Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/andreaconti/supermario-dqn
Deep Reinforcement Learning Agent for Super Mario Bros
https://github.com/andreaconti/supermario-dqn
deep-learning python3 reinforcement-learning super-mario-bros-ai
Last synced: 27 days ago
JSON representation
Deep Reinforcement Learning Agent for Super Mario Bros
- Host: GitHub
- URL: https://github.com/andreaconti/supermario-dqn
- Owner: andreaconti
- Created: 2019-09-11T13:49:56.000Z (about 5 years ago)
- Default Branch: master
- Last Pushed: 2019-09-28T21:18:10.000Z (about 5 years ago)
- Last Synced: 2024-10-23T03:33:20.986Z (28 days ago)
- Topics: deep-learning, python3, reinforcement-learning, super-mario-bros-ai
- Language: Jupyter Notebook
- Homepage:
- Size: 75.1 MB
- Stars: 2
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.rst
- Authors: AUTHORS.rst
Awesome Lists containing this project
README
supermario-dqn
==============
Deep Reinforcement Learning Agent for Super Mario Bros using OpenAI gym
[gym-super-mario-bros](https://github.com/Kautenja/gym-super-mario-bros) environment
## Usage
~~~shell
# create virtual env in the project folder
$ python3 -m venv .venv
$ source .venv/bin/activate
# install dependencies
$ pip3 install -r requirements
# use
$ supermario_train -h
usage: supermario_train [-h] [--batch_size BATCH_SIZE]
[--fit_interval FIT_INTERVAL] [--gamma GAMMA]
[--eps_start EPS_START] [--eps_end EPS_END]
[--eps_decay EPS_DECAY]
[--target_update TARGET_UPDATE]
[--save_path SAVE_PATH] [--memory_size MEMORY_SIZE]
[--num_episodes NUM_EPISODES] [--resume RESUME]
[--checkpoint CHECKPOINT] [--random] [--render]
[--world_stage WORLD_STAGE WORLD_STAGE]
[--actions ACTIONS] [--test TEST] [--log]
Handle training
optional arguments:
-h, --help show this help message and exit
--batch_size BATCH_SIZE
size of each batch used for training
--fit_interval FIT_INTERVAL
fit every `fit_interval` examples available
--gamma GAMMA discount rate used for Q-values learning
--eps_start EPS_START
start probability to choose a random action
--eps_end EPS_END end probability to choose a random action
--eps_decay EPS_DECAY
decay of eps probabilities
--target_update TARGET_UPDATE
number of episodes between each target dqn update
--save_path SAVE_PATH
where save trained model
--memory_size MEMORY_SIZE
size of replay memory
--num_episodes NUM_EPISODES
number of games to be played before end
--resume RESUME load from a checkpoint
--checkpoint CHECKPOINT
number of episodes between each network checkpoint
--random choose randomly different worlds and stages
--render rendering of frames, only for debug
--world_stage WORLD_STAGE WORLD_STAGE
select specific world and stage
--actions ACTIONS select actions used between ["simple"]
--test TEST each `test` episodes network is used and tested over
an episode
--log logs episodes results
# play
$ supermario_play -h
usage: play a game [-h] [--world_stage WORLD_STAGE WORLD_STAGE] [--skip SKIP]
[--processed]
model
positional arguments:
model neural network model
optional arguments:
-h, --help show this help message and exit
--world_stage WORLD_STAGE WORLD_STAGE
select a specific world and stage, world in [1..8],
stage in [1..4]
--skip SKIP number of frames to skip
--processed shows frames processed for neural network
~~~
### Results
~~~bash
$ supermario_play --skip 5 --world_stage 1 1 trained/train_1_1/model.pt
~~~
| rewards | play gif |
|---------|----------|
|| |