Improve Python-API.md with an example

[ucacaxm]

Hi,

First, thank you for you great work in ml-agents !

Since it was not so obvious to create a simple random agent
with the python API I would like to improve the doc/Python-API.md with an example.
And also fix the import:
from mlagents.envs.environment import UnityEnvironment

A simple example of random agent

import sys
import numpy as np

from mlagents.envs.environment import UnityEnvironment


"""The world's simplest agent!"""
class RandomAgent(object):
    def __init__(self, action_space):
        print("Action size="+str(action_space))
        self.action = np.empty((action_space), dtype=float)

    def act(self, observation, reward, done):
        for j in range(self.action.shape[0]):
            self.action[j] = -1.0 + 2.0 * np.random.random_sample()
        return self.action




if __name__ == '__main__':
    env = UnityEnvironment(file_name="3DBall", worker_id=0, seed=1)  # 3DBall executable should exist

    episode_count = 100
    reward = 0
    done = False

    brain = env.brains['BrainRoller']       # See brain name in the Unity Editor
    print("brain="+str(brain))
    print("number of agents=" + str(len(brain.vector_action_space_size)))
    print("action space type="+str(brain.vector_action_space_type))
    print("action space size="+str(brain.vector_action_space_size))
    print("observation space size=" + str(brain.vector_observation_space_size))
    print("=================================")

    info = env.reset()
    agent = RandomAgent(brain.vector_action_space_size)

    agentId = 0            # assume onmy one agent
    for i in range(episode_count):
        info = env.reset()
        gameover = False
        totalReward = 0
        while not gameover:
            brainInfo = info['BrainRoller']
            action = agent.act( brainInfo.vector_observations[agentId], brainInfo.rewards[agentId], brainInfo.local_done[agentId])
            totalReward += brainInfo.rewards[agentId]
            info = env.step(action)
            brainInfo = info['BrainRoller']
            gameover = brainInfo.local_done[agentId]

        print("Game over, total reward=" + str(totalReward))
    env.close()

[awjuliani]

Hi @ucacaxm,

Thanks for making this PR. I am somewhat hesitant to add this, since we already have an interactive jupyter notebook dedicated to describing the same process: https://github.com/Unity-Technologies/ml-agents/blob/master/notebooks/getting-started.ipynb. I see that we don't link to it from the Python-API page. I will make a point of adding a link to it, which will hopefully make things easier for first-time users.

[awjuliani]

I've made a PR with the change here: #1311. I think keeping the notebook as the main example is the best way to go, so I will close this PR for now.

[ucacaxm]

I did not found the notebook before: it is perfect for this purpose  Yes you are right just linking to the notebook would be better. Thanks pointing me to the notebook, Best Alex mercredi, 10 octobre 2018, 02:03AM +02:00 de Arthur Juliani notifications@github.com :

…

I've made a PR with the change here: #1311 . I think keeping the notebook as the main example is the best way to go, so I will close this PR for now. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub , or mute the thread .