# Public Dispatch Simulator — Experimental Code (Process Overview)

This repository contains the experimental code accompanying the experiment *Public Dispatch Simulator*. It consists of five core scripts executed in the following order to complete the full pipeline:

1. **`fast-V.py`**  
   Computes and exports the **V function** required by the MDP-based policy.

2. **`fit_envir.py`**  
   Trains a **neural network environment model** (e.g., transition/reward dynamics).

3. **`stochastic.py`**  
   Estimates and saves the **stochastic noise/perturbation artifacts** used for downstream simulation and robustness analyses.

4. **`train.py`**  
   Trains the **Transformer DQN** policy under the learned environment and noise settings.

5. **`eval.py`**  
   Evaluates and compares **multiple methods**, reporting MSE.

> The intermediate artifacts produced in earlier steps (V function, environment model, noise files, and policy weights) are consumed by later stages, forming a complete pipeline from value precomputation → environment modeling → stochastic modeling → policy training → evaluation.

## Hardware Recommendation

This project is **recommended to run on a GPU-equipped machine** (CUDA-enabled).