[pytorch]——用flatten和index_select实现embedding look_up

代码

batch_size, n_features = w.shape
emb_out = self.embeddings.index_select(0, w.flatten())
x = emb_out.view(batch_size, -1) # -1 means: n_features * embed_size

也就是

x = self.embeddings(w)