Minecraft Machine Learning Dataset: Generation Instructions

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.

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.

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.

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.

Name		Name	Last commit message	Last commit date
Latest commit History 61 Commits
bot-js		bot-js
bot-python		bot-python
images		images
make-hdf5		make-hdf5
region-dataset-1		region-dataset-1
screen-capture		screen-capture
single-block-dataset-1		single-block-dataset-1
world-processed		world-processed
world		world
.gitignore		.gitignore
README.md		README.md
SERVER.md		SERVER.md
access-world.py		access-world.py
level.dat-backup		level.dat-backup
server.properties		server.properties