def test_loss_model1(self):
layer_dim = 512
model = DefaultTranslator(
src_reader=self.src_reader,
trg_reader=self.trg_reader,
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
attender=MlpAttender(input_dim=layer_dim,
state_dim=layer_dim,
hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=MlpSoftmaxDecoder(input_dim=layer_dim,
trg_embed_dim=layer_dim,
rnn_layer=UniLSTMSeqTransducer(
input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="model.decoder.rnn_layer"),
mlp_layer=MLP(
input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_rnn_dim=layer_dim,
vocab_size=100,
yaml_path="model.decoder.rnn_layer"),
bridge=CopyBridge(dec_dim=layer_dim,
dec_layers=1)),
)
model.set_train(False)
self.assert_single_loss_equals_batch_loss(model)
class TestFreeDecodingLoss(unittest.TestCase):
def setUp(self):
layer_dim = 512
xnmt.events.clear()
ParamManager.init_param_col()
self.model = DefaultTranslator(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim, hidden_dim=layer_dim),
attender=MlpAttender(input_dim=layer_dim, state_dim=layer_dim, hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=MlpSoftmaxDecoder(input_dim=layer_dim, lstm_dim=layer_dim, mlp_hidden_dim=layer_dim,
trg_embed_dim=layer_dim, vocab_size=100,
bridge=CopyBridge(dec_dim=layer_dim, dec_layers=1)),
)
self.model.set_train(False)
self.model.initialize_generator(beam=1)
self.src_data = list(self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(self.model.trg_reader.read_sents("examples/data/head.en"))
def test_single(self):
dy.renew_cg()
self.model.initialize_generator(beam=1)
outputs = self.model.generate_output(self.src_data[0], 0,
forced_trg_ids=self.trg_data[0])
dy.renew_cg()
train_loss = self.model.calc_loss(src=self.src_data[0],
trg=outputs[0].actions,
loss_calculator=LossCalculator()).value()
self.assertAlmostEqual(-outputs[0].score, train_loss, places=4)
class TestGreedyVsBeam(unittest.TestCase):
"""
Test if greedy search produces same output as beam search with beam 1.
"""
def setUp(self):
layer_dim = 512
xnmt.events.clear()
ParamManager.init_param_col()
self.model = DefaultTranslator(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
attender=MlpAttender(input_dim=layer_dim,
state_dim=layer_dim,
hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=MlpSoftmaxDecoder(input_dim=layer_dim,
trg_embed_dim=layer_dim,
rnn_layer=UniLSTMSeqTransducer(
input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="model.decoder.rnn_layer"),
mlp_layer=MLP(
input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_rnn_dim=layer_dim,
vocab_size=100,
yaml_path="model.decoder.rnn_layer"),
bridge=CopyBridge(dec_dim=layer_dim,
dec_layers=1)),
)
self.model.set_train(False)
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
def test_greedy_vs_beam(self):
dy.renew_cg()
self.model.initialize_generator()
outputs = self.model.generate_output(self.src_data[0],
0,
BeamSearch(beam_size=1),
forced_trg_ids=self.trg_data[0])
output_score1 = outputs[0].score
dy.renew_cg()
self.model.initialize_generator()
outputs = self.model.generate_output(self.src_data[0],
0,
GreedySearch(),
forced_trg_ids=self.trg_data[0])
output_score2 = outputs[0].score
self.assertAlmostEqual(output_score1, output_score2)
class TestForcedDecodingOutputs(unittest.TestCase):
def assertItemsEqual(self, l1, l2):
self.assertEqual(len(l1), len(l2))
for i in range(len(l1)):
self.assertEqual(l1[i], l2[i])
def setUp(self):
layer_dim = 512
xnmt.events.clear()
ParamManager.init_param_col()
self.model = DefaultTranslator(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
attender=MlpAttender(input_dim=layer_dim,
state_dim=layer_dim,
hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=MlpSoftmaxDecoder(input_dim=layer_dim,
trg_embed_dim=layer_dim,
rnn_layer=UniLSTMSeqTransducer(
input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="model.decoder.rnn_layer"),
mlp_layer=MLP(
input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_rnn_dim=layer_dim,
vocab_size=100,
yaml_path="model.decoder.rnn_layer"),
