gengsinong / samq

State Aggregation Minimizing Q error

State aggregation minimizing Q error (SAmQ) is a toolkit to conduct state aggregation for dynamic discrete choice model estimation.

Installation

pip install git+https://github.com/gengsinong/SAmQ.git@master

Main Funcion

The main function is in main.py. It includes a list of parameters.

python main.py

Tuning

The tunning is conducted by submit_wandb_airline.sh, submit_wandb_bus_engine.sh and submit_wandb_bus_engine_plot.sh. This file uses wandb.

Data Generation

Bus Engine Replacement Analysis

To generate the decision-making data for dynamic discrete choice model estimation, we consider a synthetic setting similar to the bus engine application in the Rust model. The specific setting is defined and detialed in env.bus_env. Then, we conduct soft-q iteration to estimate the optimal choice-specific value function, and generate data in rl.softQ.

Airline Market Entry analysis

Experiments on airline market entry analysis leverages the data in airline_data.

State Aggregation

The state aggregation step is conducted in irl.pqr. The aggregation uses the deep PQR method.