The current self-play and training pipeline is run using a C++ implementation of Minigo.

The C++ Minigo port requires version 0.19.2 of the Bazel build system (see README.md for installation instructions).

Minigo++ depends on the TensorFlow C++ libraries, but we have not yet set up Bazel WORKSPACE and BUILD rules to automatically download and configure TensorFlow so (for now at least) you must perform a manual step to build the library. This depends on zip, so be sure that package is installed first.

If you want to run on GPU, you will have to tell Bazel where to find your CUDA install (note that recent versions of Bazel ignore LD_LIBRARY_PATH):

sudo sh -c "echo /usr/local/cuda/lib64 > /etc/ld.so.conf.d/cuda.conf"
sudo ldconfig

Now, you're ready to compile TensorFlow from source:

sudo apt-get install zip
./cc/configure_tensorflow.sh

If you want to compile for CPU and not GPU, then execute the following instead:

sudo apt-get install zip
TF_NEED_CUDA=0 ./cc/configure_tensorflow.sh

This will automatically perform the first steps of Installing TensorFlow from Sources but instead of installing the TensorFlow package, it extracts the generated C++ headers into the cc/tensorflow subdirectory of the repo. The script then builds the required TensorFlow shared libraries and copies them to the same directory. The tensorflow cc_library build target in cc/BUILD pulls these header and library files together into a format the Bazel understands how to link against.

Although Minigo uses TensorFlow as its inference engine by default, other engines can be used (e.g. Cloud TPU, TensorRT, TensorFlow Lite). See the Inferences engines section below for more details.

The TensorFlow models can found on our public Google Cloud Storage bucket. A good model to start with is 000990-cormorant from the v15 run. The C++ engine requires the frozen model in a GraphDef proto format, which is the .pb file. You can copy it locally as follows:

mkdir -p saved_models
gsutil cp gs://minigo-pub/v15-19x19/models/000990-cormorant.pb saved_models/

Minigo can also read models directly from Google Cloud Storage but it doesn't currently perform any local caching, so you're better off copying the model locally once instead of copying from GCS every time.

C++ Minigo is made up of several different binaries. All binaries can be run with --helpshort, which will display the full list of command line arguments.

A very simple example of how to perform self-play using the Minigo engine. Plays a single game using a fixed number of readouts.

bazel build -c opt cc:simple_example
bazel-bin/cc/simple_example \
  --model=tf,saved_models/000990-cormorant.pb \
  --num_readouts=160

The self-play binary used in our training pipeline. Has a lot more functionality that the simple example, including:

• Play multiple games in parallel, batching their inferences together for better GPU/TPU utilization.
• Automatic loading of the latest trained model.
• Write SGF games & TensorFlow training examples to Cloud Storage or Cloud BigTable.
• Flag file support with automatic reloading. This is used among other things to dynamically adjust the resign threshold to minimize the number of bad resigns.

bazel build -c opt cc:selfplay
bazel-bin/cc/selfplay \
  --model=saved_models/000990-cormorant.pb \
  --num_readouts=160 \
  --parallel_games=1 \
  --output_dir=data/selfplay \
  --holdout_dir=data/holdout \
  --sgf_dir=sgf

Evaluates the performance of two models by playing them against each other over multiple games in parallel.

bazel build -c opt cc:eval
bazel-bin/cc/eval \
  --model=saved_models/000990-cormorant.pb \
  --model_two=saved_models/000990-cormorant.pb \
  --num_readouts=160 \
  --parallel_games=32 \
  --sgf_dir=sgf

Play using the GTP protocol. This is also the binary we recommend using as a backend for Minigui (see minigui/README.md).

bazel build -c opt cc:gtp
bazel-bin/cc/gtp \
  --model=saved_models/000990-cormorant.pb \
  --num_readouts=160

Loads all SGF files found in the given sgf_dir and tries to predict the move made at each position in each game. After all games are processed, prints summary stats to about how many moves were correctly predicted.

bazel build -c opt cc:puzzle
bazel-bin/cc/puzzle \
  --model=saved_models/000990-cormorant.pb \
  --num_readouts=160 \
  --sgf_dir=puzzle_sgf

Minigo's C++ unit tests operate on both 9x9 and 19x19, and some tests are only enabled for a particular board size. Consequently, you must run the tests twice: once for 9x9 boards and once for 19x19 boards.

bazel test --define=board_size=9 cc/...  &&  bazel test cc/...

Note that Minigo is compiled for a 19x19 board by default, which explains the lack of a --define=board_size=19 in the second bazel test invocation.