bridge=CopyBridge(dec_dim=layer_dim,
dec_layers=1)),
)
self.model.set_train(False)
self.model.initialize_generator()
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
self.search = GreedySearch()
def assert_forced_decoding(self, sent_id):
dy.renew_cg()
outputs = self.model.generate_output(
self.src_data[sent_id],
sent_id,
self.search,
forced_trg_ids=self.trg_data[sent_id])
self.assertItemsEqual(self.trg_data[sent_id], outputs[0].actions)
def test_forced_decoding(self):
for i in range(1):
self.assert_forced_decoding(sent_id=i)
def test_loss_model2(self):
model = DefaultTranslator(
src_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
encoder=PyramidalLSTMSeqTransducer(self.model_context, layers=3),
attender=MlpAttender(self.model_context),
trg_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
decoder=MlpSoftmaxDecoder(self.model_context, vocab_size=100),
)
model.set_train(False)
self.assert_single_loss_equals_batch_loss(model, pad_src_to_multiple=4)
def test_loss_model1(self):
model = DefaultTranslator(
src_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
encoder=BiLSTMSeqTransducer(self.model_context),
attender=MlpAttender(self.model_context),
trg_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
decoder=MlpSoftmaxDecoder(self.model_context, vocab_size=100),
)
model.set_train(False)
self.assert_single_loss_equals_batch_loss(model)
class TestGreedyVsBeam(unittest.TestCase):
"""
Test if greedy search produces same output as beam search with beam 1.
"""
def setUp(self):
xnmt.events.clear()
self.model_context = ModelContext()
self.model_context.dynet_param_collection = PersistentParamCollection(
"some_file", 1)
self.model = DefaultTranslator(
src_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
encoder=BiLSTMSeqTransducer(self.model_context),
attender=MlpAttender(self.model_context),
trg_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
decoder=MlpSoftmaxDecoder(self.model_context,
vocab_size=100,
bridge=CopyBridge(self.model_context,
dec_layers=1)),
)
self.model.initialize_training_strategy(TrainingStrategy())
self.model.set_train(False)
self.training_corpus = BilingualTrainingCorpus(
train_src="examples/data/head.ja",
train_trg="examples/data/head.en",
dev_src="examples/data/head.ja",
dev_trg="examples/data/head.en")
self.corpus_parser = BilingualCorpusParser(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
training_corpus=self.training_corpus)
def test_greedy_vs_beam(self):
dy.renew_cg()
self.model.initialize_generator(beam=1)
outputs = self.model.generate_output(
self.training_corpus.train_src_data[0],
0,
forced_trg_ids=self.training_corpus.train_trg_data[0])
output_score1 = outputs[0].score
dy.renew_cg()
self.model.initialize_generator()
outputs = self.model.generate_output(
self.training_corpus.train_src_data[0],
0,
forced_trg_ids=self.training_corpus.train_trg_data[0])
output_score2 = outputs[0].score
self.assertAlmostEqual(output_score1, output_score2)
class TestForcedDecodingOutputs(unittest.TestCase):
def assertItemsEqual(self, l1, l2):
self.assertEqual(len(l1), len(l2))
for i in range(len(l1)):
self.assertEqual(l1[i], l2[i])
def setUp(self):
layer_dim = 512
xnmt.events.clear()
ParamManager.init_param_col()
self.model = DefaultTranslator(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
attender=MlpAttender(input_dim=layer_dim,
state_dim=layer_dim,
hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=AutoRegressiveDecoder(
input_dim=layer_dim,
trg_embed_dim=layer_dim,
rnn=UniLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="model.decoder.rnn"),
transform=NonLinear(input_dim=layer_dim * 2,
output_dim=layer_dim),
scorer=Softmax(input_dim=layer_dim, vocab_size=100),
bridge=CopyBridge(dec_dim=layer_dim, dec_layers=1)),
)
self.model.set_train(False)
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
def assert_forced_decoding(self, sent_id):
dy.renew_cg()
outputs = self.model.generate(
xnmt.batcher.mark_as_batch([self.src_data[sent_id]]), [sent_id],
BeamSearch(),
