2022-HybridIntelligence

Notes

test_submission_code.py and train_submission_code.py are templates provided at the competition basalt_utils and kairos_minerl are packages prepared by the authors and are included in requirements.txt folder "data" is where the dataset will be downloaded "train.zip" include trained models so feel free to use them

Instructions:

install Ananaconda

Prepare the Environment

conda-env create -f environment_gpu.yml # with gpu
conda-env create -f environment_cpu.yml # without gpu

conda activate basalt

pip install -r requirements.txt

(it might be better to install packages seperately when running each code because there has been some issues of incompatible versions) make sure that minerl version is 0.4.4

Data Processing

1. download dataset

   python downloadBasaltDataset.py
   

2. extract images and actions from the recorded videos

   python dataProcessing.py
   

3. (optimal) open GUI to label image frames.

   you can skip this and just use the labels folder in the repository after extracting it

   python labelDataGUI.py
   

4. create training data for state classifier ( I forgot that when i worked on this i manually copied images from /data to the corresponding folder in /labels before running this code. I should have written a code for that)

   python compileLabels.py
   

5. training

   python stateClassifier.py
   

6. cloning behavior

   python train.py
   

7. still doesn't work as there was weirdly an issue with CUDA just there

   python test.py 
   

Supply

Original repository: https://github.com/viniciusguigo/kairos_minerl_basalt.git

paper citation: https://arxiv.org/abs/2112.03482