gowun / rl_for_reco

Deep Reinforcement Learning for Business Structured Data

---

Item_Reco

---

A class to recommend products to customers with their any current information and product-recommended history. Class variable items indicates the products as well as their associate promotions, offers such as any recommendation type. If you want to take a case where customers have not recommendation, you can use 'none' to represent the case. States, actions and reward are respectively n-dim array, 1-d array and a float number. A transition model, state + action => (state, reward), is assumed as a multi-output neural network on TorchModel.

This framework, actually, is applicable to problems of any structured data.

Network_for_Reco

---

A class to update Q-values though a nueral network. This is also a general form avaiable to any problem.

RL_Learn

---

A class to formulate a Deep Q Learning problem(an environment, an agent and its policy and associated parameters) and to learn the agent by a Deep Q Network and its approximator.

TorchModel

---

Several classes to build a neural network by pyTorch.