forced_trg_ids=xnmt.batcher.mark_as_batch([self.trg_data[sent_id]
]))
self.assertItemsEqual(self.trg_data[sent_id].words, outputs[0].actions)
def test_forced_decoding(self):
for i in range(1):
self.assert_forced_decoding(sent_id=i)
class TestFreeDecodingLoss(unittest.TestCase):
def setUp(self):
layer_dim = 512
xnmt.events.clear()
ParamManager.init_param_col()
self.model = DefaultTranslator(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
attender=MlpAttender(input_dim=layer_dim,
state_dim=layer_dim,
hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=AutoRegressiveDecoder(
input_dim=layer_dim,
trg_embed_dim=layer_dim,
rnn=UniLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="model.decoder.rnn"),
transform=NonLinear(input_dim=layer_dim * 2,
output_dim=layer_dim),
scorer=Softmax(input_dim=layer_dim, vocab_size=100),
bridge=CopyBridge(dec_dim=layer_dim, dec_layers=1)),
)
self.model.set_train(False)
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
def test_single(self):
dy.renew_cg()
outputs = self.model.generate(
xnmt.batcher.mark_as_batch([self.src_data[0]]), [0],
GreedySearch(),
forced_trg_ids=xnmt.batcher.mark_as_batch([self.trg_data[0]]))
output_score = outputs[0].score
dy.renew_cg()
train_loss = self.model.calc_loss(
src=self.src_data[0],
trg=outputs[0],
loss_calculator=AutoRegressiveMLELoss()).value()
self.assertAlmostEqual(-output_score, train_loss, places=5)
def test_loss_model2(self):
model = DefaultTranslator(
src_reader=self.src_reader,
trg_reader=self.trg_reader,
src_embedder=SimpleWordEmbedder(self.exp_global, vocab_size=100),
encoder=PyramidalLSTMSeqTransducer(self.exp_global, layers=3),
attender=MlpAttender(self.exp_global),
trg_embedder=SimpleWordEmbedder(self.exp_global, vocab_size=100),
decoder=MlpSoftmaxDecoder(self.exp_global,
vocab_size=100,
bridge=CopyBridge(
exp_global=self.exp_global,
dec_layers=1)),
)
model.set_train(False)
self.assert_single_loss_equals_batch_loss(model, pad_src_to_multiple=4)
def test_loss_model3(self):
model = DefaultTranslator(
src_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
encoder=BiLSTMSeqTransducer(self.model_context, layers=3),
attender=MlpAttender(self.model_context),
trg_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
decoder=MlpSoftmaxDecoder(self.model_context,
vocab_size=100,
bridge=CopyBridge(self.model_context,
dec_layers=1)),
)
model.initialize_training_strategy(TrainingStrategy())
model.set_train(False)
self.assert_single_loss_equals_batch_loss(model)
class TestFreeDecodingLoss(unittest.TestCase):
def setUp(self):
xnmt.events.clear()
self.model_context = ModelContext()
self.model_context.dynet_param_collection = PersistentParamCollection(
"some_file", 1)
self.model = DefaultTranslator(
src_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
encoder=BiLSTMSeqTransducer(self.model_context),
attender=MlpAttender(self.model_context),
trg_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
decoder=MlpSoftmaxDecoder(self.model_context,
vocab_size=100,
bridge=CopyBridge(self.model_context,
dec_layers=1)),
)
self.model.initialize_training_strategy(TrainingStrategy())
self.model.set_train(False)
self.model.initialize_generator()
self.training_corpus = BilingualTrainingCorpus(
train_src="examples/data/head.ja",
train_trg="examples/data/head.en",
dev_src="examples/data/head.ja",
dev_trg="examples/data/head.en")
self.corpus_parser = BilingualCorpusParser(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
training_corpus=self.training_corpus)
def test_single(self):
dy.renew_cg()
self.model.initialize_generator()
outputs = self.model.generate_output(
self.training_corpus.train_src_data[0],
0,
forced_trg_ids=self.training_corpus.train_trg_data[0])
output_score = outputs[0].score
dy.renew_cg()
train_loss = self.model.calc_loss(
src=self.training_corpus.train_src_data[0],
trg=outputs[0].actions).value()
self.assertAlmostEqual(-output_score, train_loss, places=5)
class TestGreedyVsBeam(unittest.TestCase):
"""
Test if greedy search produces same output as beam search with beam 1.
