def train(algo_name,
params,
dataset_path,
model_save_path,
experiment_name=None,
with_timestamp=True,
logdir='d3rlpy_logs',
prev_model_path=None,
test_size=0.2):
dataset = MDPDataset.load(dataset_path)
train_data, test_data = train_test_split(dataset, test_size=test_size)
# train
algo = create_algo(algo_name, dataset.is_action_discrete(), **params)
algo.fit(train_data,
experiment_name=experiment_name,
with_timestamp=with_timestamp,
logdir=logdir,
save_interval=1000000) # never save models for now
# save final model
algo.save_model(model_save_path)
# evaluate
scores = _evaluate(algo, test_data, dataset.is_action_discrete())
# compare previous model
if prev_model_path:
base_algo = create_algo(algo_name, **params)
base_algo.load_model(prev_model_path)
score = _comapre(algo, base_algo, test_data,
dataset.is_action_discrete())
scores['algo_action_diff'] = score
return scores
def get_model(self):
from flask import jsonify, send_file
# initialize algorithm
algo = create_algo(self.algo_name, self.dataset.is_action_discrete(),
**self.algo_params)
algo.create_impl(self.dataset.get_observation_shape(),
self.dataset.get_action_size())
# load latest model
trial = self.n_trials - 1
while trial >= 0:
model_path = self.model_save_path_tmpl % trial
if os.path.exists(model_path):
algo.load_model(model_path)
break
trial -= 1
if trial < 0:
# return error
from flask import jsonify
return jsonify({'status': 'empty'}), 500
# save policy
policy_path = os.path.join(self.dir_path, 'policy.pt')
algo.save_policy(policy_path)
# send back policy data
res = send_file(policy_path,
as_attachment=True,
attachment_filename='policy.pt')
return res
def train(algo_name,
params,
dataset_path,
experiment_name=None,
logdir='d3rlpy_logs'):
# prepare dataset
dataset = MDPDataset.load(dataset_path)
train_data, test_data = train_test_split(dataset, test_size=0.2)
# evaluate
scorers = _get_scorers(dataset.is_action_discrete())
# train
algo = create_algo(algo_name, dataset.is_action_discrete(), **params)
algo.fit(train_data,
eval_episodes=test_data,
scorers=scorers,
experiment_name=experiment_name,
with_timestamp=False,
logdir=logdir,
save_interval=1,
show_progress=False,
tensorboard=False)
return True
def train(algo_name,
params,
dataset_path,
experiment_name=None,
logdir="d3rlpy_logs"):
# prepare dataset
dataset = MDPDataset.load(dataset_path)
train_data, test_data = train_test_split(dataset, test_size=0.2)
# get dataset statistics
stats = dataset.compute_stats()
# evaluate
scorers = _get_scorers(dataset.is_action_discrete(), stats)
# add action scaler if continuous action-space
if not dataset.is_action_discrete():
params["action_scaler"] = "min_max"
# train
algo = create_algo(algo_name, dataset.is_action_discrete(), **params)
algo.fit(
train_data,
n_steps=params["n_epochs"] * params["n_steps_per_epoch"],
n_steps_per_epoch=params["n_steps_per_epoch"],
eval_episodes=test_data,
scorers=scorers,
experiment_name=experiment_name,
with_timestamp=False,
logdir=logdir,
save_interval=1,
show_progress=False,
)
return True
def __init__(self,
algo_name,
algo_params,
dataset,
dir_path='d3rlpy_logs/worker',
with_timestamp=True,
model_path=None):
from flask import Flask
self.algo_name = algo_name
self.algo_params = algo_params
self.dataset = dataset
self.n_trials = 0
# setup flask server
self.app = Flask(__name__)
self.app.add_url_rule('/train',
'train',
self.train_algo,
methods=['POST'])
self.app.add_url_rule('/data',
'data',
self.append_data,
methods=['POST'])
self.app.add_url_rule('/model',
'model',
self.get_model,
methods=['GET'])
self.app.add_url_rule('/status',
'status',
self.get_status,
methods=['GET'])
# prepare directory
if with_timestamp:
dir_path += '_' + datetime.now().strftime('%Y%m%d%H%M%S')
self.dir_path = os.path.abspath(dir_path)
os.makedirs(dir_path)
# save initial dataset
self.dataset_path = os.path.join(self.dir_path, 'dataset.h5')
dataset.dump(self.dataset_path)
self.model_save_path_tmpl = os.path.join(self.dir_path, 'model_%d.pt')
self.experiment_name_tmpl = 'worker_training_%d'
self.train_uid = None
self.latest_metrics = {}
# make initial model
if model_path:
algo = create_algo(algo_name, **algo_params)
algo.create_impl(dataset.get_observation_shape(),
dataset.get_action_size())
algo.load_model(model_path)
algo.save_model(self.model_save_path_tmpl % 0)
self.n_trials += 1
else:
self._dispatch_training_job()