def __init__(self, embedding_file, graph_path, params={}):
Evaluation.__init__(self)
self._embedding_file = embedding_file
self._graph_path = graph_path
self._directed = params['directed'] if 'directed' in params else False
self.results = None
def __init__(self, config):
BaseAgent.__init__(self, config)
Evaluation.__init__(self, self.config.metrics_k)
self.metrics_k = config.metrics_k
self.state_config = self.config.hyperparameters["State"]
self.embedding = self.get_embedding().to(self.device)
self.embedding_dim = self.embedding.embedding_dim
# Initialize state-module
self.state_agg = RNNStateAgg(self.embedding,
state_config=self.state_config,
reward_range=[0, 1],
with_rewards=False).to(self.device)
self.state_size = self.state_agg.state_size
self.state_optimizer = self.create_state_optimizer()
if self.config.hyperparameters[
"state-only-pretrain"] or self.config.hyperparameters[
"pretrain"]:
save_dir = Path(config.file_to_save_model).parent / "pretrain"
self.pretrain_model_saver = ModelSaver(save_dir)
if self.config.hyperparameters["state-only-pretrain"]:
self.output_layer = torch.nn.Linear(
self.state_size,
self.environment.action_space.n).to(self.device)
self.output_layer_optimizer = torch.optim.Adam(
self.output_layer.parameters())
self.user_history_mask_items = None
self.masking_enabled = self.hyperparameters.get("history_masking")
self.model_saver = ModelSaver(Path(config.file_to_save_model).parent)
self.exploration_strategy = Epsilon_Greedy_Exploration(self.config)
self.last_done = np.zeros(self.environment.num_envs, dtype=np.bool)
def __init__(self, train_env, k):
Evaluation.__init__(self, k)
self.train_env = train_env