"""
def setUp(self):
xnmt.events.clear()
self.exp_global = ExpGlobal(
dynet_param_collection=PersistentParamCollection("some_file", 1))
self.model = DefaultTranslator(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
src_embedder=SimpleWordEmbedder(exp_global=self.exp_global,
vocab_size=100),
encoder=BiLSTMSeqTransducer(exp_global=self.exp_global),
attender=MlpAttender(exp_global=self.exp_global),
trg_embedder=SimpleWordEmbedder(exp_global=self.exp_global,
vocab_size=100),
decoder=MlpSoftmaxDecoder(exp_global=self.exp_global,
vocab_size=100,
bridge=CopyBridge(
exp_global=self.exp_global,
dec_layers=1)),
)
self.model.set_train(False)
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
def test_greedy_vs_beam(self):
dy.renew_cg()
self.model.initialize_generator(beam=1)
outputs = self.model.generate_output(self.src_data[0],
0,
forced_trg_ids=self.trg_data[0])
output_score1 = outputs[0].score
dy.renew_cg()
self.model.initialize_generator()
outputs = self.model.generate_output(self.src_data[0],
0,
forced_trg_ids=self.trg_data[0])
output_score2 = outputs[0].score
self.assertAlmostEqual(output_score1, output_score2)
class TestForcedDecodingOutputs(unittest.TestCase):
def assertItemsEqual(self, l1, l2):
self.assertEqual(len(l1), len(l2))
for i in range(len(l1)):
self.assertEqual(l1[i], l2[i])
def setUp(self):
xnmt.events.clear()
self.model_context = ModelContext()
self.model_context.dynet_param_collection = PersistentParamCollection(
"some_file", 1)
self.model = DefaultTranslator(
src_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
encoder=BiLSTMSeqTransducer(self.model_context),
attender=MlpAttender(self.model_context),
trg_embedder=SimpleWordEmbedder(self.model_context,
vocab_size=100),
decoder=MlpSoftmaxDecoder(self.model_context, vocab_size=100),
)
self.model.initialize_training_strategy(TrainingStrategy())
self.model.set_train(False)
self.model.initialize_generator()
self.training_corpus = BilingualTrainingCorpus(
train_src="examples/data/head.ja",
train_trg="examples/data/head.en",
dev_src="examples/data/head.ja",
dev_trg="examples/data/head.en")
self.corpus_parser = BilingualCorpusParser(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
training_corpus=self.training_corpus)
def assert_forced_decoding(self, sent_id):
dy.renew_cg()
outputs = self.model.generate_output(
self.training_corpus.train_src_data[sent_id],
sent_id,
forced_trg_ids=self.training_corpus.train_trg_data[sent_id])
self.assertItemsEqual(self.training_corpus.train_trg_data[sent_id],
outputs[0].actions)
def test_forced_decoding(self):
for i in range(1):
self.assert_forced_decoding(sent_id=i)
class TestForcedDecodingOutputs(unittest.TestCase):
def assertItemsEqual(self, l1, l2):
self.assertEqual(len(l1), len(l2))
for i in range(len(l1)):
self.assertEqual(l1[i], l2[i])
def setUp(self):
xnmt.events.clear()
self.exp_global = ExpGlobal(
dynet_param_collection=PersistentParamCollection("some_file", 1))
self.model = DefaultTranslator(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
src_embedder=SimpleWordEmbedder(exp_global=self.exp_global,
vocab_size=100),
encoder=BiLSTMSeqTransducer(exp_global=self.exp_global),
attender=MlpAttender(exp_global=self.exp_global),
