# Copyright 2018 Google LLC # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """Wrapper scripts to ensure that main.py commands are called correctly.""" import argh import argparse import logging import os import main import shipname import sys import time import tempfile from utils import timer from absl import flags import cloud_logging from tensorflow import gfile # Pull in environment variables. Run `source ./cluster/common` to set these. BUCKET_NAME = os.environ['BUCKET_NAME'] BOARD_SIZE = os.environ['BOARD_SIZE'] BASE_DIR = "gs://{}".format(BUCKET_NAME) MODELS_DIR = os.path.join(BASE_DIR, 'models') SELFPLAY_DIR = os.path.join(BASE_DIR, 'data/selfplay') HOLDOUT_DIR = os.path.join(BASE_DIR, 'data/holdout') SGF_DIR = os.path.join(BASE_DIR, 'sgf') TRAINING_CHUNK_DIR = os.path.join(BASE_DIR, 'data', 'training_chunks') GOLDEN_CHUNK_DIR = os.path.join(BASE_DIR, 'data', 'golden_chunks') ESTIMATOR_WORKING_DIR = 'estimator_working_dir' # How many games before the selfplay workers will stop trying to play more. MAX_GAMES_PER_GENERATION = 10000 # How many games minimum, until the trainer will train MIN_GAMES_PER_GENERATION = 500 # What percent of games to holdout from training per generation HOLDOUT_PCT = 0.05 def print_flags(): flags = { 'BUCKET_NAME': BUCKET_NAME, 'BASE_DIR': BASE_DIR, 'MODELS_DIR': MODELS_DIR, 'SELFPLAY_DIR': SELFPLAY_DIR, 'HOLDOUT_DIR': HOLDOUT_DIR, 'SGF_DIR': SGF_DIR, 'TRAINING_CHUNK_DIR': TRAINING_CHUNK_DIR, 'ESTIMATOR_WORKING_DIR': ESTIMATOR_WORKING_DIR, 'BOARD_SIZE': BOARD_SIZE, } print("Computed variables are:") print('\n'.join('--{}={}'.format(flag, value) for flag, value in flags.items())) def get_models(): """Finds all models, returning a list of model number and names sorted increasing. Returns: [(13, 000013-modelname), (17, 000017-modelname), ...etc] """ all_models = gfile.Glob(os.path.join(MODELS_DIR, '*.meta')) model_filenames = [os.path.basename(m) for m in all_models] model_numbers_names = sorted([ (shipname.detect_model_num(m), shipname.detect_model_name(m)) for m in model_filenames]) return model_numbers_names def get_latest_model(): """Finds the latest model, returning its model number and name Returns: (17, 000017-modelname) """ return get_models()[-1] def get_model(model_num): models = {k: v for k, v in get_models()} if not model_num in models: raise ValueError("Model {} not found!".format(model_num)) return models[model_num] def game_counts(n_back=20): """Prints statistics for the most recent n_back models""" all_models = gfile.Glob(os.path.join(MODELS_DIR, '*.meta')) model_filenames = sorted([os.path.basename(m).split('.')[0] for m in all_models], reverse=True) for m in model_filenames[:n_back]: games = gfile.Glob(os.path.join(SELFPLAY_DIR, m, '*.zz')) print(m, len(games)) def bootstrap(): bootstrap_name = shipname.generate(0) bootstrap_model_path = os.path.join(MODELS_DIR, bootstrap_name) print("Bootstrapping with working dir {}\n Model 0 exported to {}".format( ESTIMATOR_WORKING_DIR, bootstrap_model_path)) main.bootstrap(ESTIMATOR_WORKING_DIR, bootstrap_model_path) def selfplay(readouts=1600, verbose=2): _, model_name = get_latest_model() games = gfile.Glob(os.path.join(SELFPLAY_DIR, model_name, '*.zz')) if len(games) > MAX_GAMES_PER_GENERATION: print("{} has enough games ({})".format(model_name, len(games))) time.sleep(10*60) sys.exit(1) print("Playing a game with model {}".format(model_name)) model_save_path = os.path.join(MODELS_DIR, model_name) game_output_dir = os.path.join(SELFPLAY_DIR, model_name) game_holdout_dir = os.path.join(HOLDOUT_DIR, model_name) sgf_dir = os.path.join(SGF_DIR, model_name) main.selfplay( load_file=model_save_path, output_dir=game_output_dir, holdout_dir=game_holdout_dir, output_sgf=sgf_dir, readouts=readouts, holdout_pct=HOLDOUT_PCT, verbose=verbose, ) def gather(): print("Gathering game output...") main.gather(input_directory=SELFPLAY_DIR, output_directory=TRAINING_CHUNK_DIR) def train(load_dir=MODELS_DIR, save_dir=MODELS_DIR): model_num, model_name = get_latest_model() games = gfile.Glob(os.path.join(SELFPLAY_DIR, model_name, '*.zz')) if len(games) < MIN_GAMES_PER_GENERATION: print("{} doesn't have enough games to train a new model yet ({})".format( model_name, len(games))) print("Sleeping...") time.sleep(10*60) print("Done...") sys.exit(1) print("Training on gathered game data, initializing from {}".format(model_name)) new_model_num = model_num + 1 new_model_name = shipname.generate(new_model_num) print("New model will be {}".format(new_model_name)) training_file = os.path.join( GOLDEN_CHUNK_DIR, str(new_model_num) + '.tfrecord.zz') while not gfile.Exists(training_file): print("Waiting for", training_file) time.sleep(1*60) print("Using Golden File:", training_file) load_file = os.path.join(load_dir, model_name) save_file = os.path.join(save_dir, new_model_name) try: main.train(ESTIMATOR_WORKING_DIR, [training_file], save_file, generation_num=model_num + 1) except: logging.exception("Train error") def validate(model_num=None, validate_name=None): """ Runs validate on the directories up to the most recent model, or up to (but not including) the model specified by `model_num` """ if model_num is None: model_num, model_name = get_latest_model() else: model_num = int(model_num) model_name = get_model(model_num) # Model N was trained on games up through model N-2, so the validation set # should only be for models through N-2 as well, thus the (model_num - 1) # term. models = list( filter(lambda num_name: num_name[0] < (model_num - 1), get_models())) # Run on the most recent 50 generations, # TODO(brianklee): make this hyperparameter dependency explicit/not hardcoded holdout_dirs = [os.path.join(HOLDOUT_DIR, pair[1]) for pair in models[-50:]] main.validate(ESTIMATOR_WORKING_DIR, *holdout_dirs, checkpoint_name=os.path.join(MODELS_DIR, model_name), validate_name=validate_name) def backfill(): models = [m[1] for m in get_models()] import dual_net import tensorflow as tf from tqdm import tqdm from tensorflow.python.framework import meta_graph features, labels = dual_net.get_inference_input() dual_net.model_fn(features, labels, tf.estimator.ModeKeys.PREDICT, dual_net.get_default_hyperparams()) for model_name in tqdm(models): if model_name.endswith('-upgrade'): continue try: load_file = os.path.join(MODELS_DIR, model_name) dest_file = os.path.join(MODELS_DIR, model_name) main.convert(load_file, dest_file) except: print('failed on', model_name) continue def echo(): pass # Flags are echo'd in the ifmain block below. parser = argparse.ArgumentParser() argh.add_commands(parser, [train, selfplay, gather, echo, backfill, bootstrap, game_counts, validate]) if __name__ == '__main__': print_flags() cloud_logging.configure() remaining_argv = flags.FLAGS(sys.argv, known_only=True) argh.dispatch(parser, argv=remaining_argv[1:])