def make_material_feats(part, force):
spacy_dir = join(LOCAL_DIR, part, 'spacy')
lempos_feats_dir = '_{}/Material/lempos_feats'.format(part)
if force or not exists(lempos_feats_dir):
generate_feats(spacy_dir, lempos_feats_dir, lemma_pos_feats, nlp=nlp)
word_feats_dir = '_{}/Material/word_feats'.format(part)
if force or not exists(word_feats_dir):
generate_feats(spacy_dir,
word_feats_dir,
lambda sent: word_feats(sent, context_size=1),
nlp=nlp)
wn_feats_dir = '_{}/Material/wordnet_feats'.format(part)
if force or not exists(wn_feats_dir):
generate_feats(spacy_dir,
wn_feats_dir,
lambda s: wordnet_feats(s, context_size=2),
nlp=nlp)
dep_feats_dir = '_{}/Material/dep_feats'.format(part)
if force or not exists(dep_feats_dir):
generate_feats(spacy_dir,
dep_feats_dir,
lambda s: dep_feats(s, context_size=2),
nlp=nlp)
return [lempos_feats_dir, word_feats_dir, wn_feats_dir]
def make_task_feats(part, force):
spacy_dir = join(LOCAL_DIR, part, 'spacy')
lempos_feats_dir = '_{}/Task/lempos_feats'.format(part)
if force or not exists(lempos_feats_dir):
generate_feats(spacy_dir,
lempos_feats_dir,
lambda s: lemma_pos_feats(s, context_size=1),
nlp=nlp)
word_feats_dir = '_{}/Task/word_feats'.format(part)
if force or not exists(word_feats_dir):
generate_feats(spacy_dir,
word_feats_dir,
lambda sent: word_feats(sent, context_size=0),
nlp=nlp)
dep_feats_dir = '_{}/Task/dep_feats'.format(part)
if force or not exists(dep_feats_dir):
generate_feats(spacy_dir,
dep_feats_dir,
lambda sent: dep_feats(sent, context_size=1),
nlp=nlp)
return [lempos_feats_dir, word_feats_dir]
token_feats['{}:wnhypernym3'.format(j)] = wnhypernym3
sent_feats.append(token_feats)
return sent_feats
# Step 1: Generate features
spacy_dir = join(LOCAL_DIR, 'train', 'spacy')
feats_dir = join('_train', 'features2')
# If you want to save time by resusing feats from crf1-exp/py,
# comment out the line below:
generate_feats(spacy_dir, feats_dir, features2)
# Step 2: Collect data for running CRF classifier
true_iob_dir = join(LOCAL_DIR, 'train', 'iob')
data = collect_crf_data(true_iob_dir, feats_dir)
# Step 3: Create folds
# create folds from complete texts only (i.e. instances of the same text
# are never in different folds)
# TODO How to set seed for random generator?
group_k_fold = GroupKFold(n_splits=5)
# use same split for all three entities
for synset in synsets:
try:
token_feats['{}:{}'.format(j, synset.hypernyms()[0].hypernyms()[0].hypernyms()[0].name())] = 1
except:
pass
#token_feats['{}:lemma'.format(j)] = lemma
#token_feats['{}:pos'.format(j)] = pos
sent_feats.append(token_feats)
return sent_feats
generate_feats(spacy_dir, wn_feats_dir, wnfeats1)
# Step 2: Run experiments
crf = PruneCRF(c1=0.1, c2=0.1, all_possible_transitions=True)
feat_dirs = [base_feats_dir, wn_feats_dir, word_feats_dir]
preds = {}
for label in ENTITIES:
preds[label] = run_exp_train_cv(crf, feat_dirs, label, n_folds=5)
# Step 3: Evaluate
eval_exp_train(preds)
from os.path import join
from sklearn_crfsuite import CRF
from sie import ENTITIES, LOCAL_DIR
from sie.feats import generate_feats, features1
from sie.exp import run_exp_test, eval_exp_train
# Step 1: Generate features
train_spacy_dir = join(LOCAL_DIR, 'train', 'spacy')
train_base_feats_dir = join('_train', 'features1')
# If you want to save time by reusing existing feats, comment out the line below:
generate_feats(train_spacy_dir, train_base_feats_dir, features1)
dev_spacy_dir = join(LOCAL_DIR, 'dev', 'spacy')
dev_base_feats_dir = join('_dev', 'features1')
# If you want to save time by reusing existing feats, comment out the line below:
generate_feats(dev_spacy_dir, dev_base_feats_dir, features1)
test_spacy_dir = join(LOCAL_DIR, 'test', 'spacy')
test_base_feats_dir = join('_test', 'features1')
generate_feats(test_spacy_dir, test_base_feats_dir, features1)
# Step 2: Run experiments
crf = CRF(c1=0.1, c2=0.1, all_possible_transitions=True)
from os.path import join