trg_embedder=SimpleWordEmbedder(exp_global=self.exp_global,
vocab_size=100),
decoder=MlpSoftmaxDecoder(exp_global=self.exp_global,
vocab_size=100,
bridge=CopyBridge(
exp_global=self.exp_global,
dec_layers=1)),
)
self.model.set_train(False)
self.model.initialize_generator(beam=1)
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
def assert_forced_decoding(self, sent_id):
dy.renew_cg()
outputs = self.model.generate_output(
self.src_data[sent_id],
sent_id,
forced_trg_ids=self.trg_data[sent_id])
self.assertItemsEqual(self.trg_data[sent_id], outputs[0].actions)
def test_forced_decoding(self):
for i in range(1):
self.assert_forced_decoding(sent_id=i)
class TestFreeDecodingLoss(unittest.TestCase):
def setUp(self):
xnmt.events.clear()
self.exp_global = ExpGlobal(
dynet_param_collection=PersistentParamCollection("some_file", 1))
self.model = DefaultTranslator(
src_reader=PlainTextReader(),
trg_reader=PlainTextReader(),
src_embedder=SimpleWordEmbedder(exp_global=self.exp_global,
vocab_size=100),
encoder=BiLSTMSeqTransducer(exp_global=self.exp_global),
attender=MlpAttender(exp_global=self.exp_global),
trg_embedder=SimpleWordEmbedder(exp_global=self.exp_global,
vocab_size=100),
decoder=MlpSoftmaxDecoder(exp_global=self.exp_global,
vocab_size=100,
bridge=CopyBridge(
exp_global=self.exp_global,
dec_layers=1)),
)
self.model.set_train(False)
self.model.initialize_generator()
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
def test_single(self):
dy.renew_cg()
self.model.initialize_generator()
outputs = self.model.generate_output(self.src_data[0],
0,
forced_trg_ids=self.trg_data[0])
output_score = outputs[0].score
dy.renew_cg()
train_loss = self.model.calc_loss(
src=self.src_data[0],
trg=outputs[0].actions,
loss_calculator=LossCalculator()).value()
self.assertAlmostEqual(-output_score, train_loss, places=5)
def test_loss_model4(self):
layer_dim = 512
model = DefaultTranslator(
src_reader=self.src_reader,
trg_reader=self.trg_reader,
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
attender=DotAttender(),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=MlpSoftmaxDecoder(input_dim=layer_dim,
lstm_dim=layer_dim,
mlp_hidden_dim=layer_dim,
trg_embed_dim=layer_dim,
vocab_size=100,
bridge=CopyBridge(dec_layers=1,
dec_dim=layer_dim)),
)
model.set_train(False)
self.assert_single_loss_equals_batch_loss(model)
def test_loss_model2(self):
layer_dim = 512
model = DefaultTranslator(
src_reader=self.src_reader,
trg_reader=self.trg_reader,
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=PyramidalLSTMSeqTransducer(input_dim=layer_dim, hidden_dim=layer_dim, layers=3),
attender=MlpAttender(input_dim=layer_dim, state_dim=layer_dim, hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=AutoRegressiveDecoder(input_dim=layer_dim,
trg_embed_dim=layer_dim,
rnn=UniLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="model.decoder.rnn"),
transform=NonLinear(input_dim=layer_dim*2, output_dim=layer_dim),
scorer=Softmax(input_dim=layer_dim, vocab_size=100),
bridge=CopyBridge(dec_dim=layer_dim, dec_layers=1)),
)
model.set_train(False)
self.assert_single_loss_equals_batch_loss(model, pad_src_to_multiple=4)
class TestComposing(unittest.TestCase):
def setUp(self):
