passing kwargs to tensorforce environment via Runner

[rolandrmgservices]

Thanks for such a useful codebase. I had a question re customizing environment creation via the runner class.

Say we have an environment like the ThermostatEnvironment defined in: https://github.com/tensorforce/tensorforce/blob/master/examples/temperature-controller.ipynb

but the constructor takes tau as a parameter:

    def __init__(self, tau):
        ## Some initializations.  Will eventually parameterize this in the constructor.
        self.tau = tau
        self.current_temp = np.random.random(size=(1,))

        super().__init__()

We then want to run this environment with the Runner, like:

    ...
    runner = Runner(agent=agent, environment=ThermostatEnvironment, num_parallel=4)
    runner.run(num_episodes=100, batch_agent_calls=True)
    runner.close()

But we want to configure the environment with a different tau for a variety of different runs. So ideally we could do something like:

    ...
    runner = Runner(agent=agent, environment=ThermostatEnvironment, num_parallel=4, tau=3.0)
    runner.run(num_episodes=100, batch_agent_calls=True)
    runner.close()

    ...
    runner = Runner(agent=agent, environment=ThermostatEnvironment, num_parallel=4, tau=6.0)
    runner.run(num_episodes=100, batch_agent_calls=True)
    runner.close()

However, the Runner constructor doesn't pass kwargs to the Environment.create in:

tensorforce/tensorforce/execution/runner.py

Lines 168 to 171 in 868d12d

	environment = Environment.create(
	environment=environments[0], max_episode_timesteps=max_episode_timesteps,
	remote=remote, blocking=blocking, host=host[0], port=port[0]
	)

Given the above, is there a way to pass keywords through to customize the Tensorforce environment that is being created by the runner? I saw that it is possible to define a spec via JSON, but this would require writing different JSON files for each environment variation to be created.