def load_ltrain(predicate_resume, session):
key_group = predicate_resume["label_group"]
LFs = get_labelling_functions(predicate_resume)
labeler = LabelAnnotator(lfs=LFs)
train_cids_query = get_train_cids_with_span(predicate_resume, session)
L_train = labeler.load_matrix(session,
cids_query=train_cids_query,
key_group=key_group)
return L_train
def get_L_train(LFs, parallelism=2, split=0):
L_train = None
labeler = None
np.random.seed(1701)
labeler = LabelAnnotator(lfs=LFs)
print(datetime.datetime.now())
L_train = labeler.apply(
split=split
) # ,cids_query=session.query(Candidate.id).filter(Candidate.get_parent().id %10==1))
print(datetime.datetime.now())
print(type(L_train))
print(L_train.shape)
# print("**Total non_overlapping_coverage on L_train (percentage of labelled over all)** "+str(L_train.non_overlapping_coverage()))
return L_train
def apply_LF(lf_file):
"""
Load labeling functions and applies on the candidates extracted in train set
:param lf_file: labeling functions python file
:return: L_train
"""
labeling_func = __import__(lf_file)
LF_list = [o[1] for o in getmembers(labeling_func) if isfunction(o[1])]
labeler = LabelAnnotator(lfs=LF_list)
np.random.seed(1701)
L_train = labeler.apply(split=0)
L_train.todense()
report.append('\n#LF Stats\n')
report.append(L_train.lf_stats(session).to_csv(sep=' ', index=False, header=True))
return L_train
def predicate_candidate_labelling(predicate_resume,
parallelism=1,
limit=None,
replace_key_set=False):
logging.info("Starting labeling ")
session = SnorkelSession()
try:
candidate_subclass = predicate_resume["candidate_subclass"]
key_group = predicate_resume["label_group"]
cids_query = session.query(
candidate_subclass.id).filter(candidate_subclass.split == 0)
##skip cands already extracted
#alreadyExistsGroup=session.query(LabelKey).filter(LabelKey.group==key_group).count()>0
#if alreadyExistsGroup:
# cids_query= get_train_cids_not_labeled(predicate_resume,session)
#if limit !=None:
# cids_query=cids_query.filter(candidate_subclass.id<limit)
LFs = get_labelling_functions(predicate_resume)
labeler = LabelAnnotator(lfs=LFs)
np.random.seed(1701)
##if first run or adding a new labeling functionS is needed to set replace key set to True
#if not replace_key_set:
# replace_key_set=not alreadyExistsGroup
L_train = labeler.apply(parallelism=parallelism,
cids_query=cids_query,
key_group=key_group,
clear=True,
replace_key_set=True)
print(L_train.lf_stats(session))
logging.info(L_train.lf_stats(session))
finally:
logging.info("Finished labeling ")
def score_lfs(predicate_resume,
L_gold_test,
session,
date_time,
parallelism=8):
dump_file_path = "./results/" + "lfs_1_" + predicate_resume[
"predicate_name"] + date_time + ".csv"
key_group = predicate_resume["label_group"]
LFs = get_labelling_functions(predicate_resume)
labeler = LabelAnnotator(lfs=LFs)
test_cids_query = get_test_cids_with_span(predicate_resume, session)
L_test = labeler.apply(parallelism=parallelism,
cids_query=test_cids_query,
key_group=key_group,
clear=True,
replace_key_set=False)
data_frame = L_test.lf_stats(session)
print(data_frame)
logging.info(data_frame)
data_frame.to_csv(dump_file_path)
gen_model = GenerativeModel()
gen_model.train(L_test,
epochs=100,
decay=0.95,
step_size=0.1 / L_test.shape[0],
reg_param=1e-6)
p, r, f1 = gen_model.score(L_test, L_gold_test)
print("Prec: {0:.3f}, Recall: {1:.3f}, F1 Score: {2:.3f}".format(p, r, f1))
logging.info("Prec: {0:.3f}, Recall: {1:.3f}, F1 Score: {2:.3f}".format(
p, r, f1))
dump_file_path1 = "./results/" + "test_gen_1_" + predicate_resume[
