Skip to content

cosmoharrigan/minecraft-dataset-generation

Repository files navigation

Minecraft Machine Learning Dataset: Generation Instructions

Note: This is still in-progress. The initial release has not yet occurred.

Overview

The files needed to generate the Minecraft dataset are provided here.

The dataset is also available on Amazon S3 in HDF5 format upon request.

Contents

  • make-hdf5 - Jupyter notebook. After capturing Minecraft screenshots into png files, this utility produces an HDF5 datafile from them. Includes examples of how to create and how to read the HDF5 datafile.

  • world - The world folder in the Minecraft server folder should be replaced with this folder. For example:

$ rm -r /Users/cosmo/minecraft-server/world
$ cp -R ~/minecraft-dataset-generation/world /Users/cosmo/minecraft-server/
  • server.properties - The server.properties file in the Minecraft server folder should be replaced with this file.
  • world-processed - This is what the world folder is transformed into after it is loaded the first time by the Minecraft server.
  • bot-js - node.js bot for placing blocks in the world and changing and recording agent position and orientation
  • bot-python - Python bot for placing blocks in the world and changing and recording agent position and orientation
  • screen-capture - Optional Java utility for taking a screenshot at a regular interval and saving to a file. Currently not used, as the preferred method is to use the screenshot capability contained within bot-js.

Additional contents (optional)

  • access-world.py - Illustrates how to use the pymclevel Python interface to interact with world data by printing the 2D NumPy array for each slice of the world that contains blocks.

Screenshots

Minecraft world regions

Screenshot

Inside the Minecraft client

Screenshot Screenshot Screenshot

Using the MCEdit level editor

Screenshot

Instructions

  • Download MCEdit Unified as a GUI world editor
  • Clone the MCEdit Unified repository to obtain the Python programmatic world editor.
  • To run access-world.py, cd to the MCEdit-Unified folder that you cloned from the repository, and run python2 ~/minecraft-dataset-generation/access-world.py (replace with the correct path on your system as needed.)
  • If you want to add additional functionality beyond that contained in access-world.py, the file pymclevel/mclevel.py in the MCEdit-Unified repository provides a useful explanation of how to interact with the pymclevel API.

Running the custom Bukkit server

  • See the instructions in SERVER.md

About

Annotated Minecraft dataset for machine learning

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published