def test_beam(self):
src_voc = Vocabulary()
trg_voc = Vocabulary()
for tok in "</s> I am Philip You are a".split():
src_voc[tok]
for tok in "</s> 私 は フィリップ です 1 2 3".split():
trg_voc[tok]
model = EncDecNMT(Args("attn"), src_voc, trg_voc, optimizer=optimizers.SGD())
model_out = "/tmp/model-nmt.temp"
X, Y = src_voc, trg_voc
# Train with 1 example
src = np.array([[X["I"], X["am"], X["Philip"]]], dtype=np.int32)
trg = np.array([[Y["私"], Y["は"], Y["フィリップ"], Y["です"]]], dtype=np.int32)
model.train(src, trg)
# Save
serializer = ModelSerializer(model_out)
serializer.save(model)
# Load
model1 = EncDecNMT(InitArgs(model_out))
k = model.classify(src, beam=10)
def test_NMT_2_read_write(self):
for model in ["encdec", "attn"]:
src_voc = Vocabulary()
trg_voc = Vocabulary()
for tok in "</s> I am Philip".split():
src_voc[tok]
for tok in "</s> 私 は フィリップ です".split():
trg_voc[tok]
model = EncDecNMT(Args(model), src_voc, trg_voc, optimizer=optimizers.SGD())
model_out = "/tmp/nmt/tmp"
X, Y = src_voc, trg_voc
# Train with 1 example
src = np.array([[X["I"], X["am"], X["Philip"]]], dtype=np.int32)
trg = np.array([[Y["私"], Y["は"], Y["フィリップ"], Y["です"]]], dtype=np.int32)
model.train(src, trg)
# Save
serializer = ModelSerializer(model_out)
serializer.save(model)
# Load
model1 = EncDecNMT(InitArgs(model_out))
# Check
self.assertModelEqual(model._model, model1._model)