# Seeding
numpy.random.seed(2)
random.seed(2)
layer_dim = 64
xnmt.events.clear()
ParamManager.init_param_col()
self.segment_composer = SumComposer()
self.src_reader = CharFromWordTextReader()
self.trg_reader = PlainTextReader()
self.loss_calculator = AutoRegressiveMLELoss()
self.segmenting_encoder = SegmentingSeqTransducer(
segment_composer=self.segment_composer,
final_transducer=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
)
self.model = DefaultTranslator(
src_reader=self.src_reader,
trg_reader=self.trg_reader,
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=self.segmenting_encoder,
attender=MlpAttender(input_dim=layer_dim,
state_dim=layer_dim,
hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=AutoRegressiveDecoder(
input_dim=layer_dim,
rnn=UniLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="decoder"),
transform=AuxNonLinear(input_dim=layer_dim,
output_dim=layer_dim,
aux_input_dim=layer_dim),
scorer=Softmax(vocab_size=100, input_dim=layer_dim),
trg_embed_dim=layer_dim,
bridge=CopyBridge(dec_dim=layer_dim, dec_layers=1)),
)
self.model.set_train(True)
self.layer_dim = layer_dim
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
my_batcher = xnmt.batcher.TrgBatcher(batch_size=3,
src_pad_token=1,
trg_pad_token=2)
self.src, self.trg = my_batcher.pack(self.src_data, self.trg_data)
dy.renew_cg(immediate_compute=True, check_validity=True)
def inp_emb(self, idx=0):
self.model.start_sent(self.src[idx])
embed = self.model.src_embedder.embed_sent(self.src[idx])
return embed
def test_lookup_composer(self):
enc = self.segmenting_encoder
word_vocab = Vocab(vocab_file="examples/data/head.ja.vocab")
word_vocab.freeze()
enc.segment_composer = LookupComposer(word_vocab=word_vocab,
src_vocab=self.src_reader.vocab,
hidden_dim=self.layer_dim)
enc.transduce(self.inp_emb(0))
def test_charngram_composer(self):
enc = self.segmenting_encoder
word_vocab = Vocab(vocab_file="examples/data/head.ja.vocab")
word_vocab.freeze()
enc.segment_composer = CharNGramComposer(
word_vocab=word_vocab,
src_vocab=self.src_reader.vocab,
hidden_dim=self.layer_dim)
enc.transduce(self.inp_emb(0))
def test_add_multiple_segment_composer(self):
enc = self.segmenting_encoder
word_vocab = Vocab(vocab_file="examples/data/head.ja.vocab")
word_vocab.freeze()
enc.segment_composer = SumMultipleComposer(composers=[
LookupComposer(word_vocab=word_vocab,
src_vocab=self.src_reader.vocab,
hidden_dim=self.layer_dim),
CharNGramComposer(word_vocab=word_vocab,
src_vocab=self.src_reader.vocab,
hidden_dim=self.layer_dim)
])
enc.transduce(self.inp_emb(0))
def test_sum_composer(self):
enc = self.segmenting_encoder
enc.segment_composer = SumComposer()
enc.transduce(self.inp_emb(0))
def test_avg_composer(self):
enc = self.segmenting_encoder
enc.segment_composer = AverageComposer()
enc.transduce(self.inp_emb(0))
def test_max_composer(self):
enc = self.segmenting_encoder
enc.segment_composer = MaxComposer()
enc.transduce(self.inp_emb(0))
def test_convolution_composer(self):
enc = self.segmenting_encoder
enc.segment_composer = ConvolutionComposer(ngram_size=1,
embed_dim=self.layer_dim,
hidden_dim=self.layer_dim)
self.model.set_train(True)
enc.transduce(self.inp_emb(0))
enc.segment_composer = ConvolutionComposer(ngram_size=3,