from sie import ENTITIES, LOCAL_DIR, EXPS_DIR
from sie.crf import PruneCRF
from sie.exp import run_exp_train_cv, eval_exp_train
from sie.feats import generate_feats, wordnet_feats
# Step 1: Generate features
base_feats_dir = join(EXPS_DIR, 'crf1/_train/features1')
word_feats_dir = join(EXPS_DIR, 'wordfeats/_train/wordfeats1')
spacy_dir = join(LOCAL_DIR, 'train', 'spacy')
wn_feats_dir = join('_train', 'wnfeats1')
generate_feats(spacy_dir, wn_feats_dir, wordnet_feats)
# Step 2: Run experiments
crf = PruneCRF() #c1=0.1, c2=0.1, min_freq=5)#, all_possible_transitions=True)
feat_dirs = [base_feats_dir, wn_feats_dir, word_feats_dir]
preds = {}
for label in ENTITIES[:1]:
preds[label] = run_exp_train_cv(crf,
feat_dirs,
label,
n_folds=5,
n_jobs=-1)
# Step 3: Evaluate
with word features
"""
from os.path import join
from sie.crf import PruneCRF
from sie import ENTITIES, LOCAL_DIR, EXPS_DIR
from sie.feats import generate_feats, word_feats
from sie.exp import run_exp_train_cv, eval_exp_train
# Step 1: Generate features
spacy_dir = join(LOCAL_DIR, 'train', 'spacy')
word_feats_dir = join('_train', 'wordfeats1')
generate_feats(spacy_dir, word_feats_dir,
lambda sent: word_feats(sent, context_size=1))
# Step 2: Run experiments
crf = PruneCRF() #c1=0.1, c2=0.1, all_possible_transitions=True)
base_feats_dir = join(EXPS_DIR, 'crf1/_train/features1')
feat_dirs = [base_feats_dir, word_feats_dir]
preds = {}
for label in ENTITIES:
preds[label] = run_exp_train_cv(crf, feat_dirs, label)
# Step 3: Evaluate
eval_exp_train(preds)
from os.path import join
from sie import ENTITIES, LOCAL_DIR, EXPS_DIR
from sie.crf import PruneCRF
from sie.exp import run_exp_train_cv, eval_exp_train
from sie.feats import generate_feats, brown_feats
# Step 1: Generate features
spacy_dir = join(LOCAL_DIR, 'train', 'spacy')
brown_feats_dir = join('_train', 'brown_feats')
base_feats_dir = join(EXPS_DIR, 'crf1/_train/features1')
word_feats_dir = join(EXPS_DIR, 'wordfeats/_train/wordfeats1')
generate_feats(spacy_dir, brown_feats_dir, brown_feats)
# Step 2: Run experiments
crf = PruneCRF() #c1=0.1, c2=0.1, all_possible_transitions=True)
base_feats_dir = join(EXPS_DIR, 'crf1/_train/features1')
feat_dirs = [base_feats_dir, brown_feats_dir, word_feats_dir]
preds = {}
for label in ENTITIES:
preds[label] = run_exp_train_cv(crf, feat_dirs, label)
# Step 3: Evaluate
eval_exp_train(preds)
from os.path import join
from sie import ENTITIES, LOCAL_DIR, EXPS_DIR
from sie.crf import PruneCRF
from sie.exp import run_exp_train_cv, eval_exp_train
from sie.feats import generate_feats, dep_feats
# Step 1: Generate features
base_feats_dir = join(EXPS_DIR, 'best/_train/Material/lempos_feats')
word_feats_dir = join(EXPS_DIR, 'best/_train/Material/word_feats')
wordnet_feats_dir = join(EXPS_DIR, 'best/_train/Material/wordnet_feats')
spacy_dir = join(LOCAL_DIR, 'train', 'spacy')
dep_feats_dir = join('_train', 'dep_feats')
generate_feats(spacy_dir, dep_feats_dir,
lambda sent: dep_feats(sent, context_size=1))
# Step 2: Run experiments
crf = PruneCRF() # c1=0.1, c2=0.1, all_possible_transitions=True)
feat_dirs = [
#base_feats_dir,
#word_feats_dir,
#wordnet_feats_dir,
dep_feats_dir
]
preds = {}
for label in ENTITIES:
"""
from os.path import join
from sklearn_crfsuite import CRF
from sie import ENTITIES, LOCAL_DIR
from sie.feats import generate_feats, features1
from sie.exp import run_exp_train, eval_exp_train
# Step 1: Generate features
spacy_dir = join(LOCAL_DIR, 'train', 'spacy')
base_feats_dir = join('_train', 'features1')
# If you want to save time by reusing existing feats, comment out the line below:
generate_feats(spacy_dir, base_feats_dir, features1)
# Step 2: Run experiments
crf = CRF(c1=0.1, c2=0.1, all_possible_transitions=True)
feat_dirs = [base_feats_dir]
preds = {}
for label in ENTITIES:
preds[label] = run_exp_train(crf, feat_dirs, label)
# Step 3: Evaluate
eval_exp_train(preds)