def discovery(embedding, vocab, chord_a, chord_b, chord_c, num_output=10):
a_id = vocab.get_token_index(chord_a)
b_id = vocab.get_token_index(chord_b)
c_id = vocab.get_token_index(chord_c)
vec_a = embedding.weight[a_id]
vec_b = embedding.weight[b_id]
vec_c = embedding.weight[c_id]
cosine = CosineSimilarity(dim=0)
sims = Counter()
vec = vec_b - vec_a + vec_c
for index, token in vocab.get_index_to_token_vocabulary().items():
sim = cosine(vec, embedding.weight[index]).item()
sims[token] = sim
return sims.most_common(num_output)
vocab = Vocabulary().from_files("data/vocabulary")
EMBEDDING_DIM = 128
token_embedding = Embedding(num_embeddings=vocab.get_vocab_size("tokens"),
embedding_dim=EMBEDDING_DIM)
token_embedding.load_state_dict(
torch.load("saved_models/word2vec.th", map_location="cpu"))
print(discovery(token_embedding, vocab, "C", "G", "G"))
except Exception as e:
print(e)
print(label)
marker_list = [".", ".", ".", "v", "p", "D", "x", "*", "s"]
return color_list[key_number], marker_list[form_index]
vocab = Vocabulary().from_files("data/vocabulary")
EMBEDDING_DIM = 128
token_embedding = Embedding(num_embeddings=vocab.get_vocab_size("tokens"),
embedding_dim=EMBEDDING_DIM)
token_embedding.load_state_dict(torch.load("saved_models/word2vec.th"))
token_ids = torch.tensor([x for x in range(2, vocab.get_vocab_size())],
dtype=torch.long)
if torch.cuda.is_available():
cuda_device = 0
token_embedding = token_embedding.cuda(cuda_device)
token_ids = token_ids.cuda(cuda_device)
else:
cuda_device = -1
token_embedding.eval()
with torch.no_grad():
embeddings = token_embedding(token_ids).cpu().numpy()
