def save(self, directory: str):
shutil.copyfile(self.model_file, os.path.join(directory, 'dan.pt'))
shell(f'rm -f {self.model_file}')
with open(os.path.join(directory, 'dan.pkl'), 'wb') as f:
cloudpickle.dump({
'page_field': self.page_field,
'combined_text_field': self.text_field,
'unigram_text_field': self.unigram_field,
'bigram_text_field': self.bigram_field,
'trigram_text_field': self.trigram_field,
'combined_ngrams': self.combined_ngrams,
'unigrams': self.unigrams,
'bigrams': self.bigrams,
'trigrams': self.trigrams,
'combined_max_vocab_size': self.combined_max_vocab_size,
'unigram_max_vocab_size': self.unigram_max_vocab_size,
'bigram_max_vocab_size': self.bigram_max_vocab_size,
'trigram_max_vocab_size': self.trigram_max_vocab_size,
'qanta_id_field': self.qanta_id_field,
'n_classes': self.n_classes,
'gradient_clip': self.gradient_clip,
'n_hidden_units': self.n_hidden_units,
'n_hidden_layers': self.n_hidden_layers,
'nn_dropout': self.nn_dropout,
'batch_size': self.batch_size,
'use_wiki': self.use_wiki,
'n_wiki_sentences': self.n_wiki_sentences,
'wiki_title_replace_token': self.wiki_title_replace_token,
'lowercase': self.lowercase,
'pooling': self.pooling,
'random_seed': self.random_seed,
'config_num': self.config_num
}, f)
def slurm(
partition,
qos,
mem_per_cpu,
max_time,
nodelist,
cpus_per_task,
luigi_module,
luigi_task,
):
env = Environment(loader=PackageLoader("qanta", "slurm/templates"))
template = env.get_template("luigi-template.sh.jinja2")
sbatch_script = template.render({
"luigi_module": luigi_module,
"luigi_task": luigi_task,
"partition": partition,
"qos": qos,
"mem_per_cpu": mem_per_cpu,
"max_time": max_time,
"nodelist": nodelist,
"cpus_per_task": cpus_per_task,
})
tmp_file = get_tmp_filename()
with open(tmp_file, "w") as f:
f.write(sbatch_script)
shell(f"sbatch {tmp_file}")
shell(f"rm -f {tmp_file}")
def load(cls, directory: str) -> AbstractGuesser:
guesser = DANGuesser()
embeddings, embedding_lookup = _load_embeddings(root_directory=directory)
guesser.embeddings = embeddings
guesser.embedding_lookup = embedding_lookup
params_path = os.path.join(directory, DEEP_DAN_PARAMS_TARGET)
with open(params_path, 'rb') as f:
params = pickle.load(f)
guesser.max_len = params['max_len']
guesser.class_to_i = params['class_to_i']
guesser.i_to_class = params['i_to_class']
guesser.vocab = params['vocab']
guesser.n_classes = params['n_classes']
if (guesser.max_len is None
or guesser.class_to_i is None
or guesser.i_to_class is None
or guesser.vocab is None
or guesser.n_classes is None):
raise ValueError('Attempting to load uninitialized model parameters')
model_path = os.path.join(directory, DEEP_DAN_MODEL_TARGET)
shell('cp -r {} {}'.format(model_path, safe_path(DEEP_DAN_MODEL_TMP_DIR)))
we_path = os.path.join(directory, TF_DAN_WE)
shutil.copyfile(TF_DAN_WE_TMP, we_path)
return guesser
def save(self, directory: str):
shutil.copyfile(self.model_file, os.path.join(directory, 'dan.pt'))
shell(f'rm -f {self.model_file}')
with open(os.path.join(directory, 'dan.pkl'), 'wb') as f:
cloudpickle.dump(
{
'page_field': self.page_field,
'combined_text_field': self.text_field,
'unigram_text_field': self.unigram_field,
'bigram_text_field': self.bigram_field,
'trigram_text_field': self.trigram_field,
'combined_ngrams': self.combined_ngrams,
'unigrams': self.unigrams,
'bigrams': self.bigrams,
'trigrams': self.trigrams,
'combined_max_vocab_size': self.combined_max_vocab_size,
'unigram_max_vocab_size': self.unigram_max_vocab_size,
'bigram_max_vocab_size': self.bigram_max_vocab_size,
