Code Structure

The TensorTrade library is modular. The tensortrade library usually has a common setup when using components. If you wish to make a particular class a component all you need to do is subclass Component.

class Example(Component):
    """An example component to show how to subclass."""

    def foo(self, arg1, arg2) -> str:
        """A method to return a string."""
        raise NotImplementedError()

    def bar(self, arg1, arg2, **kwargs) -> int:
        """A method to return an integer."""

From this abstract base class, more concrete and custom subclasses can be made that provide the implementation of these methods.

Example of Structure
A good example of this structure is the RewardScheme component. This component controls the reward mechanism of a TradingEnv.

The beginning of the code in RewardScheme is seen here.

from abc import abstractmethod

from tensortrade.core.component import Component
from tensortrade.core.base import TimeIndexed

class RewardScheme(Component, TimeIndexed):
    """A component to compute the reward at each step of an episode."""

    registered_name = "rewards"

    def reward(self, env: 'TradingEnv') -> float:
        """Computes the reward for the current step of an episode.

        env : `TradingEnv`
            The trading environment

            The computed reward.
        raise NotImplementedError()

    def reset(self) -> None:
        """Resets the reward scheme."""

As you can see above, the RewardScheme has a majority of the structural and mechanical details that guide all other representations of that type of class. When creating a new reward scheme, one needs to add further details for how information from then environment gets converted into a reward.