liveBook · Manning

Action

As explained in Experimentation for Engineers

A candidate choice made by a contextual bandit, such as displaying one particular post.

An action is a candidate decision available to a contextual bandit. In a post-display system, displaying one particular post is an action; the selected action depends on the user context and produces a measurable reward such as viewing time. section 5.1.1

Choosing a post

When a user opens the app, the system chooses one post from many possible posts. Different user features can lead the system to choose different actions, while viewing time measures the outcome of that choice. section 5.1.1

Listing 5.2 A logged sample
class Sample:
    def __init__(self, context, action, reward):
        self.context = context    #1
        self.action = action      #2
        self.reward = reward      #3

Inspect the logged sample structure, which records the context, selected action, and resulting reward for later modeling. section 5.1.1

How Experimentation for Engineers covers it

How deep it goes
Defines action through post-display examples and logging, but stops short of detailed action-selection policies.
Examples
Uses displaying one post among many and logging context, action, and reward as concrete examples.
What it assumes
Assumes contextual-bandit vocabulary while defining context and reward alongside action.
How it connects
Connects action to contextual bandits, context, reward, viewing time, and prediction models.

Also explained in

Compare how 13 books explain action →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Experimentation for Engineers. The section links point to the parts of the book it draws on.