'trigram_max_vocab_size': self.trigram_max_vocab_size,
'qanta_id_field': self.qanta_id_field,
'n_classes': self.n_classes,
'gradient_clip': self.gradient_clip,
'n_hidden_units': self.n_hidden_units,
'n_hidden_layers': self.n_hidden_layers,
'nn_dropout': self.nn_dropout,
'batch_size': self.batch_size,
'use_wiki': self.use_wiki,
'n_wiki_sentences': self.n_wiki_sentences,
'wiki_title_replace_token': self.wiki_title_replace_token,
'lowercase': self.lowercase,
'pooling': self.pooling,
'random_seed': self.random_seed,
'config_num': self.config_num
}, f)
def run(self):
if is_aws_authenticated():
s3_location = 's3://pinafore-us-west-2/public/wiki_redirects.csv'
shell('aws s3 cp {} {}'.format(s3_location, ALL_WIKI_REDIRECTS))
else:
https_location = 'https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/wiki_redirects.csv'
shell('wget -O {} {}'.format(ALL_WIKI_REDIRECTS, https_location))
def run(self):
with open(QANTA_MAPPED_DATASET_PATH) as f:
qanta_questions = json.load(f)['questions']
tmp_db = get_tmp_filename()
questions_to_sqlite(qanta_questions, tmp_db)
shell(f'mv {tmp_db} {QANTA_SQL_DATASET_PATH}')
def run(self):
with open(QANTA_MAPPED_DATASET_PATH) as f:
qanta_questions = json.load(f)['questions']
tmp_db = get_tmp_filename()
questions_to_sqlite(qanta_questions, tmp_db)
shell(f'mv {tmp_db} {QANTA_SQL_DATASET_PATH}')
def run(self):
safe_path(WIKI_DISAMBIGUATION_PAGES)
if is_aws_authenticated():
s3_location = 's3://pinafore-us-west-2/public/disambiguation_pages.json'
shell('aws s3 cp {} {}'.format(s3_location, WIKI_DISAMBIGUATION_PAGES))
else:
https_location = 'https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/disambiguation_pages.json'
shell('wget -O {} {}'.format(WIKI_DISAMBIGUATION_PAGES, https_location))
def run(self):
safe_path(ALL_WIKI_REDIRECTS)
if is_aws_authenticated():
s3_location = 's3://pinafore-us-west-2/public/wiki_redirects.csv'
shell('aws s3 cp {} {}'.format(s3_location, ALL_WIKI_REDIRECTS))
else:
https_location = 'https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/wiki_redirects.csv'
shell('wget -O {} {}'.format(ALL_WIKI_REDIRECTS, https_location))
def run(self):
safe_path(ALL_WIKI_REDIRECTS)
if is_aws_authenticated():
s3_location = "s3://pinafore-us-west-2/public/wiki_redirects.csv"
shell("aws s3 cp {} {}".format(s3_location, ALL_WIKI_REDIRECTS))
else:
https_location = "https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/wiki_redirects.csv"
shell("wget -O {} {}".format(ALL_WIKI_REDIRECTS, https_location))
def run(self):
safe_path(WIKI_DISAMBIGUATION_PAGES)
if is_aws_authenticated():
s3_location = "s3://pinafore-us-west-2/public/disambiguation_pages.json"
shell("aws s3 cp {} {}".format(s3_location, WIKI_DISAMBIGUATION_PAGES))
else:
https_location = "https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/disambiguation_pages.json"
shell("wget -O {} {}".format(WIKI_DISAMBIGUATION_PAGES, https_location))
def run(self):
make_dirs('output/predictions/')
shell(
('vw -t '
'-d output/vw_input/{fold}.vw.txt '
'--loss_function logistic '
'-i {vw_model} --audit '
'| python cli.py format_vw_audit '
'> output/predictions/{fold}.audit').format(fold='test', vw_model=c.VW_MODEL)
)
def save(self, directory: str) -> None:
shutil.copyfile(self.model_file, os.path.join(directory, 'elmo.pt'))
shell(f'rm -f {self.model_file}')
with open(os.path.join(directory, 'elmo.pkl'), 'wb') as f:
cloudpickle.dump({
'class_to_i': self.class_to_i,
'i_to_class': self.i_to_class,
'config_num': self.config_num,
