def testSeqtoSeqWithLuongMulAttentionSaveAndLoad(self):
filename = "full_ml_tests/toy_nmt/toy_nmt_luong_mul_attention.xml"
block_filepath = self.setup_holder.filepath_handler.get_test_block_path(
filename)
data_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"nmt/toy/")
embedding_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"embeddings/")
interface = BasicInterface()
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_folder", embedding_filepath)
interface.initialize()
f = open(data_filepath + "tgt.txt")
lines = [l.strip() for l in f]
gold_sentences = [l + " EOS" for l in lines]
f.close()
interface.train()
predictions = interface.predict()
self.assertEqual(len(gold_sentences), len(predictions))
for i, s in enumerate(gold_sentences):
pred_sent = " ".join(predictions[i])
self.assertEqual(s, pred_sent)
model_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"stored_models/toy_nmt/luong")
if os.path.exists(model_filepath):
shutil.rmtree(model_filepath)
interface.save(model_filepath)
interface_2 = BasicInterface()
interface_2.load_file(block_filepath)
interface_2.set_variable("data_folder", data_filepath)
interface_2.set_variable("embedding_folder", embedding_filepath)
interface_2.initialize()
interface_2.load(model_filepath)
predictions = interface_2.predict()
self.assertEqual(len(gold_sentences), len(predictions))
for i, s in enumerate(gold_sentences):
pred_sent = " ".join(predictions[i])
self.assertEqual(s, pred_sent)
def testSeqtoSeqWithVariableDecoderLayers(self):
filename = "full_ml_tests/toy_nmt/toy_nmt_variable_output_lstm_layers.xml"
block_filepath = self.setup_holder.filepath_handler.get_test_block_path(
filename)
data_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"nmt/toy/")
embedding_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"embeddings/")
interface = BasicInterface()
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_folder", embedding_filepath)
interface.set_variable("lstm_layers", "2")
interface.initialize()
f = open(data_filepath + "tgt.txt")
lines = [l.strip() for l in f]
gold_sentences = [l + " EOS" for l in lines]
f.close()
interface.train()
predictions = interface.predict()
self.assertEqual(len(gold_sentences), len(predictions))
for i, s in enumerate(gold_sentences):
pred_sent = " ".join(predictions[i])
self.assertEqual(s, pred_sent)
interface = BasicInterface()
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_folder", embedding_filepath)
interface.set_variable("lstm_layers", "1")
interface.initialize()
f = open(data_filepath + "tgt.txt")
lines = [l.strip() for l in f]
gold_sentences = [l + " EOS" for l in lines]
f.close()
interface.train()
predictions = interface.predict()
self.assertEqual(len(gold_sentences), len(predictions))
for i, s in enumerate(gold_sentences):
pred_sent = " ".join(predictions[i])
self.assertEqual(s, pred_sent)
def testReverseSequenceTaskWithCombinedVocab(self):
filename = "full_ml_tests/pointer_network/pointer_plus_decoder.xml"
block_filepath = self.setup_holder.filepath_handler.get_test_block_path(
filename)
data_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"nmt/reverse_sequence_plus_vocab/")
embedding_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"embeddings/")
interface = BasicInterface()
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_folder", embedding_filepath)
interface.initialize()
f = open(data_filepath + "tgt.txt")
lines = [l.strip() for l in f]
gold_sentences = [l + " EOS" for l in lines]
f.close()
interface.train()
predictions = interface.predict()
self.assertEqual(len(gold_sentences), len(predictions))
for i, s in enumerate(gold_sentences):
pred_sent = " ".join(predictions[i])
self.assertEqual(s, pred_sent)
def testVariationalLuongMulAttentionWithConditionalPrior(self):
filename = "full_ml_tests/toy_nmt/variational_toy_nmt_conditional_prior.xml"
block_filepath = self.setup_holder.filepath_handler.get_test_block_path(
filename)
data_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"nmt/toy/")
embedding_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"embeddings/")
interface = BasicInterface()
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_folder", embedding_filepath)
interface.initialize()
f = open(data_filepath + "tgt.txt")
lines = [l.strip() for l in f]
