mila-iqia/atari-representation-learning

Issues

• Which version of gym[atari] was used?

  #74 opened a year ago by Sopralapanca
  0

• Conflict to the requirement

  #72 opened a year ago by hyyh28
  0

• How to get ball direction/velocity in Atari Pong?

  #71 opened 3 years ago by pedrohpf
  0

• Can't reproduce experimental results

  #69 opened 4 years ago by Yingdong-Hu
  5

• robot x/y coordinate mismatch in Berzerk game dict

  #68 opened 4 years ago by zacharyhorvitz
  0

• How necessary is local-local infomax?

  #66 opened 5 years ago by seann999
  1

• Generating Enough Episodes for Tennis

  #65 opened 5 years ago by adamtupper
  2

• Ram annotation of (x, y) are not aligned

  #64 opened 5 years ago by happywu
  3

• unused naff_fc_size parameter

  #63 opened 5 years ago by AnxietyYoungPoet
  2

• Incorrect/ambiguous features in Seaquest

  #59 opened 5 years ago by damnOblivious
  3

• Nearest neighbor visualizations

  #1 opened 5 years ago by ankeshanand
  0

• Add actions to dataloader

  #7 opened 5 years ago by eracah
  0

• Inverse Model

  #8 opened 5 years ago by eracah
  0

• Intuition on cross entropy.

  #58 opened 5 years ago by biggzlar
  4

• Benchmark the pretrained-rl-agent method

  #55 opened 5 years ago by liuyuezhang
  3

• Pillow version problem

  #52 opened 5 years ago by DuaneNielsen
  9

• Confusing comment about scaling frame observations

  #54 opened 5 years ago by Xemnas0
  1

• Should the weights and biases account be configurable?

  #53 opened 5 years ago by DuaneNielsen
  2

• Evaluate "transporter" network

  #51 opened 5 years ago by DuaneNielsen
  2

• Pixel positions

  #47 opened 5 years ago by frederikschubert
  2

• Overwriting of src

  #44 opened 5 years ago by eracah
  1

• Any tips on extracting RAM locations?

  #40 opened 5 years ago by neighthan
  5

• Reproducing published score

  #41 opened 5 years ago by htdt
  6

• Pretrained PPO trajectories

  #39 opened 5 years ago by bmazoure
  1

• Figure out a better way to collect samples to train the encoder (other than a random policy)

  #13 opened 5 years ago by ankeshanand
  1

• Run estimators and linear probes on the full (210, 160) image.

  #16 opened 5 years ago by ankeshanand
  0

• CPC implementation

  #2 opened 5 years ago by ankeshanand
  1

• Control: Evaluate how well the pretrained representations perform on control tasks

  #12 opened 5 years ago by ankeshanand
  0

• Positive examples within some window of t-5 to t+5 instead of always t-1

  #32 opened 6 years ago by eracah
  1

• Add VAE baseline

  #24 opened 6 years ago by eracah
  1

• Probe representations from trained models using Uber Atari library

  #30 opened 6 years ago by eracah
  1

• add atari uber rollouts to run_contrastive code

  #33 opened 6 years ago by eracah
  1

• Redo fully supervised, random-cnn probes using atari rollouts

  #29 opened 6 years ago by eracah
  1

• Make interface to generate rollouts using uber atari library

  #28 opened 6 years ago by eracah
  1

• Spatio-Temporal Estimator

  #25 opened 6 years ago by ankeshanand
  1

• Make encoders work for (210, 160) images

  #31 opened 6 years ago by ankeshanand
  1

• Random CNN probe baseline

  #26 opened 6 years ago by eracah
  1

• Run the fully supervised baseline by backproping loss over probes.

  #17 opened 6 years ago by ankeshanand
  2

• Add validation evaluation to contrastive training

  #23 opened 6 years ago by eracah
  1

• Add all verified Atari ram values to the "master dictionary" for probing

  #9 opened 6 years ago by eracah
  2

• Support frame-stacking for the whole training / probing pipeline

  #21 opened 6 years ago by ankeshanand
  1

• Fix bucketing issues

  #15 opened 6 years ago by eracah
  3

• Add validation set for probing

  #14 opened 6 years ago by ankeshanand
  1

• Add a bilinear classifier for Appo estimators

  #4 opened 6 years ago by ankeshanand
  1

• Design document for the probing benchmark

  #10 opened 6 years ago by ankeshanand
  1

• Setup the linear probe pipeline

  #3 opened 6 years ago by ankeshanand
  1

• Create a docker image to run code easily on MSR infra

  #5 opened 6 years ago by ankeshanand
  1