/minecraft_dreaming

Primary LanguageJupyter Notebook

Minecraft Dreaming

  • Extracting chunk data from a Minecraft server.
  • Training various models on the chunk data.
  • Insert model output back into the Minecraft server
  • ????
  • Profit

Files/notebooks

world_file.py / exact_region_files.ipynb

  • Extracts .mca region files from Minecraft server docker container

replace_world_file.py / copy_mca_to_container.ipynb

  • Creates a backup of region file in container, then inserts source region file into Minecraft docker container

plotting_chunks.ipynb

  • Extracts Minecraft NBT chunk data from .mca files and plots the chunks as voxels.

single_chunk_plot.ipynb

  • Extracts Minecraft NBT chunk data from .mca files and plots the blocks as voxels.

nbt_to_mca_file.ipynb

  • Takes Minecraft NBT chunks and compresses and stores them in .mca region format to be consumed by a Minecraft server

chunk_dummy_default.ipynb chunk_dummy_good.ipynb

  • Simple FFN model that takes an input of a NxNxN numpy array.
  • Splits the NxNxN array into sub chunks.
  • Generate dataset where the input is a sub chunk and the output is the up to 6 sub chunks it shares a face with. The sub chunk is rotated such that input chunk is always facing its output chunk.
  • Training the model to overfit on the dataset to get an idea of number of parameters and settings of hyperparameters.

actual_chunk_2.ipynb

  • Training a FFN model on actual chunk data.
  • Splitting a 16x16x16 into 2x2x2 chunks.
  • Checking to see how many one-to-many relationships this causes as with sub chunks this small its very possible to have exact chunks have a large variance in adjacent chunks.

actual_chunk_4.ipynb

  • Training a FFN model on actual chunk data.
  • Splitting a 16x16x16 into 4x4x4 chunks.
  • Increased sub chunk size reduces the number of one-to-many relationships at the cost of model size.

create_chunks_dataset.ipynb

  • Loads region files, extracts NBT chunks.
  • Creates a combined palette that contains all block_state + biome combinations.
  • Creates a 16x16x16 numpy array for each sub chunk, where the value represents an index in the combined block+biome palette.