def load_glove_embedding(self, freeze=False):
initial_arr = self.m.u_encoder.embedding.weight.data.cpu().numpy()
embedding_arr = torch.from_numpy(get_glove_matrix(self.reader.vocab, initial_arr))
self.m.u_encoder.embedding.weight.data.copy_(embedding_arr)
self.m.z_decoder.emb.weight.data.copy_(embedding_arr)
self.m.m_decoder.emb.weight.data.copy_(embedding_arr)
def load_glove_embedding(self, freeze=False):
initial_arr = self.sedst.u_encoder.embedding.weight.data.cpu().numpy()
embedding_arr = torch.from_numpy(get_glove_matrix(self.reader.vocab, initial_arr))
self.sedst.u_encoder.embedding.weight.data.copy_(embedding_arr)
self.sedst.m_encoder.embedding.weight.data.copy_(embedding_arr)
self.sedst.m_decoder.emb.weight.data.copy_(embedding_arr)
self.sedst.qz_decoder.mu.weight.data.copy_(embedding_arr.transpose(1,0))
self.sedst.pz_decoder.mu.weight.data.copy_(embedding_arr.transpose(1,0))
if freeze:
self.freeze_module(self.sedst.u_encoder.embedding)
self.freeze_module(self.sedst.m_encoder.embedding)
self.freeze_module(self.sedst.m_decoder.emb)
def load_glove_embedding(self, freeze=False):
initial_arr = self.m.u_encoder.embedding.weight.data.cpu().numpy()
mat = get_glove_matrix(self.reader.vocab, initial_arr)
# np.save('./data/embedding.npy',mat)
# mat = np.load('./data/embedding.npy')
embedding_arr = torch.from_numpy(mat)
self.m.u_encoder.embedding.weight.data.copy_(embedding_arr)
self.m.p_encoder.embedding.weight.data.copy_(embedding_arr)
self.m.m_decoder.emb.weight.data.copy_(embedding_arr)
self.m.p_decoder.emb.weight.data.copy_(embedding_arr)
self.m.qz_decoder.mu.weight.data.copy_(embedding_arr.transpose(1, 0))
self.m.pz_decoder.mu.weight.data.copy_(embedding_arr.transpose(1, 0))
if freeze:
self.freeze_module(self.m.u_encoder.embedding)
self.freeze_module(self.m.m_e.embedding)
self.freeze_module(self.m.m_decoder.emb)
def load_glove_embedding(self):
initial_arr = self.m.u_encoder.embedding.weight.data.cpu().numpy()
embedding_arr = torch.from_numpy(
get_glove_matrix(self.reader.vocab, initial_arr))
self.m.u_encoder.embedding.weight.data.copy_(embedding_arr)
self.m.u_encoder.embedding.weight.requires_grad = cfg.emb_trainable
if cfg.separate_enc:
self.m.z_encoder.embedding.weight.data.copy_(embedding_arr)
self.m.z_encoder.embedding.weight.requires_grad = cfg.emb_trainable
for i in range(cfg.num_head):
self.m.z_decoders[i].emb.weight.data.copy_(embedding_arr)
self.m.z_decoders[i].emb.weight.requires_grad = cfg.emb_trainable
self.m.req_classifiers.emb.weight.data.copy_(embedding_arr)
self.m.req_classifiers.emb.weight.requires_grad = cfg.emb_trainable
self.m.res_classifiers.emb.weight.data.copy_(embedding_arr)
self.m.res_classifiers.emb.weight.requires_grad = cfg.emb_trainable
self.m.m_decoder.emb.weight.data.copy_(embedding_arr)
self.m.m_decoder.emb.weight.requires_grad = cfg.emb_trainable
def init_embedding(embedding, r):
initial_arr = embedding.weight.data.cpu().numpy()
embedding_arr = torch.from_numpy(reader.get_glove_matrix(r.vocab, initial_arr))
embedding.weight.data.copy_(embedding_arr)
return embedding
def init_embedding_model(model, r):
""" Set glove embeddings for model, r is a reader instance """
initial_arr = model.embedding.weight.data.cpu().numpy()
embedding_arr = torch.from_numpy(reader.get_glove_matrix(r.vocab, initial_arr))
model.embedding.weight.data.copy_(embedding_arr)