Export BRAIN to run locally

[sergiobarrionuevo]

Hi,
I've tried the carpole demo, trained, then exported the BRAIN. Pull the docker to my machine, run it...
Went to locahost:5000 on my browser and it returns "Could not find a matching route". Am I doing something wrong?
Is there any example I can use to understand how to use the docker image?

Thank You!

Regards,

[#2 · Accepted solution]

@sergiobarrionuevo 

Thank you for trying the Cartpole example.

You must first run the docker image:
docker run -d \
  -p 5000:5000 \
  --name cartpole-brain \
  myworkspace.azurecr.io/bonsai/cartpole-sample:1-linux-amd64
 
Note as of now, only three types of process architectures are supported: 1) x64, 2) arm64v8, 3) arm32v7.
Then you must test with curl on a host:
 
curl -X GET http://localhost:5000/v1/prediction \
     -d '{"cart_position": 0.0, "cart_velocity": 0.1, "pole_angle": 0.0, "pole_angular_velocity": 0.1}'
 
Then you can test with a python script test-cartpole.py:
 
import sys
import requests
__HOST = sys.argv[1]
def predict(host: str, 
            cart_position: float, 
            cart_velocity: float,
            pole_angle: float,
            pole_angular_velocity: float) -> dict:
  url = "{host}/v1/prediction"
  state = {
    "cart_position": cart_position,
    "cart_velocity": cart_velocity,
    "pole_angle": pole_angle,
    "pole_angular_velocity": pole_angular_velocity
  }
  response = requests.get(url, json=state)
  action = response.json()
  return action
action = predict(__HOST, 0.0, 0.1, 0.0, 0.1)
 
Then you must run the python script on host:
python test-cartpole.py http://localhost:5000`