def _makeToyTaggerData(self):
data_config = {}
features_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "src.txt"),
["M . Smith went to Washington .", "I live in New Zealand ."],
)
labels_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "labels.txt"),
["B-PER I-PER E-PER O O S-LOC O", "O O O B-LOC E-LOC O"],
)
data_config["source_vocabulary"] = test_util.make_vocab_from_file(
os.path.join(self.get_temp_dir(), "src_vocab.txt"), features_file)
data_config["target_vocabulary"] = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "labels_vocab.txt"),
[
"O",
"B-LOC",
"I-LOC",
"E-LOC",
"S-LOC",
"B-PER",
"I-PER",
"E-PER",
"S-PER",
],
)
return features_file, labels_file, data_config
def _makeToyEnDeData(self, with_alignments=False):
data_config = {}
features_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "src.txt"),
["Parliament Does Not Support Amendment Freeing Tymoshenko",
"Today , the Ukraine parliament dismissed , within the Code of Criminal Procedure "
"amendment , the motion to revoke an article based on which the opposition leader , "
"Yulia Tymoshenko , was sentenced .",
"The amendment that would lead to freeing the imprisoned former Prime Minister was "
"revoked during second reading of the proposal for mitigation of sentences for "
"economic offences ."])
labels_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "tgt.txt"),
["Keine befreiende Novelle für Tymoshenko durch das Parlament",
"Das ukrainische Parlament verweigerte heute den Antrag , im Rahmen einer Novelle "
"des Strafgesetzbuches denjenigen Paragrafen abzuschaffen , auf dessen Grundlage die "
"Oppositionsführerin Yulia Timoshenko verurteilt worden war .",
"Die Neuregelung , die den Weg zur Befreiung der inhaftierten Expremierministerin hätte "
"ebnen können , lehnten die Abgeordneten bei der zweiten Lesung des Antrags auf Milderung "
"der Strafen für wirtschaftliche Delikte ab ."])
data_config["source_vocabulary"] = test_util.make_vocab_from_file(
os.path.join(self.get_temp_dir(), "src_vocab.txt"), features_file)
data_config["target_vocabulary"] = test_util.make_vocab_from_file(
os.path.join(self.get_temp_dir(), "tgt_vocab.txt"), labels_file)
if with_alignments:
# Dummy and incomplete alignments.
data_config["train_alignments"] = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "aligne.txt"),
["0-0 1-0 2-2 3-4 4-4 5-6",
"0-1 1-1 1-3 2-3 4-4",
"0-0 1-0 2-2 3-4 4-4 5-6"])
return features_file, labels_file, data_config
def testSequenceToSequenceInputter(self):
source_vocabulary = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "src_vocab.txt"),
["<blank>", "<s>", "</s>", "a", "b", "c", "d"],
)
target_vocabulary = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "tgt_vocab.txt"),
["<blank>", "<s>", "</s>", "e", "f", "g", "h"],
)
source_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "src.txt"),
["a c c", "b d", "a e"])
target_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "tgt.txt"),
["f h g", "e h", "a e"])
inputter = sequence_to_sequence.SequenceToSequenceInputter(
text_inputter.WordEmbedder(embedding_size=20),
text_inputter.WordEmbedder(embedding_size=20),
)
inputter.initialize(
dict(source_vocabulary=source_vocabulary,
target_vocabulary=target_vocabulary))
dataset = inputter.make_dataset([source_file, target_file])
element = iter(dataset).next()
features, labels = inputter.make_features(element)
self.assertIn("ids_out", labels)
self.assertAllEqual(labels["ids"], [1, 4, 6, 5])
self.assertAllEqual(labels["ids_out"], [4, 6, 5, 2])
self.assertEqual(labels["length"], 4)
def _makeToyClassifierData(self):
data_config = {}
features_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "src.txt"),
["This product was not good at all , it broke on the first use !",
"Perfect , it does everything I need .",
"How do I change the battery ?"])
labels_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "labels.txt"), ["negative", "positive", "neutral"])
data_config["source_vocabulary"] = test_util.make_vocab_from_file(
os.path.join(self.get_temp_dir(), "src_vocab.txt"), features_file)
data_config["target_vocabulary"] = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "labels_vocab.txt"), ["negative", "positive", "neutral"])
return features_file, labels_file, data_config
def testDatasetSize(self):
path = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "file.txt"),
list(map(str, range(15))))
dataset = tf.data.TextLineDataset(path)
size = dataset_util.get_dataset_size(dataset)
self.assertEqual(self.evaluate(size), 15)
def testTrainLanguageModel(self):
src = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "src.txt"),
["1 2 3 4", "5 6 7 8 9", "3 2"])
vocab = test_util.make_vocab(
os.path.join(self.get_temp_dir(), "vocab.txt"),
list(map(str, range(10))))
config = {
"data": {
"train_features_file": src,
"vocabulary": vocab,
},
"params": {
"learning_rate": 0.0005,
"optimizer": "Adam"
},
"train": {
"batch_size": 10,
"max_step": 2,
},
}
model = models.LanguageModel(decoders.SelfAttentionDecoder(
2, num_units=32, ffn_inner_dim=32),
embedding_size=16,
reuse_embedding=False)
runner = Runner(model, config)
runner.train()
def _makeTransliterationData(self):
ar = [
"آ ت ز م و ن",
"آ ت ش ي س و ن",
"آ ر ب ا ك ه",
"آ ر ث ر",
"آ ز ا",
]
en = ["a t z m o n", "a c h e s o n", "a a r b a k k e", "a r t h u r", "a s a"]
ar_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "ar.txt"), ar
)
en_file = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "en.txt"), en
)
return ar_file, en_file
def _make_model(name, src_vocab, tgt_vocab, random_slots=False):
model, _ = _seq2seq_model(training=True)
optimizer = tf.keras.optimizers.Adam()
data = {}
data["source_vocabulary"] = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "%s-src-vocab.txt" % name),
src_vocab)
data["target_vocabulary"] = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "%s-tgt-vocab.txt" % name),
tgt_vocab)
model.initialize(data)
model.create_variables(optimizer=optimizer)
if random_slots:
for variable in model.trainable_variables:
for slot_name in optimizer.get_slot_names():
slot = optimizer.get_slot(variable, slot_name)
slot.assign(tf.random.uniform(slot.shape))
return model, optimizer
def testLoadSentencePieceVocab(self):
vocab_path = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "vocab_sp"), [
"<unk> 0", "<s> 0", "</s> 0", ", -3.0326", ". -3.41093",
"▁the -3.85169", "s -4.05468", "▁die -4.15914", "▁in -4.2419",
"▁der -4.36135"
])
vocab = Vocab(from_file=vocab_path, from_format="sentencepiece")
self.assertEqual(len(vocab), 7)
self.assertNotIn("<unk>", vocab)
self.assertNotIn("<s>", vocab)
self.assertNotIn("</s>", vocab)
self.assertIn("▁the", vocab)
def _create_dataset(model, temp_dir):
data_path = os.path.join(temp_dir, "data.txt")
test_util.make_data_file(data_path, ["a a a b b d", "a b b b", "c c"])
dataset = model.examples_inputter.make_inference_dataset(data_path, 1)
return dataset
def _run_scorer(self, scorer, refs, hyps):
ref_path = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "ref.txt"), refs)
hyp_path = test_util.make_data_file(
os.path.join(self.get_temp_dir(), "hyp.txt"), hyps)
return scorer(ref_path, hyp_path)
