def generate_sent1(description_id, label):
"""
Create description from text inputs and save description into database.
"""
# getting the description inputs
title, cat, features = format_inputs(Create.query.get(description_id))
# inputs to LM
if 'shoes' in cat.lower():
model = 'perciapp/models/unisex_shoes_description_beginning'
args['model_name_or_path'] = model
elif cat.lower().startswith('men'):
model = 'perciapp/models/mens_clothing_description_beginning'
args['model_name_or_path'] = model
else:
model = 'perciapp/models/womens_clothing_description_beginning'
args['model_name_or_path'] = model
args['seed'] = random.randint(1, 100001)
args['prompt'] = f'<bos> <category> {cat} <features> \
{features} <brand> <model> {title} \t<desc1> '
print(title + ' ' + label + 'generating now')
sent = brand_remove(generate(args)[0], title)
sent = clean_up_sent(sent)
update = Create.query.filter_by(id=description_id).update({label: sent})
db.session.commit()
return description_id
def generate_sent2(description_id, label):
"""
Create description from text inputs and save description into database.
"""
# getting the description inputs
title, cat, features = format_inputs(Create.query.get(description_id))
# inputs to LM
if 'shoes' in cat.lower():
model = 'perciapp/models/unisex_shoes_description_middle'
args['model_name_or_path'] = model
#give shoes model extra length in case it creates new input
args['length'] = 90
elif cat.lower().startswith('men'):
model = 'perciapp/models/mens_clothing_description_middle'
args['model_name_or_path'] = model
else:
model = 'perciapp/models/womens_clothing_description_middle'
args['model_name_or_path'] = model
args['seed'] = random.randint(1, 100001)
args['prompt'] = f'<bos> <category> {cat} <features> \
{features} <brand> <model> {title} \t<middle> '
print(title + ' ' + label + 'generating now')
sent = generate(args)[0]
# If model has started in <features> again, cut out extra input
if '<middle>' in sent:
sent = sent.split('<middle>')[1]
sent = clean_up_sent(sent)
update = Create.query.filter_by(id=description_id).update({label: sent})
db.session.commit()
return description_id
def edit_sent3(id):
"""
Cull sent3 candidates and put best candidate back into database
"""
# initialize model
from transformers import GPT2Tokenizer, GPT2LMHeadModel
model = GPT2LMHeadModel.from_pretrained('perciapp/models/openai_gpt')
model.eval()
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
import time
start = time.time()
# repeatedly pull candidates until all have been generated
description = Create.query.get(id)
candidates = [
description.sent3, description.sent3_2, description.sent3_3,
description.sent3_4, description.sent3_5, description.sent3_6,
description.sent3_7, description.sent3_8, description.sent3_9,
description.sent3_10, description.sent3_11, description.sent3_12,
description.sent3_13, description.sent3_14, description.sent3_15
]
while None in candidates:
time.sleep(5)
db.session.commit()
description = Create.query.get(id)
candidates = [
description.sent3, description.sent3_2, description.sent3_3,
description.sent3_4, description.sent3_5, description.sent3_6,
description.sent3_7, description.sent3_8, description.sent3_9,
description.sent3_10, description.sent3_11, description.sent3_12,
description.sent3_13, description.sent3_14, description.sent3_15
]
now = time.time()
print(description.title + ' edit_sent3:' + str(int(now - start)))
print('None number: ' + str(candidates.count(None)))
print(candidates)
if now - start > 240:
break
#remove Nones from candidate list
descriptions = list(filter(None, candidates))
print(description.title + ' edit_sent3 starting now')
print(description.title + ' sentences:')
for sentence in descriptions:
print(sentence)
print()
descriptions = remove_bad_sentences(descriptions)
scores = [score(i, tokenizer, model) for i in descriptions]
print('scores/descriptions after edit:')
for i, (desc, scr) in enumerate(zip(descriptions, scores)):
try:
print(int(scr), desc)
except ValueError:
print('ValueError!')
