Deep Q-learning for playing flappy bird game

License: MIT License

Python 100.00%

reinforcement-learning deep-q-network deep-q-learning pytorch pygame

flappy-bird-deep-q-learning-pytorch's Introduction

[PYTORCH] Deep Q-learning for playing Flappy Bird

Introduction

Here is my python source code for training an agent to play flappy bird. It could be seen as a very basic example of Reinforcement Learning's application.

Result

How to use my code

With my code, you can:

Train your model from scratch by running python train.py
Test your trained model by running python test.py

Trained models

You could find my trained model at trained_models/flappy_bird

Requirements

python 3.6
pygame
cv2
pytorch
numpy

flappy-bird-deep-q-learning-pytorch's People

Contributors

Stargazers

Watchers

Forkers

tramper2 faisalshahbaz 7tony treetrees mgonz21 juanlp jahidme thatsankur nguyenducnhaty burakakrishna chichak lamsking alexlevn aminekha hlehong041817 nondukishor ksrath0re hanntonkin insad pete1313 amrlotfy77 michaelbernstein samimideksa zorrock mountain-ai victor8733 einspyon astraylinux sprinterzzj jiaodaxiaozi trendingtechnology b4nk4i arunpoochelvan cmcai0104 phrmgb yoyojacky ankitnamdeo34 andrecnf quangtm09 nooralahzadeh kaliham anandpawara narendraakumar jamesawright zoid-anurag kunal-varma esmaeilinia sgrives 5tr1k3r dp167 znatz miendinh tdtrinh11 tomheaven thanhhoang283 gene-smith henriquedezani caoshunwu quan821223 warlock2195 batermj ambiguouserror chenmeng0508 zhangzhao4444 qintuyuan lhcshine yeahalti ming-er saajidpasha cxuxin wbyz wyfzidane xjzpguob mickgray1990 bigrobinson barryzm iamsile rochzhangyan huguensjean bqdqj mustafaomerguclu pepijnt dieptran43 roschiweiming mayankanand007 zhuhang0796 tinaxv vuongmt2000 mkygogo lipanr ab-jiteshlalwani wesley-yang timefly-1989 xueliu8617112 prettywork2021 devherles chuongloc nuomifan zhy109 ankurhcu

flappy-bird-deep-q-learning-pytorch's Issues

Error Running Train.py

Flappy-bird-deep-Q-learning-pytorch/src/deep_q_network.py:21: UserWarning: nn.init.uniform is now deprecated in favor of nn.init.uniform_.
nn.init.uniform(m.weight, -0.01, 0.01)
Perform a random action
Iteration: 2/2000000, Action: 1, Loss: 0.010123813524842262, Epsilon 0.1, Reward: 0.1, Q-value: 0.0005683371564373374
Traceback (most recent call last):
File "train.py", line 133, in
train(opt)
File "train.py", line 74, in train
action = torch.argmax(prediction)[0]
IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number
(python36) bash-3.2$

Can we use this code to train other games

I want to train dino dragon chrome game can I use this code for it

Error in dimension of image tensor

First of all, thanks for taking the time to put this pytorch implementation together. I am getting the following error. Also can you explain what the purpose of the torch.cat line is?

Traceback (most recent call last):
File "train.py", line 133, in
train(opt)
File "train.py", line 58, in train
state = torch.cat(tuple(image for _ in range(4)))[None, :, :, :]
IndexError: too many indices for tensor of dimension 2

IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number

I'm trying out your code and I followed the instructions to leave the environment compatable, I have the following modules installed:

numpy==1.16.3
opencv-contrib-python==4.1.0.25
opencv-python==4.1.0.25
Pillow==6.0.0
protobuf==3.7.1
pygame==1.9.6
six==1.12.0
tensorboardX==1.6
torch==1.0.1
torchvision==0.2.2.post3

But with minor nuances, as I'm using virtualenv to isolate the installed modules.

When executing your code I get the following message:

$ python train.py
pygame 1.9.6
Hello from the pygame community. https://www.pygame.org/contribute.html
libpng warning: iCCP: known incorrect sRGB profile
libpng warning: iCCP: known incorrect sRGB profile
libpng warning: iCCP: known incorrect sRGB profile
libpng warning: iCCP: known incorrect sRGB profile
libpng warning: iCCP: known incorrect sRGB profile
.\workspace\MachineLearning\Flappy-bird-deep-Q-learning-pytorch\src\deep_q_network.py:21: UserWarning: nn.init.uniform is now deprecated in favor of nn.init.uniform_.
  nn.init.uniform(m.weight, -0.01, 0.01)
tensor(0)
Traceback (most recent call last):
  File "train.py", line 133, in <module>
    train(opt)
  File "train.py", line 74, in train
    action = torch.argmax(prediction)[0]
IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python number

What I suggest to fix this problem, I have not programmed with Python for some years, and I do not know the frameworks used with Python. I have already used OpenCV with pure C / C ++ on board and succeeded.

Change in code after pytorch upgrade

HI,

on line 74:
action = torch.argmax(prediction)[0]

should be:

action = torch.argmax(prediction)

after that the code ran

AttributeError: 'Conv2D' object has no attribute 'padding_mode'

This is the error I'm getting if I'm trying to run test.py with PyTorch 1.1

It's just a heads-up - this doesn't seem to be a problem on your end.
Downgrading to PyTorch 1.0.1 solved the issue for me.

uvipen / flappy-bird-deep-q-learning-pytorch Goto Github PK