def load_policy_agent(h5file):
model = kerasutil.load_model_from_hdf5_group(h5file['model'])
encoder_name = h5file['encoder'].attrs['name']
board_width = h5file['encoder'].attrs['board_width']
board_height = h5file['encoder'].attrs['board_height']
encoder = encoders.get_encoder_by_name(encoder_name,
(board_width, board_height))
return PolicyAgent(model, encoder)
def load_policy_agent(h5file_dir):
h5file = h5py.File(h5file_dir, 'r')
model = kerasutil.load_model_from_hdf5_group(h5file['model'])
encoder_name = h5file['encoder'].attrs['name']
board_width = h5file['encoder'].attrs['board_width']
board_height = h5file['encoder'].attrs['board_height']
encoder = get_encoder_by_name(encoder_name, board_height)
return SeegaAgent(model, encoder)
def load(cls, filename):
inp = h5py.File(filename, 'r')
model = kerasutil.load_model_from_hdf5_group(inp['model'])
training_run = cls(filename, model,
inp['metadata'].attrs['epochs_completed'],
inp['metadata'].attrs['chunks_completed'],
inp['metadata'].attrs['num_chunks'])
inp.close()
return training_run
def load_prediction_agent(h5file):
model = kerasutil.load_model_from_hdf5_group(h5file['model'])
encoder_name = h5file['encoder'].attrs['name']
if not isinstance(encoder_name, str):
encoder_name = encoder_name.decode('ascii')
board_width = h5file['encoder'].attrs['board_width']
board_height = h5file['encoder'].attrs['board_height']
encoder = encoders.get_encoder_by_name(encoder_name, (board_width, board_height))
return DeepLearningAgent(model, encoder)
def load_policy_agent(h5file):
model = kerasutil.load_model_from_hdf5_group(
h5file['model'],
custom_objects={'policy_gradient_loss': policy_gradient_loss})
encoder_name = h5file['encoder'].attrs['name']
if not isinstance(encoder_name, str):
encoder_name = encoder_name.decode('ascii')
board_width = h5file['encoder'].attrs['board_width']
board_height = h5file['encoder'].attrs['board_height']
encoder = ElevenPlaneEncoder((board_width, board_height))
return PolicyAgent(model, encoder)
def load_policy_agent(h5file):
model = kerasutil.load_model_from_hdf5_group(
h5file['model']
) # Uses built in Keras functions to load the model structure and weights
encoder_name = h5file['encoder'].attrs[
'name'] # Recovers the board encoder
board_width = h5file['encoder'].attrs['board_width']
board_height = h5file['encoder'].attrs['board_height']
if type(encoder_name) == bytes:
encoder_name = encoder_name.decode()
encoder = encoders.get_encoder_by_name(encoder_name,
(board_width, board_height))
return PolicyAgent(model, encoder) # Reconstructs the agent
def load_zero_agent(h5file):
model = kerasutil.load_model_from_hdf5_group(
h5file['model']
) # Uses built in Keras functions to load the model structure and weights
encoder_name = h5file['encoder'].attrs[
'name'] # Recovers the board encoder
board_size = h5file['encoder'].attrs['board_size']
num_rounds = h5file['meta'].attrs['num_rounds']
c = h5file['meta'].attrs['c']
# zero encoder isn't in dlgo.encoders ... should it be?
encoder = ZeroEncoder(board_size)
return ZeroAgent(model, encoder, num_rounds, c)
def main():
pth = '//home//nail//Code_Go//checkpoints//'
pth_experience = '//home//nail//Experience//'
experience = []
os.chdir(pth_experience)
lst_files = os.listdir(pth_experience)
pattern = input('Паттерн для выборки файлов для обучения: ')
if len(pattern) == 0:
pattern = "exp*.h5"
for entry in lst_files:
if fnmatch.fnmatch(entry, pattern):
experience.append(entry)
experience.sort()
learning_agent = input('learning_agent:')
learning_agent = pth + learning_agent + '.h5'
print('learning_agent: ', learning_agent)
agent_out = input('agent_out:')
agent_out = pth + agent_out + '.h5'
board_size = 19
print('agent_out: ', agent_out)
try:
lr = float(input('lr = '))
except:
lr = 0.000001
try:
bs = int(input('bs = '))
except:
bs = 1024
# ==================================================
import tensorflow as tf
config = tf.compat.v1.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.98
config.gpu_options.allow_growth = True
config.log_device_placement = True
sess = tf.compat.v1.Session(config=config)
tf.compat.v1.keras.backend.set_session(sess)
# ==================================================
encoder = SimpleEncoder((board_size, board_size))
try:
h5file = h5py.File(learning_agent, "r")
learning_agent = rl.load_value_agent(h5file)
model_v = kerasutil.load_model_from_hdf5_group(h5file['model'])
except:
learning_agent = create_v_model(lr=lr)
i = 1
num_files = len(experience)
for exp_filename in experience:
print(50 * '=')
print('Файл для обучения: %s...' % exp_filename)
print(50 * '=')
exp_buffer = rl.load_experience(h5py.File(exp_filename, "r"))
model_v = my_train_v(model_v,
encoder,
exp_buffer,
lr=lr,
batch_size=bs)
print('Обработано файлов: ', i, ' из ', num_files)
i += 1
learning_agent = rl.ValueAgent(model_v, encoder)
with h5py.File(agent_out, 'w') as updated_agent_outf:
learning_agent.serialize(updated_agent_outf)
