gym-display-advertising

An OpenAI Gym for Display Advertisment Reinforcement Learning

Installation

pip install gym-display-advertising

Usage

import gym_display_advertising

env = gym.make("StaticDisplayAdvertising-v0")
episode_over = False
while not episode_over:
    state, reward, episode_over, _ = env.step(env.action_space.sample())
    print(state, reward)

import gym_display_advertising

env = gym.make("DisplayAdvertising-v0")
episode_over = False
while not episode_over:
    state, reward, episode_over, _ = env.step(env.action_space.sample())
    print(state, reward)

Real Ad Bidding Data

The repository contains real-life bidding data from a single merchant and loads this by default. If you want to load more data follow the instructions in the make-ipinyou-data repository to create the data.

Then use the helper class ProcessedIPinYouData to load the data and pass the dataframe into the gym.make command.

import pathlib
import gym
import gym_display_advertising

ipinyou = gym_display_advertising.data.ProcessedIPinYouData(directory=pathlib.Path("path/to/file"))
training_data, _ = ipinyou.get_merchant_data(2997)
env = gym.make("DisplayAdvertising-v0", data=training_data)
state, reward, _, _ = env.step(env.action_space.sample())
print(state, reward)

Acknowledgements

This is a project by Winder Research, a Cloud-Native Data Science consultancy.