embed_dim=self.layer_dim,
hidden_dim=self.layer_dim)
self.model.set_train(True)
enc.transduce(self.inp_emb(0))
def test_transducer_composer(self):
enc = self.segmenting_encoder
enc.segment_composer = SeqTransducerComposer(
seq_transducer=BiLSTMSeqTransducer(input_dim=self.layer_dim,
hidden_dim=self.layer_dim))
self.model.set_train(True)
enc.transduce(self.inp_emb(0))
class TestSegmentingEncoder(unittest.TestCase):
def setUp(self):
# Seeding
numpy.random.seed(2)
random.seed(2)
layer_dim = 64
xnmt.events.clear()
ParamManager.init_param_col()
self.segment_encoder_bilstm = BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim)
self.segment_composer = SumComposer()
self.src_reader = CharFromWordTextReader()
self.trg_reader = PlainTextReader()
self.loss_calculator = AutoRegressiveMLELoss()
baseline = Linear(input_dim=layer_dim, output_dim=1)
policy_network = Linear(input_dim=layer_dim, output_dim=2)
self.poisson_prior = PoissonPrior(mu=3.3)
self.eps_greedy = EpsilonGreedy(eps_prob=0.0, prior=self.poisson_prior)
self.conf_penalty = ConfidencePenalty()
self.policy_gradient = PolicyGradient(input_dim=layer_dim,
output_dim=2,
baseline=baseline,
policy_network=policy_network,
z_normalization=True,
conf_penalty=self.conf_penalty,
sample=5)
self.length_prior = PoissonLengthPrior(lmbd=3.3, weight=1)
self.segmenting_encoder = SegmentingSeqTransducer(
embed_encoder=self.segment_encoder_bilstm,
segment_composer=self.segment_composer,
final_transducer=BiLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim),
policy_learning=self.policy_gradient,
eps_greedy=self.eps_greedy,
length_prior=self.length_prior,
)
self.model = DefaultTranslator(
src_reader=self.src_reader,
trg_reader=self.trg_reader,
src_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
encoder=self.segmenting_encoder,
attender=MlpAttender(input_dim=layer_dim,
state_dim=layer_dim,
hidden_dim=layer_dim),
trg_embedder=SimpleWordEmbedder(emb_dim=layer_dim, vocab_size=100),
decoder=AutoRegressiveDecoder(
input_dim=layer_dim,
rnn=UniLSTMSeqTransducer(input_dim=layer_dim,
hidden_dim=layer_dim,
decoder_input_dim=layer_dim,
yaml_path="decoder"),
transform=AuxNonLinear(input_dim=layer_dim,
output_dim=layer_dim,
aux_input_dim=layer_dim),
scorer=Softmax(vocab_size=100, input_dim=layer_dim),
trg_embed_dim=layer_dim,
bridge=CopyBridge(dec_dim=layer_dim, dec_layers=1)),
)
self.model.set_train(True)
self.layer_dim = layer_dim
self.src_data = list(
self.model.src_reader.read_sents("examples/data/head.ja"))
self.trg_data = list(
self.model.trg_reader.read_sents("examples/data/head.en"))
my_batcher = xnmt.batcher.TrgBatcher(batch_size=3,
src_pad_token=1,
trg_pad_token=2)
self.src, self.trg = my_batcher.pack(self.src_data, self.trg_data)
dy.renew_cg(immediate_compute=True, check_validity=True)
def test_reinforce_loss(self):
self.model.global_fertility = 1.0
loss = self.model.calc_loss(self.src[0], self.trg[0],
AutoRegressiveMLELoss())
reinforce_loss = self.model.calc_additional_loss(
self.trg[0], self.model, loss)
pl = self.model.encoder.policy_learning
# Ensure correct length
src = self.src[0]
mask = src.mask.np_arr
outputs = self.segmenting_encoder.compose_output
actions = self.segmenting_encoder.segment_actions