"predicate_name"] + date_time + ".csv"
with open(dump_file_path1, 'w+b') as f:
writer = csv.writer(f, delimiter=',', quoting=csv.QUOTE_MINIMAL)
writer.writerow(["Precision", "Recall", "F1"])
writer.writerow(
["{0:.3f}".format(p), "{0:.3f}".format(r), "{0:.3f}".format(f1)])
test_marginals = gen_model.marginals(L_test)
dump_file_path2 = "./results/" + "plt_1_" + predicate_resume[
"predicate_name"] + date_time + ".csv"
#plt.hist(test_marginals, bins=20)
#plt.savefig(dump_file_path2)
#plt.show()
dump_file_path3 = "./results/" + "gen_2_" + predicate_resume[
"predicate_name"] + date_time + ".csv"
data_frame3 = gen_model.learned_lf_stats()
data_frame3.to_csv(dump_file_path3)
dump_file_path4 = "./results/" + "gen_3_" + predicate_resume[
"predicate_name"] + date_time + ".csv"
tp, fp, tn, fn = gen_model.error_analysis(session, L_test, L_gold_test)
with open(dump_file_path4, 'w+b') as f:
writer = csv.writer(f, delimiter=',', quoting=csv.QUOTE_MINIMAL)
writer.writerow(["TP", "FP", "TN", "FN"])
writer.writerow(
[str(len(tp)),
str(len(fp)),
str(len(tn)),
str(len(fn))])
dump_file_path5 = "./results/" + "gen_4_" + predicate_resume[
"predicate_name"] + date_time + ".csv"
data_frame4 = L_test.lf_stats(session, L_gold_test,
gen_model.learned_lf_stats()['Accuracy'])
data_frame4.to_csv(dump_file_path5)
def LF_distant_supervision(c):
v, h = c.virus.get_span(), c.host.get_span()
return 1 if (v, h) in known_pairs else 0
# list of all LFs
LFs = [
LF_detect, LF_infect, LF_isolate, LF_positive, LF_positive2, LF_misc,
LF_v_cause_h, LF_v_h, LF_h_v, LF_other_verbs, LF_far_v_h, LF_far_h_v,
LF_neg_h, LF_neg_assertions, LF_distant_supervision
]
# set up the label annotator class
labeler = LabelAnnotator(lfs=LFs)
# -------------------------------------------
# START CROSS VALIDATION SPLIT in a loop:
# Make an array of indexes (should equal number of documents 88). In a loop, split the index array into train, test, and dev arrays. The sentences get added to the respective t,t,d sets and the candidates are extracted.
index_array = np.arange(0, 88)
# for roc
tprs = []
aucs = []
mean_fpr = np.linspace(0, 1, 100)
# for recording prec, rec, f1 scores
candidates = session.query(DiseaseGene).filter(
DiseaseGene.split == 0).limit(1).all()
LF_DEBUG(candidates[0])
# In[ ]:
LFs = get_lfs()
# # Label The Candidates
# Label each candidate based on the provided labels above. This code runs with realtive ease, but optimization is definitely needed when the number of label functions increases linearly.
# In[ ]:
labeler = LabelAnnotator(lfs=LFs)
cids = session.query(Candidate.id).filter(Candidate.split == 0)
get_ipython().magic(
u'time L_train = labeler.apply(split=0, cids_query=cids, parallelism=5)')
cids = session.query(Candidate.id).filter(Candidate.split == 1)
get_ipython().magic(
u'time L_dev = labeler.apply_existing(split=1, cids_query=cids, parallelism=5, clear=False)'
)
cids = session.query(Candidate.id).filter(Candidate.split == 2)
get_ipython().magic(
u'time L_test = labeler.apply_existing(split=2, cids_query=cids, parallelism=5, clear=False)'
)
from utils.label_functions.compound_disease_lf import CD_LFS
#from utils.gene_gene_lf import GG_LFS
# # Label The Candidates
# Label each candidate based on the provided labels above. This code runs with realtive ease, but optimization is definitely needed when the number of label functions increases linearly.
# In[ ]:
label_functions = list(CG_LFS["CbG_DB"].values()) +
list(CG_LFS["CbG_TEXT"].values()) +
list(DG_LFS["DaG_TEXT"].values())
labeler = LabelAnnotator(lfs=label_functions)