'random_seed': self.random_seed,
'dropout': self.dropout
}, f)
def save(self, directory: str) -> None:
shutil.copyfile(self.model_file, os.path.join(directory, 'elmo.pt'))
shell(f'rm -f {self.model_file}')
with open(os.path.join(directory, 'elmo.pkl'), 'wb') as f:
cloudpickle.dump({
'class_to_i': self.class_to_i,
'i_to_class': self.i_to_class,
'config_num': self.config_num,
'random_seed': self.random_seed,
'dropout': self.dropout
}, f)
def run(self):
if is_aws_authenticated():
s3_location = 's3://pinafore-us-west-2/public/wikipedia-dumps/parsed-wiki.tar.lz4'
shell('aws s3 cp {} data/external/wikipedia/parsed-wiki.tar.lz4'.format(s3_location))
else:
https_location = 'https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/wikipedia-dumps/parsed-wiki.tar.lz4'
shell('wget -O {} {}'.format('data/external/wikipedia/parsed-wiki.tar.lz4', https_location))
shell('lz4 -d data/external/wikipedia/parsed-wiki.tar.lz4 | tar -x -C data/external/wikipedia/')
shell('rm data/external/wikipedia/parsed-wiki.tar.lz4')
shell('touch data/external/wikipedia/parsed-wiki_SUCCESS')
def run(self):
archive = safe_path('data/external/wikipedia/parsed-wiki.tar.lz4')
if is_aws_authenticated():
s3_location = f's3://pinafore-us-west-2/public/parsed-wiki.tar.lz4'
shell(f'aws s3 cp {s3_location} {archive}')
else:
https_location = 'https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/parsed-wiki.tar.lz4'
shell(f'wget -O {archive} {https_location}')
shell(f'lz4 -d {archive} | tar -x -C data/external/wikipedia/')
shell(f'rm {archive}')
shell('touch data/external/wikipedia/parsed-wiki_SUCCESS')
def run(self):
archive = safe_path("data/external/wikipedia/parsed-wiki.tar.lz4")
if is_aws_authenticated():
s3_location = f"s3://pinafore-us-west-2/public/parsed-wiki.tar.lz4"
shell(f"aws s3 cp {s3_location} {archive}")
else:
https_location = "https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/parsed-wiki.tar.lz4"
shell(f"wget -O {archive} {https_location}")
shell(f"lz4 -d {archive} | tar -x -C data/external/wikipedia/")
shell(f"rm {archive}")
shell("touch data/external/wikipedia/parsed-wiki_SUCCESS")
def run(self):
archive = safe_path('data/external/wikipedia/parsed-wiki.tar.lz4')
if is_aws_authenticated():
s3_location = f's3://pinafore-us-west-2/public/parsed-wiki.tar.lz4'
shell(f'aws s3 cp {s3_location} {archive}')
else:
https_location = 'https://s3-us-west-2.amazonaws.com/pinafore-us-west-2/public/parsed-wiki.tar.lz4'
shell(f'wget -O {archive} {https_location}')
shell(f'lz4 -d {archive} | tar -x -C data/external/wikipedia/')
shell(f'rm {archive}')
shell('touch data/external/wikipedia/parsed-wiki_SUCCESS')
def save(self, directory: str) -> None:
shutil.copyfile(self.model_file, os.path.join(directory, "elmo.pt"))
shell(f"rm -f {self.model_file}")
with open(os.path.join(directory, "elmo.pkl"), "wb") as f:
cloudpickle.dump(
{
"class_to_i": self.class_to_i,
"i_to_class": self.i_to_class,
"config_num": self.config_num,
"random_seed": self.random_seed,
"dropout": self.dropout,
},
f,
)
def guess(
self, questions: List[QuestionText],
max_n_guesses: Optional[int]) -> List[List[Tuple[Answer, float]]]:
with open('/tmp/vw_test.txt', 'w') as f:
for q in questions:
features = format_question(q)
f.write('1 |words {features}\n'.format(features=features))
shell(
'vw -t -i /tmp/vw_guesser.model -p /tmp/predictions.txt -d /tmp/vw_test.txt'
)
predictions = []
with open('/tmp/predictions.txt') as f:
for line in f:
label = int(line)
predictions.append([(self.i_to_label[label], 0)])
