This repo contains experimental scripts for Vereya mod, which go beyond examples presented in minecraft-demo
You need tagilmo library (VereyaPython module) to be installed to work with this experiments.
pip install git+https://github.com/trueagi-io/minecraft-demo.git
dataset for pixel classification model
https://drive.google.com/drive/folders/1FpcZvUGVWG_oz1s_gnlc9X82hx6qaIXA?usp=sharing
Some experiments require additional libraries.
To run experiments with goodpoint detector you'll need: scikit-learn - pip3 install -U scikit-learn PyTorch-Ignite - pip install pytorch-ignite
To run get_recipes_from_minecraft_data.py pip install minecraft_data
To run experiments in test_skills.py you'll need pip install -U scikit-image
Vision training experiments are deprecated (.pt reading needs to be fixed + torch version)
cliff_v1 - non-visual target-approaching agent (moved to deprecated)
cliff_v2 - visual target-approaching agent. Used in 5_mine_and_craft_with_a_pinch_of_rl_skills.py, which is available for MalmoPython lib and minecraft 1.11.2. (moved to deprecated)
cliff_v3 - Q-learning based visual target-approaching agent (moved to deprecated)
behaviour folder contains template of And-Or behaviour tree and implementation of experimental behaviours
block_analysis folder contains experiments that collect different visual features statistics for every block seen by an agent
deprecated folder contains experiments that normally work with minecraft 1.12.1 but no longer supported for 1.18.* and higher
external folder contains necessary modules to run the experiments
trainable folder contains experimental scripts for trained/(potentially) trainable agent's micro skills
utils folder contains auxiliary modules for training, preparation of datasets and other non-primary functionality.
vision_tests folder contains experimental scripts for testing "visual perception" procedures