Advantage actor-critic models
Q-learning, as we saw in the previous sections, is quite useful but it does have its drawbacks. For example, as we have to estimate a Q value for each action, there has to be a discrete, limited set of actions. So, what if the action space is continuous or extremely large? Say you are using an RL algorithm to build a portfolio of stocks.
In this case, even if your universe of stocks consisted only of two stocks, say, AMZN and AAPL, there would be a huge amount of ways to balance them: 10% AMZN and 90% AAPL, 11% AMZM and 89% AAPL, and so on. If your universe gets bigger, the amount of ways you can combine stocks explodes.
A workaround to having to select from such an action space is to learn the policy, , directly. Once ...
Get Machine Learning for Finance now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