return predictions
def guess(self, questions: List[QuestionText],
max_n_guesses: Optional[int]) -> List[List[Tuple[Page, float]]]:
temp_dir = get_tmp_dir()
with tempfile.NamedTemporaryFile("w", delete=False, dir=temp_dir) as f:
file_name = f.name
for q in questions:
features = format_question(q)
f.write(f"1 |words {features}\n")
shell(
f"vw -t -i {self.model_file} -p {file_name}_preds -d {file_name}")
predictions = []
with open(f"{file_name}_preds") as f:
for line in f:
label = int(line)
predictions.append([(self.i_to_label[label], 0)])
shell(f"rm -f {file_name}.preds {file_name}")
return predictions
def guess(self,
questions: List[QuestionText],
max_n_guesses: Optional[int]) -> List[List[Tuple[Page, float]]]:
temp_dir = get_tmp_dir()
with tempfile.NamedTemporaryFile('w', delete=False, dir=temp_dir) as f:
file_name = f.name
for q in questions:
features = format_question(q)
f.write(f'1 |words {features}\n')
shell(f'vw -t -i {self.model_file} -p {file_name}_preds -d {file_name}')
predictions = []
with open(f'{file_name}_preds') as f:
for line in f:
label = int(line)
predictions.append([(self.i_to_label[label], 0)])
shell(f'rm -f {file_name}.preds {file_name}')
return predictions
def train(self, training_data: TrainingData) -> None:
questions = training_data[0]
answers = set(training_data[1])
x_data = []
y_data = []
for q, ans in zip(questions, answers):
for sent in q:
x_data.append(sent)
y_data.append(ans)
label_set = set(answers)
self.label_to_i = {label: i for i, label in enumerate(label_set, 1)}
self.i_to_label = {i: label for label, i in self.label_to_i.items()}
self.max_label = len(self.label_to_i)
with open('/tmp/vw_train.txt', 'w') as f:
zipped = list(zip(x_data, y_data))
random.shuffle(zipped)
for x, y in zipped:
features = format_question(x)
label = self.label_to_i[y]
f.write('{label} |words {features}\n'.format(
label=label, features=features))
if self.multiclass_online_trees:
multiclass_flag = '--log_multi'
elif self.multiclass_one_against_all:
multiclass_flag = '--oaa'
else:
raise ValueError(
'The options multiclass_one_against_all and multiclass_online_trees are XOR'
)
shell(
'vw -k {multiclass_flag} {max_label} -d /tmp/vw_train.txt -f /tmp/vw_guesser.model --loss_function '
'logistic --ngram 1 --ngram 2 --skips 1 -c --passes {passes} -b {bits} '
'--l1 {l1} --l2 {l2} -l {learning_rate} --decay_learning_rate {decay_learning_rate}'
.format(max_label=self.max_label,
multiclass_flag=multiclass_flag,
bits=self.bits,
l1=self.l1,
l2=self.l2,
passes=self.passes,
learning_rate=self.learning_rate,
decay_learning_rate=self.decay_learning_rate))
def run(self):
s3_location = 's3://pinafore-us-west-2/public/wikipedia-dumps/parsed-wiki.tar.lz4'
shell(
'aws s3 cp {} data/external/wikipedia/parsed-wiki.tar.lz4'.format(
s3_location))
shell(
'lz4 -d data/external/wikipedia/parsed-wiki.tar.lz4 | tar -x -C data/external/wikipedia/'
)
shell('rm data/external/wikipedia/parsed-wiki.tar.lz4')
shell('touch data/external/wikipedia/parsed-wiki_SUCCESS')
def slurm(partition, qos, mem_per_cpu, max_time, nodelist, cpus_per_task, luigi_module, luigi_task):
env = Environment(loader=PackageLoader('qanta', 'slurm/templates'))
template = env.get_template('luigi-template.sh.jinja2')
sbatch_script = template.render({
'luigi_module': luigi_module,
'luigi_task': luigi_task,
'partition': partition,
'qos': qos,
'mem_per_cpu': mem_per_cpu,
'max_time': max_time,
'nodelist': nodelist,
'cpus_per_task': cpus_per_task
})
tmp_file = get_tmp_filename()
with open(tmp_file, 'w') as f:
f.write(sbatch_script)
shell(f'sbatch {tmp_file}')
shell(f'rm -f {tmp_file}')
def load(cls, directory: str):
model_path = os.path.join(directory, 'vw_guesser.model')
shell('cp {} /tmp/vw_guesser.model'.format(model_path))
data_pickle_path = os.path.join(directory, 'vw_guesser.pickle')
with open(data_pickle_path, 'rb') as f:
data = pickle.load(f)
guesser = VWGuesser()
guesser.label_to_i = data['label_to_i']
guesser.i_to_label = data['i_to_label']
guesser.max_label = data['max_label']
guesser.multiclass_one_against_all = data['multiclass_one_against_all']
guesser.multiclass_online_trees = data['multiclass_online_trees']
guesser.l1 = data['l1']
guesser.l2 = data['l2']
guesser.passes = data['passes']
guesser.learning_rate = data['learning_rate']
guesser.decay_learning_rate = data['decay_learning_rate']
guesser.bits = data['bits']
return guesser
def slurm(partition, qos, mem_per_cpu, max_time, nodelist, cpus_per_task,
luigi_module, luigi_task):
env = Environment(loader=PackageLoader('qanta', 'slurm/templates'))
template = env.get_template('luigi-template.sh.jinja2')
sbatch_script = template.render({
'luigi_module': luigi_module,
'luigi_task': luigi_task,
'partition': partition,
'qos': qos,
'mem_per_cpu': mem_per_cpu,
'max_time': max_time,
'nodelist': nodelist,
'cpus_per_task': cpus_per_task
})
tmp_file = get_tmp_filename()
with open(tmp_file, 'w') as f:
f.write(sbatch_script)
shell(f'sbatch {tmp_file}')
shell(f'rm -f {tmp_file}')
def save(self, directory: str) -> None:
model_path = os.path.join(directory, 'vw_guesser.model')
shell('cp /tmp/vw_guesser.model {}'.format(model_path))
data = {
'label_to_i': self.label_to_i,
'i_to_label': self.i_to_label,
'max_label': self.max_label,
'multiclass_one_against_all': self.multiclass_one_against_all,
'multiclass_online_trees': self.multiclass_online_trees,
'l1': self.l1,
'l2': self.l2,
'passes': self.passes,
'learning_rate': self.learning_rate,
'decay_learning_rate': self.decay_learning_rate,
'bits': self.bits
}
data_pickle_path = os.path.join(directory, 'vw_guesser.pickle')
with open(data_pickle_path, 'wb') as f:
pickle.dump(data, f)
def run(self):
shell('rm -rf {}'.format(c.WIKIFIER_OUTPUT_TARGET))
shell('mkdir -p {}'.format(c.WIKIFIER_OUTPUT_TARGET))
command = (
'(cd data/external/Wikifier2013 '
'&& java -Xmx10G -jar dist/wikifier-3.0-jar-with-dependencies.jar '
'-annotateData '
'../../../{} '
'../../../{} '
'false configs/STAND_ALONE_NO_INFERENCE.xml)')
shell(command.format(c.WIKIFIER_INPUT_TARGET,
c.WIKIFIER_OUTPUT_TARGET))
shell('touch {}/_SUCCESS'.format(c.WIKIFIER_OUTPUT_TARGET))
def save(self, directory: str) -> None:
params_path = os.path.join(directory, DEEP_DAN_PARAMS_TARGET)
with safe_open(params_path, 'wb') as f:
if (self.max_len is None
or self.class_to_i is None
or self.i_to_class is None
or self.vocab is None
or self.n_classes is None):
raise ValueError('Attempting to save uninitialized model parameters')
pickle.dump({
'max_len': self.max_len,
'class_to_i': self.class_to_i,
'i_to_class': self.i_to_class,
'vocab': self.vocab,
'n_classes': self.n_classes
}, f)
model_path = os.path.join(directory, DEEP_DAN_MODEL_TARGET)
shell('cp -r {} {}'.format(DEEP_DAN_MODEL_TMP_DIR, safe_path(model_path)))
we_path = os.path.join(directory, TF_DAN_WE)
shutil.copyfile(TF_DAN_WE_TMP, safe_path(we_path))
def run(self):
guesser_class = get_class(self.guesser_module, self.guesser_class)
reporting_directory = AbstractGuesser.reporting_path(
self.guesser_module, self.guesser_class, self.config_num, "")