# # Quickly Relabel Candidates
# Use this block here to re-label candidates that have already been labled from the above process.
# In[ ]:
train_df = pd.read_excel(spreadsheet_names['train'])
train_cids = train_df.candidate_id.astype(int).tolist()
train_df.head(2)
# In[ ]:
docs = session.query(Document).order_by(Document.name).all()
sentences = session.query(Sentence).all()
#print(sentences)
sents=set();
for i,doc in enumerate(docs):
for s in doc.sentences:
sents.add(s)
cand_extractor.apply(sents)
print("Number of candidates:", session.query(pairs).count())
labeler = LabelAnnotator(lfs=LFs)
L_train = labeler.apply()
print(L_train.lf_stats(session))
# generative model, training_marginals are probabilistic training labels
gen_model = GenerativeModel()
gen_model.train(L_train, epochs=100, decay=0.95, step_size=0.1 / L_train.shape[0], reg_param=1e-6)
print(gen_model.weights.lf_accuracy)
train_marginals = gen_model.marginals(L_train)
candidate_dfs = {
key: load_candidate_dataframes(spreadsheet_names[key])
for key in spreadsheet_names
}
for key in candidate_dfs:
print("Size of {} set: {}".format(key, candidate_dfs[key].shape[0]))
# In[8]:
label_functions = (list(DG_LFS["DaG_DB"].values()) +
list(DG_LFS["DaG_TEXT"].values()))
if quick_load:
labeler = LabelAnnotator(lfs=[])
label_matricies = {
key:
labeler.load_matrix(session,
cids_query=make_cids_query(session,
candidate_dfs[key]))
for key in candidate_dfs
}
else:
labeler = LabelAnnotator(lfs=label_functions)
label_matricies = {
key:
label_candidates(labeler,
LF_common_1000, LF_common_2000
]
LFs_BD = [
LF_colon, LF_known_abs, LF_single_letter, LF_roman_numeral, LF_common_2000,
LF_same_thing_BD
]
LFs_BM = [
LF_distance_far, LF_colon, LF_known_abs, LF_single_letter,
LF_roman_numeral, LF_common_2000, LF_same_thing
]
LFs_BT = [
LF_colon, LF_known_abs, LF_single_letter, LF_roman_numeral, LF_common_2000,
LF_same_thing
]
labeler_BC = LabelAnnotator(lfs=LFs_BC)
labeler_BD = LabelAnnotator(lfs=LFs_BD)
labeler_BM = LabelAnnotator(lfs=LFs_BM)
labeler_BT = LabelAnnotator(lfs=LFs_BT)
# Training
L_train_BC = labeler_BC.apply(split=0)
L_train_BD = labeler_BD.apply(split=0)
L_train_BM = labeler_BM.apply(split=0)
L_train_BT = labeler_BT.apply(split=0)
L_train_BC
L_train_BD
L_train_BM
L_train_BT
# Labeling Function Performance - Coverage, Overlaps, Conflicts
def _get_labeler(predicate_resume):
LFs = get_labelling_functions(predicate_resume)
labeler = LabelAnnotator(lfs=LFs)
return labeler
"Number of candidates:",
session.query(candidate_class).filter(candidate_class.split == 0).count())
print("==============================")
# Split to pull eval candidates from
eval_split = 0
# Executing query for eval candidates
eval_cands = session.query(candidate_class).filter(
candidate_class.split == eval_split).order_by(candidate_class.id).all()
print(f'Loaded {len(eval_cands)} candidates...')
# Applying LFs
print("Applying LFs...")
from snorkel.annotations import LabelAnnotator
labeler = LabelAnnotator(lfs=LFs)
L_eval = labeler.apply(split=eval_split, parallelism=parallelism)
# defining model
from snorkel.learning import GenerativeModel
# Creating generative model
gen_model = GenerativeModel()
# defining saved weights directory and name
model_name = 'Price_Gen_20K' # this was provided when the model was saved!
save_dir = '/dfs/scratch0/jdunnmon/data/memex-data/extractor_checkpoints/Price_Gen_20K' # this was provided when the model was saved!
# loading
print("Loading generative model...")
gen_model.load(model_name=model_name, save_dir=save_dir, verbose=True)
BiomarkerCondition).filter(BiomarkerCondition.split == 1).count()
session.commit()
# In[ ]:
from LF import *
LFs_BC = [
LF_markerDatabase, LF_keyword, LF_distance, LF_abstract_titleWord,
LF_single_letter, LF_auxpass, LF_known_abs, LF_same_thing_BC,
LF_common_1000, LF_common_2000
]
# In[ ]:
from snorkel.annotations import LabelAnnotator
BC_labeler = LabelAnnotator(lfs=LFs_BC)
# In[ ]:
np.random.seed(1701)
get_ipython().magic(u'time L_train_BC = BC_labeler.apply(split=0)')
L_train_BC
# In[ ]:
get_ipython().magic(
u'time L_train_BC = BC_labeler.load_matrix(session, split=0)')
L_train_BC
# In[ ]:
#candidates = [session.query(DiseaseGene).filter(DiseaseGene.id == ids).one() for ids in [19817,19818,19830,19862,19980,20001,20004]]
for c in candidates:
if c[0].get_parent().id != 14264:
continue
print c
print get_tagged_text(c)
print c[1].sentence.entity_cids[c[1].get_word_start()]
# # Label The Candidates
# This block of code will run through the label functions and label each candidate in the training and development groups.
