An in-development book teaching foundational ideas in reinforcement learning with examples in finance.
Reinforcement Learning (RL) is emerging as a viable and powerful technique for solving a variety of complex business problems across industries that involve Sequential Optimal Decisioning under Uncertainty. Although RL is classified as a branch of Machine Learning (ML), it tends to be viewed and treated quite differently from other branches of ML (Supervised and Unsupervised Learning). Indeed, RL seems to hold the key to unlocking the promise of AI – machines that adapt their decisions to vagaries in observed information, while continuously steering towards the optimal outcome. It’s penetration in high-profile problems like self-driving cars, robotics and strategy games points to a future where RL algorithms will have decisioning abilities far superior to humans.
The book and its codebase are under active development. A recent PDF version is available online, but the only way to get the most up-to-date PDF is to generate it yourself following the instructions in this README.
Basic setup for working with Pandoc and TeX.
To work on the book, you need to install Nix.
On macOS, first install the XCode command-line tools:
xcode-select --install
the install Nix with:
sh <(curl https://nixos.org/nix/install) --darwin-use-unencrypted-nix-store-volume
On Linux, install Nix with:
curl -L https://nixos.org/nix/install | sh
Once you have Nix installed, run nix-shell to get access to Pandoc, LaTeX and all the other tools you need. The first time you run nix-shell will take a while to finish as it downloads and installs all the packages you need.
Once inside the Nix shell, you'll have access to Pandoc and you'll be able to generate PDFs with XeTeX. The to-pdf script can do this for a single chapter in the book directory:
[nix-shell:~/Documents/RL-book]$ bin/to-pdf chapter0
Converting book/chapter0/chapter0.md to book/chapter0/chapter0.pdf
You can also generate the entire book to a file called book.pdf:
[nix-shell:~/Documents/RL-book]$ bin/to-pdf
Combining
book/chapter0/chapter0.md
book/chapter2/chapter2.md
book/chapter3/chapter3.md
book/chapter4/chapter4.md
book/chapter5/chapter5.md
into book.pdf
Note that this can take a little while (10–20 seconds for chapters 0–5).
We can define labels for chapters and headings:
# Overview {#sec:overview}
## Learning Reinforcement Learning {#sec:learning-rl}
Because of limitations with the system I'm using for managing internal references, labels for sections and chapters always have to start with sec:.
Once you have defined a label for a section or chapter, you can reference its number as follows:
Take a look at Chapter [-@sec:mdp].
Take a look at Chapter 3.
For sections, you can also use:
Take a look at [@sec:learning-rl].
Take a look at sec. 1.
(The [-@sec:foo] syntax drops the "sec. " text.)
For references across chapters to render correctly, you have to compile the entire book PDF (following the instructions above).
We can manage our Python dependencies with a venv.
First, create a venv from inside a Nix shell:
> nix-shell
[nix-shell:~/Documents/RL-book]$ python -m venv .venv
Then, each time you're working on this project, make sure to activate the venv:
> source .venv/bin/activate
(This can now be done even outside a Nix shell.)
Once the venv is activated, you should see a (.venv) in your shell prompt:
(.venv) RL-book:RL-book>
Now you can use pip to install dependencies inside the venv:
(.venv) RL-book:RL-book> pip install matplotlib
To make this reproducible, we can save the libraries to a requirements.txt file:
(.venv) RL-book:RL-book>pip freeze > requirements.txt
Then, when somebody is starting, they can install every Python package they need using:
(.venv) RL-book:RL-book>pip install -r requirements.txt