def consume_data(self, data, passback, output_dir):
open_threads = []
_, max_processes, _ = config_parser_singleton.read_execution_options()
max_processes = multiprocessing.cpu_count() if max_processes == -1 else max_processes
with multiprocessing.Pool(max_processes) as pool:
out_dir = output_dir + path.sep + "ann_models"
if not path.exists(out_dir):
os.mkdir(out_dir)
for ticker, training_data in data.items():
open_threads.append(pool.apply_async(handle_data, [ticker, training_data,
out_dir, self._overwite_existing],
{'trend_lookahead': self._trend_lookahead,
'combined_examples': self._combined_examples_factor}))
for t in tqdm.tqdm(open_threads):
t.get()
def consume_data(self, data, passback, output_dir):
out_dir = output_dir + path.sep + 'svm_strength_models'
if not path.exists(out_dir):
os.mkdir(out_dir)
exec_options = config_parser_singleton.read_execution_options()
max_processes = exec_options[1]
max_processes = multiprocessing.cpu_count(
) if max_processes == -1 else max_processes
with multiprocessing.Pool(max_processes) as pool:
open_jobs = []
for ticker, training_data in data.items():
job = pool.apply_async(
handle_data,
[ticker, training_data, out_dir, self._overwrite_existing],
{'combined_examples': self._combined_examples_factor})
open_jobs.append(job)
for job in tqdm.tqdm(open_jobs):
job.get()
def consume_data(self, data, passback, output_dir):
out_dir = output_dir + path.sep + 'random_forest_models'
if not path.exists(out_dir):
os.mkdir(out_dir)
_, max_processes, _ = config_parser_singleton.read_execution_options()
max_processes = multiprocessing.cpu_count(
) if max_processes == -1 else max_processes
with multiprocessing.Pool(max_processes) as pool:
tasks = []
for ticker, training_data in data.items():
tasks.append(
pool.apply_async(
handle_model_creation, [
ticker, training_data, out_dir,
self._overwrite_existing
], {"combined_examples": self._periods_per_example}))
for task in tqdm.tqdm(tasks):
task.get()
def predict_data(self, data, passback, in_model_dir):
out_dir = in_model_dir + path.sep + 'similarity_analysis'
if not path.exists(out_dir):
os.mkdir(out_dir)
exec_options = config_parser_singleton.read_execution_options()
max_processes = exec_options[1]
max_processes = multiprocessing.cpu_count() if max_processes == -1 else max_processes
with multiprocessing.Pool(max_processes) as pool:
open_jobs = []
for ticker, prediction_data in data.items():
open_jobs.append(pool.apply_async(
predict_data,
[ticker, out_dir, prediction_data],
{'combined_examples': self._combined_examples_factor,
'num_similar_regions': self._num_similar_regions}
))
for job in open_jobs:
job.get()
def predict_data(self, data, passback, in_model_dir):
model_dir = in_model_dir + path.sep + 'ann_models'
if not path.exists(model_dir):
raise FileNotFoundError("Model storage directory for ANN prediction does not exist. Please run model "
"creation without the prediction flag set to true to create models used in "
"prediction.")
predictions = {}
_, max_processes, _ = config_parser_singleton.read_execution_options()
max_processes = multiprocessing.cpu_count() if max_processes == -1 else max_processes
with multiprocessing.Pool(max_processes) as pool:
working_threads = []
for ticker, prediction_data in data.items():
working_threads.append(pool.apply_async(predict_data,
[ticker, model_dir, prediction_data],
{'trend_lookahead': self._trend_lookahead,
'combined_examples': self._combined_examples_factor}))
for worker in tqdm.tqdm(working_threads):
result = worker.get()
if result is not None:
ticker, actual_prediction, accuracy = result
predictions[ticker] = (actual_prediction, accuracy)
return predictions
def pass_data(self, output_dir, stop_for_errors=False, print_errors=True):
provider = None
consumer = None
columns = None
predict, _, export_data = read_execution_options()
ret_predictions = {}
for provKey, provider in self.providers.items():
try:
if provKey not in self.consumers.keys():
continue
registeredConsumers = self.consumers[provKey]
for consumer_set in registeredConsumers:
consumer = None
args = None
passback = None
keyword_args = {}
if len(consumer_set) == 3:
consumer, args, passback = consumer_set
elif len(consumer_set) == 2:
consumer, args = consumer_set
elif len(consumer_set) == 1:
consumer = consumer_set[0]
elif len(consumer_set) == 4:
consumer, args, passback, keyword_args = consumer_set
else:
raise ValueError(
"Invalid number of consumer registration arguments"
)
if not predict:
consumer.consume_data(
provider.generate_data(*args, **keyword_args),
passback, output_dir)
else:
predictions = consumer.predict_data(
provider.generate_prediction_data(
*args, **keyword_args), passback, output_dir)
consumer_passback_id = str(
type(consumer)) + str(passback)
if consumer_passback_id in self._prediction_string_serializers and not export_data:
predictions = self._prediction_string_serializers[
consumer_passback_id](predictions)
elif consumer_passback_id in self._data_exportation_functions and export_data:
self._data_exportation_functions[
consumer_passback_id](predictions, output_dir)
ret_predictions[consumer_passback_id] = predictions
except Exception:
if print_errors:
traceback.print_exc()
logger.logger.log(
logger.NON_FATAL_ERROR,
"Above error was encountered during processing "
"of the following provider/consumer pair")
logger.logger.log(
logger.NON_FATAL_ERROR,
"\t%s %s" % (type(provider), type(consumer)))
logger.logger.log(
logger.NON_FATAL_ERROR,
"With the following columns as a data argument")
logger.logger.log(logger.NON_FATAL_ERROR,
"\t%s" % str(columns))
if stop_for_errors:
return
return ret_predictions
from data_providing_module import configurable_registry
from data_providing_module.data_provider_registry import registry
from general_utils.config.config_parser_singleton import parser, update_config, read_execution_options
from general_utils.exportation import csv_amalgamation
from general_utils.logging import logger
# os.environ["KERAS_BACKEND"] = "plaidml.keras.backend"
if __name__ == "__main__":
import sys
args = sys.argv[1:]
providers = os.listdir("data_providing_module/data_providers")
for provider in providers:
if provider.startswith('__'):
continue
importlib.import_module('data_providing_module.data_providers.' +
provider.replace('.py', ''))
consumers = os.listdir("training_managers")
for consumer in consumers:
if consumer.startswith('__'):
continue
importlib.import_module("training_managers." +
consumer.replace('.py', ''))
configurable_registry.config_registry.handle_configurables(parser)
predict, max_processes, export_predictions = read_execution_options()
update_config()
ret_predictions = registry.pass_data(args[0], stop_for_errors=False)
if predict and not export_predictions:
for passback, predictions in ret_predictions.items():
logger.logger.log(logger.OUTPUT, predictions)
