This repository contains code to generate datasets to train and benchmark a pix2pix model for predicting the outcome of a move in the game Spread: Transmission.

Implementation of pix2pix | Spread: Transmission on Steam

Prerequisites

• Node.js 18

Getting Started

Installation

• Clone this repo:

git clone https://github.com/cxve/transmission-dataset-maker
cd transmission-dataset-maker

• Install dependencies with npm install

Generate a dataset for training

• Run the generate script for a specified number of simulations (e.g. 10):

node generate 10

• To sort the dataset by amount of iterations used to resolve the changes (flat distribution), use the -fd flag

  • Use the flatdist_compiler script, to sort the dataset for usage with the pix2pix implementation

• Use the dataset to train a pix2pix model

  • Copy the ./datasets/name/sorted/ folder to /path/to/pix2pix/datasets/name
  • Learn how to use the pix2pix implementation here

Sort a dataset based on the changes

• Run the sort_by_changes script for a specified path (e.g. ./datasets/name):

node sort_by_changes ./datasets/name

• To sort the dataset based on the relative amount of changes, use the -% flag

• Use the flatdist_compiler script, to sort the dataset for usage with the pix2pix implementation

Compile a flat distribution dataset

• Run the flatdist_compiler script for a specified path (e.g. ./datasets/name/flatdist2) and an optional amount of images to include (e.g. 60):

node flatdist_compiler ./datasets/name/flatdist2 60

• Use the dataset to train a pix2pix model
  • Copy the ./datasets/name/flatdist2/sorted/ folder to /path/to/pix2pix/datasets/name

Analyze training / test results quality

• Run the analyze script for a specified path (e.g. ./results/name):

node analyze ./results/name

• Results are saved as name.csv

Benchmark recursive implementation

• Run the benchmark script:

node benchmark

• Use the dataset to benchmark a pix2pix model
  • Copy the ./datasets_test/testset_name/ folder to /path/to/pix2pix/datasets/name/testset_name