Leela Chess Zero supervised learning easy start for Windows users

I assume one have created "chunks" for training from pgn games collection (as described by Dietrich Kappe on his Wiki page) and those "chunks" are placed in some separate directory.

Step 0. Go to lczero-training GitHub page and download lczero-training-master.zip. From lczero-common page download lczero-common-master.zip file. Unzip both files preserving directory structure. Compile chunk.proto and net.proto files with protobuf compiler protoc - see init.sh for details. As result uoy get two .py files in subdirectory tf\proto. Create empty (zero-length) init.py file in tf\proto directory.

Step 1. Download latest version Miniconda 64-bit Windows installer for Python 3.6 from Miniconda site. Install Miniconda following installation instructions. Open Anaconda Prompt and type command: python -V and expect answer: Python 3.6.5

Don't upgrade Python to version 3.7 because there isn't yet TensorFlow for this version!

Step 2. Install TensorFlow with GPU support. Open Anaconda Prompt and run command: conda install tensorflow-gpu TensorFlow will be downloaded and installed with all required additional packages - in particular CUDA and cuDNN dll's. At end install PyYAML module: conda install pyyaml because Miniconda installed other yaml by default.

Step 3. Create and edit your own .yaml configuration file as described on Wiki page. Run: python train.py ... with your parameters and enjoy supervised Leela Chess learning :-)