Bazel supports building with AddressSanitizer to check for C++ memory errors:

bazel build cc:selfplay \
  --copt=-fsanitize=address \
  --linkopt=-fsanitize=address \
  --copt=-fno-omit-frame-pointer \
  --copt=-O1

C++ Minigo currently supports multiple separate engines for performing inference. Which engines are compiled into the C++ binaries are controlled by passing Bazel --define arguments at compile time. The inference engine to use is specified as part of the --model or --model_two command line arguments:

• tf: peforms inference using the TensorFlow libraries built by cc/configure_tensorflow.sh. Compiled & used as the inference by default, disable the engine with --define=tf=0. Use by passing --model=tf,$PATH, where $PATH is the path to a frozen TensorFlow GraphDef proto (as generated by freeze_graph.py).
• lite: performs inference using TensorFlow Lite, which runs in software on the CPU. Compile by passing --define=lite=1 to bazel build. Use by passing --model=lite,$PATH, where $PATH is the path to a Toco- optimized TFLite flat buffer (see below).
• tpu: perform inference on a Cloud TPU. Your code must run on a Cloud TPU-equipped VM for this to work. Compile by passing --define=tpu=1 to bazel build. Use by passed --model=tpu:$TPU_NAME,$PATH, where $TPU_NAME is the name of a Cloud TPU (e.g. grpc://10.240.2.10:8470) and $PATH is the path to a frozen TensorFlow GraphDef proto (as generated by freeze_graph.py).
• trt: uses NVIDIA TensorRT. Compile by passing --define=trt=1 to bazel build. Use by passing --model=trt,$PATH, where $PATH is a TensorRT model.
• fake: a simple fake that is only useful in so much as you can use it to test that the tree search code compiles without also having to compile a full inference engine. Enabled by default. There's no way to disable the fake inference engine because this guaratees that there's always an engine available (even if it's a useless one). Use by passing --model=fake.
• random: a model that returns random samples from a normal distribution, which can be useful for bootstrapping the reinforcement learning pipeline. Use by passing --model=random:$SEED,$POLICY_STD_DEV:$VALUE_STD_DEV, where $SEED is a random seed (set to 0 to choose one based on the current time), $POLICY_STD_DEV is the standard deviation of the distribution of policy samples (0.4 is a reasonable choice) and $VALUE_STD_DEV is the standard deviation for the distribution of value samples (again, 0.4 is a reasonable choice). That was a bit of a long-winded explanation, so just try --model=random:0,0.4:0.4 to start with.

First, run a frozen graph through Toco, the TensorFlow optimizing compiler:

BATCH_SIZE=8
./cc/tensorflow/toco \
  --input_file=saved_models/000256-opossum.pb \
  --input_format=TENSORFLOW_GRAPHDEF \
  --output_format=TFLITE \
  --output_file=saved_models/000256-opossum.tflite \
  --inference_type=FLOAT \
  --input_type=FLOAT \
  --input_arrays=pos_tensor \
  --output_arrays=policy_output,value_output \
  --input_shapes=8,19,19,17

Minigo supports running inference on Cloud TPU. Build //cc:selfplay with --define=tpu=1 and run with --model=tpu,$PATH (see above).

To freeze a model into a GraphDef proto that can be run on Cloud TPU, use freeze_graph.py:

python freeze_graph.py \
  --model_path=$MODEL_PATH \
  --use_tpu=true \
  --tpu_name=$TPU_NAME \
  --num_tpu_cores=8

Where $MODEL_PATH is the path to your model (either a local file or one on Google Cloud Storage), and $TPU_NAME is the gRPC name of your TPU, e.g. grpc://10.240.2.10:8470. This can be found from the output of gcloud beta compute tpus list.

This command must be run from a Cloud TPU-ready GCE VM.

This invocation to freeze_graph.py will replicate the model 8 times so that it can run on all eight cores of a Cloud TPU. To take advantage of this parallelism when running selfplay, virtual_losses * parallel_games must be at least 8, ideally 128 or higher.

The C++ Minigo implementation requires that the board size be defined at compile time, using the MINIGO_BOARD_SIZE preprocessor define. This allows us to significantly reduce the number of heap allocations performed. The build scripts are configured to compile with MINIGO_BOARD_SIZE=19 by default. To compile a version that works with a 9x9 board, invoke Bazel with --define=board_size=9.

Minigo supports writing eval and selfplay results to Bigtable.

Build with --define=bt=1 and run with --output_bigtable=<PROJECT>,<INSTANCE>,<TABLE>.

For eval this would look something like

bazel-bin/cc/eval \
  --model=<MODEL_1> \
  --model_two=<MODEL_2> \
  --parallel_games=4 \
  --num_readouts=32
  --sgf_dir=sgf \
  --output_bigtable=<PROJECT>,minigo-instance,games

See number of eval games with

cbt -project=<PROJECT> -instance=minigo-instance \
  read games columns="metadata:eval_game_counter"

See eval results with

cbt -project=<PROJECT> -instance=minigo-instance \
  read games prefix="e_"

The C++ code follows Google's C++ style guide and we use cpplint to delint.