print(scr)
print(desc)
# getting the description inputs
title, cat, features = format_inputs(description)
candidate1, descriptions, scores = pop_best_sentence(descriptions, scores)
print('candidate1:')
print(candidate1)
candidate2, descriptions, scores = pop_best_sentence(descriptions, scores)
print('candidate2:')
print(candidate2)
candidate3, descriptions, scores = pop_best_sentence(descriptions, scores)
print('candidate3:')
print(candidate3)
sent = return_most_similar(features,
[candidate1, candidate2, candidate3])[0]
print('winner:')
print(sent)
update = Create.query.filter_by(id=id).update({'sent3_winner': sent})
db.session.commit()
description = Create.query.get(id)
winners = [description.sent1_winner, description.sent2_winner]
while None in winners:
time.sleep(5)
db.session.commit()
description = Create.query.get(id)
winners = [description.sent1_winner, description.sent2_winner]
final_output = period_check(description.sent1_winner) + period_check(
description.sent2_winner) + period_check(description.sent3_winner)
update = Create.query.filter_by(id=id).update(
{'description': final_output})
db.session.commit()
print()
print(description.title + ' edit_sent3 complete')
return id
def edit_sent1(id):
"""
Cull sent1 candidates and put best candidate back into database
"""
# initialize model
from transformers import GPT2Tokenizer, GPT2LMHeadModel
model = GPT2LMHeadModel.from_pretrained('perciapp/models/openai_gpt')
model.eval()
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
import time
start = time.time()
# repeatedly pull candidates until all have been generated
description = Create.query.get(id)
candidates = [
description.sent1, description.sent1_2, description.sent1_3,
description.sent1_4, description.sent1_5, description.sent1_6,
description.sent1_7, description.sent1_8, description.sent1_9,
description.sent1_10, description.sent1_11, description.sent1_12,
description.sent1_13, description.sent1_14, description.sent1_15
]
while None in candidates:
time.sleep(5)
db.session.commit()
description = Create.query.get(id)
candidates = [
description.sent1, description.sent1_2, description.sent1_3,
description.sent1_4, description.sent1_5, description.sent1_6,
description.sent1_7, description.sent1_8, description.sent1_9,
description.sent1_10, description.sent1_11, description.sent1_12,
description.sent1_13, description.sent1_14, description.sent1_15
]
now = time.time()
print(description.title + ' edit_sent1:' + str(int(now - start)))
print('None number: ' + str(candidates.count(None)))
print(candidates)
now = time.time()
if now - start > 240:
break
#remove Nones from candidate list
descriptions = list(filter(None, candidates))
print(description.title + ' edit_sent1 starting now')
print(description.title + ' sentences:')
for sentence in descriptions:
print(sentence)
print()
descriptions = remove_bad_sentences(descriptions)
# remove 1st sentences without the product title in them
for item in descriptions:
if str(description.title) not in item:
descriptions.remove(item)
scores = [score(i, tokenizer, model) for i in descriptions]
print(description.title + ' scores/descriptions after edit:')
for i, (desc, scr) in enumerate(zip(descriptions, scores)):
try:
print(int(scr), desc)
except ValueError:
print('ValueError!')
print(scr)
print(desc)
# getting the description inputs
title, cat, features = format_inputs(description)
# pulling the best written sentence
candidate1, descriptions, scores = pop_best_sentence(descriptions, scores)
print('candidate1:')
print(candidate1)
# pulling next best written sentence
candidate2, descriptions, scores = pop_best_sentence(descriptions, scores)
print('candidate2:')
print(candidate2)
candidate3, descriptions, scores = pop_best_sentence(descriptions, scores)
print('candidate3:')
print(candidate3)
# picking the one which is most similar to the features list
first_sent = return_most_similar(features,
[candidate1, candidate2, candidate3])[0]
print('winner:')
print(first_sent)
# saving the best first sentence
update = Create.query.filter_by(id=id).update({'sent1_winner': first_sent})
db.session.commit()
print()
print(description.title + ' edit_sent1 complete')
return id