PyTorch 1.x Reinforcement Learning Cookbook

There's more...

You may want to know whether the optimal policy really works. Let's act like smart gamblers and play 10,000 episodes of the game. We are going to compare the optimal policy with two other strategies: conservative (betting one dollar each round) and random (betting a random amount):

1. We start by defining the three aforementioned betting strategies.

We define the optimal strategy first:

>>> def optimal_strategy(capital):...     return optimal_policy[capital].item()

Then we define the conservative strategy:

>>> def conservative_strategy(capital):...     return 1

Finally, we define the random strategy:

>>> def random_strategy(capital):...     return torch.randint(1, capital + 1, (1,)).item()

2. Define a wrapper function that runs one episode ...