIZDOOM - reinitialize state within each episode, adjust plotter boun…
a5179b2 
…daries
juice1000 and others added 30 commits April 14, 2020 21:06
@juice1000
VIZDOOM - correct A2C and A3C, adjust plotter boundaries
53beb17
@juice1000
CARTPOLE - create save_model.py, enable save data mode for all
5dc7e25
@juice1000
CARTPOLE - correct coding mistakes with save_data
8065952
@juice1000
CARTPOLE - separate training and test data
aed3bb4
@juice1000
CARTPOLE - remove time stamp in plots
05b2b14
@juice1000
CARTPOLE - add new plots, add plot data, add newly trained models
6fa38cb
@juice1000
VIZDOOM - save_data for A3C, create algo_compare for doom
534a930
@juice1000
VIZDOOM - new synchronous implementation, reinitialize threads and sa…
f98f2e1 
…ve stats everytime, create shared NN algorithms
@juice1000
VIZDOOM - refactor viz_utils
439e964
@juice1000
VIZDOOM - refactor plotter
8f2e140
@juice1000
VIZDOOM - redefine save model condition
d71c28a
@juice1000
VIZDOOM - refactor action_queue, take action_queue out from viz_utils
89b3e6d
@juice1000
VIZDOOM - set ylim to 1000 in viz_utils
bd7b79e
@juice1000
VIZDOOM - combine CSVs and plot for A2C-sync
19b5fb4
@juice1000
VIZDOOM - algo_compare with subplots of training and testing data
fbe2680
@juice1000
CARTPOLE - make subplots for training and testing data
3c8501a
@juice1000
CART/VIZ - probabilities plot
77975c3
@juice1000
CART/VIZ - code corrections
220c2ed
@juice1000
VIZDOOM - fix probabilities plot and ep-rew-time plot, fix action_que…
5bc01fc 
…ue in A3C
@juice1000
VIZDOOM - code grooming and refactor algo_compare
983bc7c
@juice1000
Update README.md
5a1fb36
VIZDOOM - push plots and plot data
0adf081
Merge branch 'master' of https://github.com/juice1000/Synchronous-vs-…
2ffe7d0 
…Asynchronous-Learning-Pytorch-
@juice1000
Update README.md
3a2b84c
@juice1000
Update README.md
c106ea0
CARTPOLE - move results to higher directory
14e4c57
Merge branch 'master' of https://github.com/juice1000/Synchronous-vs-…
8aef27b 
…Asynchronous-Learning-Pytorch-
@juice1000
Update README.md
8f16737
@juice1000
Update README.md
72d755d
@juice1000
Update README.md
e606ecf
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@juice1000