Skip to content

Commit

Permalink
arxiv preprint ready
Browse files Browse the repository at this point in the history
  • Loading branch information
tae898 committed Dec 5, 2022
1 parent fe33398 commit bda586b
Show file tree
Hide file tree
Showing 2,130 changed files with 280,327 additions and 555,137 deletions.
1 change: 1 addition & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -141,3 +141,4 @@ junks/

# tests
train_test_dir_to_remove/
DELETE_SOON
6 changes: 4 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
For the documentation of [the older paper](https://arxiv.org/abs/2204.01611), check out
[this documentation](./v0/README-v0.md)

This repo is to train an agent that interacts with the [RoomEnv-v2](https://github.com/tae898/room-env).
This repo is to train an agent that interacts with the [RoomEnv-v1](https://github.com/tae898/room-env).
See the [paper](todo/update/the/paper) for more information.

## Prerequisites
Expand All @@ -26,7 +26,9 @@ checkpoints will be saved at `./training_results/`

| Average loss, training. | Average total rewards per episode, validation. | Average total rewards per episode, test. | Average total rewards, varying capacities, test. |
| :-----------------------------------------------------: | :----------------------------------------------------------------: | :-----------------------------------------------------------------: | :------------------------------------------------------------------: |
| ![](./figures/des_size=l-capacity=32-train_loss-v1.svg) | ![](./figures/des_size=l-capacity=32-val_total_reward_mean-v1.svg) | ![](./figures/des_size=l-capacity=32-test_total_reward_mean-v1.svg) | ![](./figures/des_size=l-capacity=all-test_total_reward_mean-v1.svg) |
| ![](./figures/des_size=l-capacity=32-train_loss-v1.png) | ![](./figures/des_size=l-capacity=32-val_total_reward_mean-v1.png) | ![](./figures/des_size=l-capacity=32-test_total_reward_mean-v1.png) | ![](./figures/des_size=l-capacity=all-test_total_reward_mean-v1.png) |

There is one different thing in the RoomEnv from the original AAAI-2023 paper: The reward is either +1 or -1, instead of +1 or 0.

Also check out [`./models/`](./models) to see the saved training results. The `test_debug`
results might especially be interesting to you.
Expand Down
Loading

0 comments on commit bda586b

Please sign in to comment.