def build_models(self):
'''
Build all models.
'''
self.pretrained_models['bert'] = BertModel.from_pretrained(
'bert-base-uncased',
output_hidden_states=True,
output_attentions=True).to(self.args.device)
self.train_models['attn_concept'] = Attention_Concepts(
input_size=768,
n_concepts=self.args.n_concepts,
device=self.args.device).to(self.args.device)
self.train_models['attn_self'] = AttentionSelf(
input_size=768, hidden_size=768,
device=self.args.device).to(self.args.device)
self.train_models['ff'] = torch.nn.Linear(768,
768).to(self.args.device)
self.train_models['classifier'] = torch.nn.Linear(
768, self.args.n_class).to(self.args.device)
self.train_models['drop'] = torch.nn.Dropout(self.args.drop_rate)
self.loss_criterion = torch.nn.CrossEntropyLoss(ignore_index=-1).to(
self.args.device)
def build_models(self):
'''
Build all models.
'''
self.base_models['embedding'] = torch.nn.Embedding(
self.batch_data['vocab_size'],
self.args.emb_dim).to(self.args.device)
self.train_models['encoder'] = EncoderRNN(
emb_dim=self.args.emb_dim,
hidden_size=self.args.rnn_hidden_dim,
nLayers=self.args.rnn_nLayers,
rnn_network=self.args.rnn_network,
device=self.args.device).to(self.args.device)
self.train_models['attn_self'] = AttentionSelf(
input_size=self.args.rnn_hidden_dim * 2,
hidden_size=self.args.rnn_hidden_dim * 2,
device=self.args.device).to(self.args.device)
self.train_models['ff'] = torch.nn.Linear(self.args.rnn_hidden_dim * 2,
self.args.rnn_hidden_dim *
2).to(self.args.device)
self.train_models['classifier'] = torch.nn.Linear(
self.args.rnn_hidden_dim * 2,
self.args.n_class).to(self.args.device)
self.train_models['drop'] = torch.nn.Dropout(self.args.drop_rate)
self.loss_criterion = torch.nn.CrossEntropyLoss(ignore_index=-1).to(
self.args.device)