gold_sentences = [l + " EOS" for l in lines]
f.close()
interface.train()
predictions = interface.predict()
self.assertEqual(len(gold_sentences), len(predictions))
for i, s in enumerate(gold_sentences):
pred_sent = " ".join(predictions[i])
self.assertEqual(s, pred_sent)
def testSeqtoSeqWithSgdLrDecease(self):
filename = "full_ml_tests/toy_nmt/toy_nmt_sgd_learning_rate_decay.xml"
block_filepath = self.setup_holder.filepath_handler.get_test_block_path(
filename)
data_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"nmt/toy/")
embedding_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"embeddings/")
interface = BasicInterface()
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_folder", embedding_filepath)
interface.initialize()
f = open(data_filepath + "tgt.txt")
lines = [l.strip() for l in f]
gold_sentences = [l + " EOS" for l in lines]
f.close()
upd_component = interface.get_execution_component("upd")
learning_rate_variable = upd_component.value_model.get_learning_rate()
lr = interface.ml_helper.tensorflow_session_model.run(
learning_rate_variable, {})
self.assertAlmostEqual(0.1, lr)
interface.train(iterations=10)
lr = interface.ml_helper.tensorflow_session_model.run(
learning_rate_variable, {})
self.assertAlmostEqual(0.1, lr)
interface.train(iterations=1991)
lr = interface.ml_helper.tensorflow_session_model.run(
learning_rate_variable, {})
self.assertAlmostEqual(0.0999, lr)
interface.train(iterations=1)
lr = interface.ml_helper.tensorflow_session_model.run(
learning_rate_variable, {})
self.assertAlmostEqual(0.0998001, lr)
interface.train()
predictions = interface.predict()
self.assertEqual(len(gold_sentences), len(predictions))
for i, s in enumerate(gold_sentences):
pred_sent = " ".join(predictions[i])
self.assertEqual(s, pred_sent)
def testPredictWithoutTraining(self):
interface = BasicInterface()
block_filename = "full_ml_tests/relation_prediction/pure_relation_prediction.xml"
block_filepath = self.setup_holder.filepath_handler.get_test_block_path(block_filename)
data_filepath = self.setup_holder.filepath_handler.get_test_data_path("relation_prediction/toy-125/")
embedding_filepath = self.setup_holder.filepath_handler.get_test_data_path("embeddings/glove.6B.100d.txt")
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_filepath", embedding_filepath)
interface.initialize()
pred = interface.predict()
self.assertEqual(51, len(pred))
self.assertIn("->", pred[0])
def testSeqtoSeqWithUnkReplacement(self):
filename = "full_ml_tests/toy_nmt/toy_nmt_replace_unks.xml"
block_filepath = self.setup_holder.filepath_handler.get_test_block_path(
filename)
data_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"nmt/toy/")
embedding_filepath = self.setup_holder.filepath_handler.get_test_data_path(
"embeddings/")
interface = BasicInterface()
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_folder", embedding_filepath)
interface.add_console_logger({"formatting": ["token_replacement"]})
interface.initialize()
f = open(data_filepath + "tgt.txt")
lines = [l.strip() for l in f]
gold_sentences = [l + " EOS" for l in lines]
f.close()
interface.train()
predictions = interface.predict()
self.assertEqual(len(gold_sentences), len(predictions))
unk_count = 0
missed_count = 0
for i, s in enumerate(gold_sentences):
gold_tokens = s.split(" ")
for j, t in enumerate(gold_tokens):
if predictions[i][j] != "UNK":
if predictions[i][j] != t:
missed_count += 1
else:
unk_count += 1
self.assertEqual(0, unk_count)
self.assertGreater(10, missed_count)
def testPredictEmbeddings(self):
interface = BasicInterface()
block_filename = "full_ml_tests/relation_prediction/predict_embeddings.xml"
block_filepath = self.setup_holder.filepath_handler.get_test_block_path(block_filename)
data_filepath = self.setup_holder.filepath_handler.get_test_data_path("relation_prediction/toy-125/")
embedding_filepath = self.setup_holder.filepath_handler.get_test_data_path("embeddings/glove.6B.100d.txt")
interface.load_file(block_filepath)
interface.set_variable("data_folder", data_filepath)
interface.set_variable("embedding_filepath", embedding_filepath)
interface.initialize()
embs = interface.predict()
self.assertEqual(51, len(embs))
self.assertEqual(81, len(embs[0]))
self.assertEqual(60, len(embs[1]))
self.assertEqual(181, len(embs[0][0]))
