def train_parser(parser,
optimizer,
dataset,
n_epochs=1,
n_train_insts=1000,
name="eng"):
for epoch in range(n_epochs):
model_name = name + "-bakeoff-" + str(epoch + 1) + ".model"
print("Epoch {}".format(epoch + 1), "Model:", model_name)
parser.train() # turn on dropout layers if they are there
parsing.train(dataset.training_data[:n_train_insts],
parser,
optimizer,
verbose=True)
print("Dev Evaluation")
parser.eval() # turn them off for evaluation
parsing.evaluate(dataset.dev_data, parser, verbose=True)
print("F-Score: {}".format(
evaluation.compute_metric(parser, dataset.dev_data,
evaluation.fscore)))
print("Attachment Score: {}".format(
evaluation.compute_attachment(parser, dataset.dev_data)))
print("\n")
print("Saving Model:")
torch.save(parser.state_dict(), model_name)
def train_english(model_file_name=None):
# Params
bakeoff_ETA_0_en = 0.01
bakeoff_DROPOUT_en = 0.5
bakeoff_LSTM_NUM_LAYERS_en = 1
pretrained_embeds = pickle.load(open("data/polyglot-en.pkl", 'rb'),
encoding='latin1')
pretrained_embeds = transform_tuple_to_dict(pretrained_embeds)
en_dataset = Dataset(EN_TRAIN_FILE, EN_DEV_FILE, EN_TEST_FILE)
word_to_ix_en = {word: i for i, word in enumerate(en_dataset.vocab)}
en_dev_data = [i.sentence for i in en_dataset.dev_data]
train(ETA=bakeoff_ETA_0_en,
DROPOUT=bakeoff_DROPOUT_en,
LSTM_NUM_LAYERS=bakeoff_LSTM_NUM_LAYERS_en,
n_epochs=4,
dataset=en_dataset,
word_to_ix=word_to_ix_en,
pretrained_embeds=pretrained_embeds,
output_preds_filename="bakeoff-dev-en.preds",
dev_data=en_dev_data,
name="eng",
model_file_name=model_file_name,
bakeoff_csv_name="KAGGLE-bakeoff-preds-en.csv",
output_test_filename="bakeoff-test-en.preds")
def train_norwegian():
# Params
bakeoff_ETA_0_nr = 0.01
bakeoff_DROPOUT_nr = 0.5
bakeoff_LSTM_NUM_LAYERS_nr = 1
pretrained_embeds_nr = pickle.load(
open(PRETRAINED_EMBEDS_FILE_NR,
'rb'), encoding='latin1') # NOT DOING ANYTHING FOR NORWEGIAN
pretrained_embeds = transform_tuple_to_dict(pretrained_embeds_nr)
nr_dataset = Dataset(NR_TRAIN_FILE, NR_DEV_FILE, NR_TEST_FILE)
word_to_ix_nr = {word: i for i, word in enumerate(nr_dataset.vocab)}
nr_dev_data = [i.sentence for i in nr_dataset.dev_data]
train(ETA=bakeoff_ETA_0_nr,
DROPOUT=bakeoff_DROPOUT_nr,
LSTM_NUM_LAYERS=bakeoff_LSTM_NUM_LAYERS_nr,
n_epochs=5,
dataset=nr_dataset,
word_to_ix=word_to_ix_nr,
pretrained_embeds=pretrained_embeds,
output_preds_filename="bakeoff-dev-nr.preds",
dev_data=nr_dev_data,
name="norweg",
bakeoff_csv_name="KAGGLE-bakeoff-preds-nr.csv",
output_test_filename="bakeoff-test-nr.preds")
def test_predict_after_train_d3_1():
global test_sent, gold, word_to_ix, vocab
torch.manual_seed(1)
feat_extract = SimpleFeatureExtractor()
word_embed = VanillaWordEmbedding(word_to_ix, TEST_EMBEDDING_DIM)
act_chooser = FFActionChooser(TEST_EMBEDDING_DIM * NUM_FEATURES)
combiner = FFCombiner(TEST_EMBEDDING_DIM)
parser = TransitionParser(feat_extract, word_embed, act_chooser, combiner)
# Train
for i in range(75):
train([ (test_sent[:-1], gold) ], parser, optim.SGD(parser.parameters(), lr=0.01), verbose=False)
# predict
pred = parser.predict(test_sent[:-1])
gold_graph = dependency_graph_from_oracle(test_sent[:-1], gold)
assert pred == gold_graph
def test_predict_after_train_d3_1():
""" 1 point(s) """
global test_sent, gold, word_to_ix, vocab
torch.manual_seed(1)
feat_extract = SimpleFeatureExtractor()
word_embed = VanillaWordEmbeddingLookup(word_to_ix, TEST_EMBEDDING_DIM)
act_chooser = ActionChooserNetwork(TEST_EMBEDDING_DIM * NUM_FEATURES)
combiner = MLPCombinerNetwork(TEST_EMBEDDING_DIM)
parser = TransitionParser(feat_extract, word_embed, act_chooser, combiner)
# Train
for i in xrange(75):
train([ (test_sent[:-1], gold) ], parser, optim.SGD(parser.parameters(), lr=0.01), verbose=False)
# predict
pred = parser.predict(test_sent[:-1])
gold_graph = dependency_graph_from_oracle(test_sent[:-1], gold)
assert pred == gold_graph