# Ensure sample == outputs
self.assertEqual(len(outputs), pl.sample)
self.assertEqual(len(actions), pl.sample)
for sample_action in actions:
for i, sample_item in enumerate(sample_action):
# The last segmentation is 1
self.assertEqual(sample_item[-1], src[i].len_unpadded())
# Assert that all flagged actions are </s>
list(
self.assertEqual(pl.actions[j][0][i], 1)
for j in range(len(mask[i])) if mask[i][j] == 1)
self.assertTrue("mle" in loss.expr_factors)
self.assertTrue("fertility" in loss.expr_factors)
self.assertTrue("rl_reinf" in reinforce_loss.expr_factors)
self.assertTrue("rl_baseline" in reinforce_loss.expr_factors)
self.assertTrue("rl_confpen" in reinforce_loss.expr_factors)
# Ensure we are sampling from the policy learning
self.assertEqual(self.model.encoder.segmenting_action,
SegmentingSeqTransducer.SegmentingAction.POLICY)
def calc_loss_single_batch(self):
loss = self.model.calc_loss(self.src[0], self.trg[0],
AutoRegressiveMLELoss())
reinforce_loss = self.model.calc_additional_loss(
self.trg[0], self.model, loss)
return loss, reinforce_loss
def test_gold_input(self):
self.model.encoder.policy_learning = None
self.model.encoder.eps_greedy = None
self.calc_loss_single_batch()
self.assertEqual(self.model.encoder.segmenting_action,
SegmentingSeqTransducer.SegmentingAction.GOLD)
@unittest.skipUnless(has_cython(), "requires cython to run")
def test_sample_input(self):
self.model.encoder.eps_greedy.eps_prob = 1.0
self.calc_loss_single_batch()
self.assertEqual(
self.model.encoder.segmenting_action,
SegmentingSeqTransducer.SegmentingAction.POLICY_SAMPLE)
self.assertEqual(self.model.encoder.policy_learning.sampling_action,
PolicyGradient.SamplingAction.PREDEFINED)
def test_global_fertility(self):
# Test Global fertility weight
self.model.global_fertility = 1.0
self.segmenting_encoder.policy_learning = None
loss1, _ = self.calc_loss_single_batch()
self.assertTrue("fertility" in loss1.expr_factors)
def test_policy_train_test(self):
self.model.set_train(True)
self.calc_loss_single_batch()
self.assertEqual(self.model.encoder.policy_learning.sampling_action,
PolicyGradient.SamplingAction.POLICY_CLP)
self.model.set_train(False)
self.calc_loss_single_batch()
self.assertEqual(self.model.encoder.policy_learning.sampling_action,
PolicyGradient.SamplingAction.POLICY_AMAX)
def test_no_policy_train_test(self):
self.model.encoder.policy_learning = None
self.model.set_train(True)
self.calc_loss_single_batch()
self.assertEqual(self.model.encoder.segmenting_action,
SegmentingSeqTransducer.SegmentingAction.PURE_SAMPLE)
self.model.set_train(False)
self.calc_loss_single_batch()
self.assertEqual(self.model.encoder.segmenting_action,
SegmentingSeqTransducer.SegmentingAction.PURE_SAMPLE)
def test_sample_during_search(self):
self.model.set_train(False)
self.model.encoder.sample_during_search = True
self.calc_loss_single_batch()
self.assertEqual(self.model.encoder.segmenting_action,
SegmentingSeqTransducer.SegmentingAction.POLICY)
@unittest.skipUnless(has_cython(), "requires cython to run")
def test_policy_gold(self):
self.model.encoder.eps_greedy.prior = GoldInputPrior("segment")
self.model.encoder.eps_greedy.eps_prob = 1.0
self.calc_loss_single_batch()