# In the cases of huge parameter sweeps on SLURM its easy to accidentally run out of /fs/ storage.
# Since we only care about the results we can get them, then delete the models. We can use the regular
# GuesserReport to preserve the model
guesser_directory = AbstractGuesser.output_path(
self.guesser_module, self.guesser_class, self.config_num, "")
param_path = AbstractGuesser.output_path(
self.guesser_module,
self.guesser_class,
self.config_num,
f"guesser_params.pickle",
)
guesses_files = []
if os.path.exists(c.QANTA_EXPO_DATASET_PATH):
folds = [c.GUESSER_DEV_FOLD, c.GUESSER_TEST_FOLD, c.EXPO_FOLD]
else:
folds = [c.GUESSER_DEV_FOLD, c.GUESSER_TEST_FOLD]
for f in folds:
guesses_files.extend([
f"guesses_char_{f}.pickle",
f"guesses_full_{f}.pickle",
f"guesses_first_{f}.pickle",
])
guesses_paths = [
AbstractGuesser.output_path(self.guesser_module,
self.guesser_class, self.config_num, f)
for f in guesses_files
]
log.info(f'Running: "cp {param_path} {reporting_directory}"')
shell(f"cp {param_path} {reporting_directory}")
for g_path in guesses_paths:
log.info(f'Running: "cp {g_path} {reporting_directory}"')
shell(f"cp {g_path} {reporting_directory}")
guesser_instance = guesser_class(self.config_num)
for f in folds:
guesser_instance.create_report(reporting_directory, f)
log.info(f'Running: "rm -rf {guesser_directory}"')
shell(f"rm -rf {guesser_directory}")
for g_path in guesses_paths:
shell(f"rm -f {g_path}")
def run_guesser(n_times, workers, guesser_qualified_class):
for _ in range(n_times):
if 'qanta.guesser' not in guesser_qualified_class:
log.error('qanta.guesser not found in guesser_qualified_class, this is likely an error, exiting.')
return
shell('rm -rf /tmp/qanta')
shell(f'rm -rf output/guesser/{guesser_qualified_class}')
shell(f'luigi --local-scheduler --module qanta.pipeline.guesser --workers {workers} AllSingleGuesserReports')
def save(self, directory: str):
shutil.copyfile(self.model_file, os.path.join(directory, "dan.pt"))
shell(f"rm -f {self.model_file}")
with open(os.path.join(directory, "dan.pkl"), "wb") as f:
cloudpickle.dump(
{
"page_field": self.page_field,
"combined_text_field": self.text_field,
"unigram_text_field": self.unigram_field,
"bigram_text_field": self.bigram_field,
"trigram_text_field": self.trigram_field,
"combined_ngrams": self.combined_ngrams,
"unigrams": self.unigrams,
"bigrams": self.bigrams,
"trigrams": self.trigrams,
"combined_max_vocab_size": self.combined_max_vocab_size,
"unigram_max_vocab_size": self.unigram_max_vocab_size,
"bigram_max_vocab_size": self.bigram_max_vocab_size,
"trigram_max_vocab_size": self.trigram_max_vocab_size,
"qanta_id_field": self.qanta_id_field,
"n_classes": self.n_classes,
"gradient_clip": self.gradient_clip,
"n_hidden_units": self.n_hidden_units,
"n_hidden_layers": self.n_hidden_layers,
"nn_dropout": self.nn_dropout,
"batch_size": self.batch_size,
"use_wiki": self.use_wiki,
"n_wiki_sentences": self.n_wiki_sentences,
"wiki_title_replace_token": self.wiki_title_replace_token,
"lowercase": self.lowercase,
"pooling": self.pooling,
"random_seed": self.random_seed,
"config_num": self.config_num,
},
f,
)
def save(self, directory: str) -> None:
model_path = safe_path(os.path.join(directory, "vw_guesser.vw"))
shell(f"mv {self.model_file}.vw {model_path}")
self.model_file = model_path
data = {
"label_to_i": self.label_to_i,
"i_to_label": self.i_to_label,
"max_label": self.max_label,
"multiclass_one_against_all": self.multiclass_one_against_all,
"multiclass_online_trees": self.multiclass_online_trees,
"l1": self.l1,
"l2": self.l2,
"passes": self.passes,
