Skip to content
DotaService is a service to play Dota 2 through gRPC
Branch: master
Clone or download
Latest commit dfe8823 Mar 18, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
botcpp
docker Fix loading of steam lib. Fixes #54. Mar 11, 2019
dotaservice
resources
scripts
.gitignore Add PO trainer that moves with reward from xp and hp Dec 6, 2018
.style.yapf Add YAPF autoformatter and autoformat the py files Dec 3, 2018
LICENSE
NOTES.md
README.md
demo.gif
dotaservice-icon.png Added logo on Conrado's request Dec 4, 2018
dotaservice.png
setup.py

README.md

DotaService

dotaservice icon


NOTE: The project that uses the dotaservice in a k8s environment is the DotaClient repo.

DotaService is a service to play Dota 2 through gRPC. There are first class python bindings and examples, so you can play dota as you would use the OpenAI gym API.

It's fully functional and super lightweight. Starting Dota obs = env.reset() takes 5 seconds, and each obs = env.step(action) in the environment takes between 10 and 30 ms.

You can even set the config of render=True and you can watch the game play live. Each game will have a uuid and folder associated where there's a Dota demo (replay) and console logs.

demo

Run DotaService Locally

Run the DotaService so you can connect your client to it later. Only one client per server is supported, and only one DotaService per VM (eg local or one per docker container).

python3 -m dotaservice
>>> Serving on 127.0.0.1:13337

Run DotaService Distributed

See docker/README.md.

To run two dockerservice instances, one on port 13337 and one on 13338, f.e. run:

docker run -dp 13337:13337 ds
docker run -dp 13338:13337 ds

You can run as many as you want, until you run out of ports or ip addresses. If you are wearing your fancy pants, use Kubernetes to deploy gazillions.

Client Code

from grpclib.client import Channel
from protobuf.DotaService_grpc import DotaServiceStub
from protobuf.DotaService_pb2 import Action
from protobuf.DotaService_pb2 import Config

# Connect to the DotaService.
env = DotaServiceStub(Channel('127.0.0.1', 13337))

# Get the initial observation.
observation = await env.reset(Config())
for i in range(8):
    # Sample an action from the action protobuf
    action = Action.MoveToLocation(x=.., y=.., z=..)
    # Take an action, returning the resulting observation.
    observation = await env.step(action)

This is very useful to provide an environment for reinforcement learning, and service aspect of it makes it especially useful for distributed training. I am planning to provide a client python module for this (PyDota) that mimics typical OpenAI gym APIs. Maybe I won't even make PyDota and the gRPC client is enough.

dotaservice connections

Requirements

  • Python 3.7
  • Unix: MacOS, Ubuntu. A dockerfile is also provided see: docker/README.md.

Installation

Installing from pypi:

pip3 install dotaservice

For development; installing from source:

pip3 install -e .

(Optional) Compile the protos for Python (run from repository root):

python3 -m grpc_tools.protoc -I. --python_out=. --python_grpc_out=. --grpc_python_out=. dotaservice/protos/*.proto

Notes

My dev notes: NOTES.md.


Acknowledgements

You can’t perform that action at this time.