# In[ ]:
labeler = LabelAnnotator(f=LFs)
get_ipython().magic(u'time L_train = labeler.apply(split=0)')
get_ipython().magic(u'time L_dev = labeler.apply_existing(split=1)')
get_ipython().magic(u'time L_test = labeler.apply_existing(split=2)')
# In[ ]:
featurizer = FeatureAnnotator()
get_ipython().magic(u'time F_train = featurizer.apply(split=0)')
get_ipython().magic(u'time F_dev = featurizer.apply_existing(split=1)')
get_ipython().magic(u'time F_test = featurizer.apply_existing(split=2)')
# # Generate Coverage Stats
GeneGene = candidate_subclass('GeneGene', ['Gene1', 'Gene2'])
elif edge_type == "cg":
CompoundGene = candidate_subclass('CompoundGene', ['Compound', 'Gene'])
elif edge_type == "cd":
CompoundDisease = candidate_subclass('CompoundDisease',
['Compound', 'Disease'])
else:
print("Please pick a valid edge type")
# # Load the data
# Here is where we load the test dataset in conjunction with the previously trained disc models. Each algorithm will output a probability of a candidate being a true candidate.
# In[6]:
labeler = LabelAnnotator(lfs=[])
# In[7]:
get_ipython().run_cell_magic(u'time', u'',
u'L_test = labeler.load_matrix(session,split=2)')
# In[8]:
L_test.shape
# In[9]:
marginal_files = [
"stratified_data/lstm_disease_gene_holdout/LR_data/LR_test_marginals.csv",
"stratified_data/lstm_disease_gene_holdout/lstm_one_test_marginals.csv",
# In[ ]: database_str = "sqlite:///" + os.environ['WORKINGPATH'] + "/Database/epilepsy.db" os.environ['SNORKELDB'] = database_str session = SnorkelSession() # # Load preprocessed data # To save time, this code will automatically load our labels that were generated in the previous file. # In[ ]: labeler = LabelAnnotator(f=None) L_train = labeler.load_matrix(session,split=0) L_dev = labeler.load_matrix(session,split=1) L_test = labeler.load_matrix(session,split=2) # In[ ]: print "Total Data Shape:" print L_train.shape print L_dev.shape print L_test.shape print print "The number of positive candiadtes (in KB) for each division:"
tweet = Tweet(tweet=raw_text, split=split)
session.add(tweet)
session.commit()
print("Commit to snorkel database done...")
#writing label generator
def worker_label_generator(t):
for worker_id in cand_dict[t.tweet.stable_id]:
yield worker_id, cand_dict[t.tweet.stable_id][worker_id]
np.random.seed(1701)
labeler = LabelAnnotator(label_generator=worker_label_generator)
L_train = labeler.apply(split=0)
print(L_train.lf_stats(session))
print("Creat training data done...")
print(" -train data shape", (L_train.shape))
print("Start to train a generative model")
gen_model = GenerativeModel(lf_propensity=True)
gen_model.train(L_train, reg_type=2, reg_param=0.1, epochs=30)
#doing statistics
print(gen_model.learned_lf_stats())
print("Train a genetive model done...!")
def LF_police_at_location_left(c):
if len(police_tags.intersection(get_right_tokens(c[1], window=7))) > 0:
return 1
else:
return 0
LFs = [
LF_crime_detect, LF_location_left_window, LF_police_at_location_left,
LF_location_left_per_vic_window
]
from snorkel.annotations import LabelAnnotator
labeler1 = LabelAnnotator(f=LFs)
np.random.seed(1701)
get_ipython().magic(u'time L_train = labeler1.apply(split=0)')
L_train
L_train = labeler1.load_matrix(session, split=0)
L_train
for docno in session.query(Document).all():
print(docno)
for i in range(L_train.shape[0]):
if (L_train[i, :].toarray()[0][0] == 1.0):
if (session.query(LocationPer).filter(LocationPer.split == 0)
[i].get_parent().get_parent() == docno):
print(