"learning_rate": self.learning_rate,
"decay_learning_rate": self.decay_learning_rate,
"bits": self.bits,
"ngrams": self.ngrams,
"skips": self.skips,
"config_num": self.config_num,
"random_seed": self.random_seed,
}
data_pickle_path = os.path.join(directory, "vw_guesser.pickle")
with open(data_pickle_path, "wb") as f:
pickle.dump(data, f)
def save(self, directory: str) -> None:
model_path = safe_path(os.path.join(directory, 'vw_guesser.vw'))
shell(f'mv {self.model_file}.vw {model_path}')
self.model_file = model_path
data = {
'label_to_i': self.label_to_i,
'i_to_label': self.i_to_label,
'max_label': self.max_label,
'multiclass_one_against_all': self.multiclass_one_against_all,
'multiclass_online_trees': self.multiclass_online_trees,
'l1': self.l1,
'l2': self.l2,
'passes': self.passes,
'learning_rate': self.learning_rate,
'decay_learning_rate': self.decay_learning_rate,
'bits': self.bits,
'ngrams': self.ngrams,
'skips': self.skips,
'config_num': self.config_num,
'random_seed': self.random_seed
}
data_pickle_path = os.path.join(directory, 'vw_guesser.pickle')
with open(data_pickle_path, 'wb') as f:
pickle.dump(data, f)
def run_guesser(n_times, workers, guesser_qualified_class):
for _ in range(n_times):
if 'qanta.guesser' not in guesser_qualified_class:
log.error(
'qanta.guesser not found in guesser_qualified_class, this is likely an error, exiting.'
)
return
shell('rm -rf /tmp/qanta')
shell(f'rm -rf output/guesser/{guesser_qualified_class}')
shell(
f'luigi --local-scheduler --module qanta.pipeline.guesser --workers {workers} AllSingleGuesserReports'
)
def run(self):
guesser_types = set()
for g_spec in AbstractGuesser.list_enabled_guessers():
guesser_types.add(
f"{g_spec.guesser_module}.{g_spec.guesser_class}")
_, _, all_dfs, _ = merge_reports(guesser_types)
best_guessers = find_best_guessers(all_dfs)
for g, config_num in best_guessers.items():
inp = f"output/guesser/{g}/{config_num}"
out = f"output/guesser/best/{g}/"
shell(f"touch {inp}/best.touch")
shell(f"mkdir -p {out}")
shell(f"cp -r {inp}/* {out}")
def generate_guesser_slurm(slurm_config_file, task, output_dir):
with open(slurm_config_file) as f:
slurm_config = yaml.load(f)
default_slurm_config = slurm_config['default']
env = Environment(loader=PackageLoader('qanta', 'slurm/templates'))
template = env.get_template('guesser-luigi-template.sh')
enabled_guessers = list(AbstractGuesser.list_enabled_guessers())
for i, gs in enumerate(enabled_guessers):
if gs.guesser_class == 'ElasticSearchGuesser':
raise ValueError('ElasticSearchGuesser is not compatible with slurm')
elif gs.guesser_class in slurm_config:
guesser_slurm_config = slurm_config[gs.guesser_class]
else:
guesser_slurm_config = None
partition = get_slurm_config_value('partition', default_slurm_config, guesser_slurm_config)
qos = get_slurm_config_value('qos', default_slurm_config, guesser_slurm_config)
mem_per_cpu = get_slurm_config_value('mem_per_cpu', default_slurm_config, guesser_slurm_config)
gres = get_slurm_config_value('gres', default_slurm_config, guesser_slurm_config)
max_time = get_slurm_config_value('max_time', default_slurm_config, guesser_slurm_config)
cpus_per_task = get_slurm_config_value('cpus_per_task', default_slurm_config, guesser_slurm_config)
account = get_slurm_config_value('account', default_slurm_config, guesser_slurm_config)
if task == 'GuesserReport':
folds = GUESSER_GENERATION_FOLDS
else:
folds = []
script = template.render({
'task': task,
'guesser_module': gs.guesser_module,
'guesser_class': gs.guesser_class,
'dependency_module': gs.dependency_module,
'dependency_class': gs.dependency_class,
'config_num': gs.config_num,
'partition': partition,
'qos': qos,
'mem_per_cpu': mem_per_cpu,
'max_time': max_time,
'gres': gres,
'cpus_per_task': cpus_per_task,
'account': account,
'folds': folds
})
slurm_file = path.join(output_dir, f'slurm-{i}.sh')
with safe_open(slurm_file, 'w') as f:
f.write(script)
singleton_path = 'qanta/slurm/templates/guesser-singleton.sh'
singleton_output = path.join(output_dir, 'guesser-singleton.sh')
shell(f'cp {singleton_path} {singleton_output}')
master_template = env.get_template('guesser-master-template.sh')
master_script = master_template.render({
'script_list': [
path.join(output_dir, f'slurm-{i}.sh') for i in range(len(enabled_guessers))
] + [singleton_output],
'gres': gres,
'partition': partition,
'qos': qos,
'mem_per_cpu': mem_per_cpu,
'max_time': max_time,
'gres': gres,
'cpus_per_task': cpus_per_task,
'account': account
})
with safe_open(path.join(output_dir, 'slurm-master.sh'), 'w') as f:
f.write(master_script)
def run(self):
s3_location = 's3://entilzha-us-west-2/wikidata/wikidata-claims_instance-of.jsonl'
shell('aws s3 cp {} {}'.format(s3_location, WIKIDATA_CLAIMS))
def train(self, training_data: TrainingData) -> None:
log.info(f'Config:\n{pformat(self.parameters())}')
questions = training_data[0]
answers = training_data[1]
x_data = []
y_data = []
for q, ans in zip(questions, answers):
for sent in q:
x_data.append(sent)
y_data.append(ans)
label_set = set(answers)
self.label_to_i = {label: i for i, label in enumerate(label_set, 1)}
self.i_to_label = {i: label for label, i in self.label_to_i.items()}
self.max_label = len(self.label_to_i)
temp_dir = get_tmp_dir()
with tempfile.NamedTemporaryFile('w', delete=False, dir=temp_dir) as f:
file_name = f.name
zipped = list(zip(x_data, y_data))
random.shuffle(zipped)
for x, y in zipped:
features = format_question(x)
label = self.label_to_i[y]
f.write('{label} |words {features}\n'.format(label=label, features=features))
if self.multiclass_online_trees:
multiclass_flag = '--log_multi'
elif self.multiclass_one_against_all:
multiclass_flag = '--oaa'
else:
raise ValueError('The options multiclass_one_against_all and multiclass_online_trees are XOR')
self.model_file = get_tmp_filename()
options = [
'vw',
'-k',
f'{multiclass_flag}',
f'{self.max_label}',
f'-d {file_name}',
f'-f {self.model_file}.vw',
'--loss_function logistic',
'-c',
f'--passes {self.passes}',
f'-b {self.bits}',
f'-l {self.learning_rate}',
f'--decay_learning_rate {self.decay_learning_rate}',
f'--random_seed {self.random_seed}'
]
for n in self.ngrams:
options.append(f'--ngram {n}')
for n in self.skips:
options.append(f'--skips {n}')
if self.l1 != 0:
options.append(f'--l1 {self.l1}')
if self.l2 != 0:
options.append(f'--l2 {self.l2}')
command = ' '.join(options)
log.info(f'Running:\n{command}')
try:
shell(command)
finally:
shell(f'rm -f {file_name} {file_name}.cache')
def run(self):
shell('rm -rf {}'.format(c.WIKIFIER_INPUT_TARGET))
shell('mkdir -p {}'.format(c.WIKIFIER_INPUT_TARGET))
shell('python3 cli.py wikify {}/'.format(c.WIKIFIER_INPUT_TARGET))
shell('touch {}/_SUCCESS'.format(c.WIKIFIER_INPUT_TARGET))
def run(self):
shell('pdftk output/reporting/*.pdf cat output /tmp/report.pdf')
shell('mv /tmp/report.pdf output/reporting/report.pdf')
def run(self):
s3_location = 's3://entilzha-us-west-2/wikidata/wikidata-claims_instance-of.jsonl'
shell('aws s3 cp {} {}'.format(s3_location, WIKIDATA_CLAIMS))
def run(self):
tmp_file = get_tmp_filename()
shell(f'wget {self.url} -O {tmp_file}')
shell(f'mv {tmp_file} {self.path}')
shell(f'rm -f {tmp_file}')
