MCPcopy Create free account
hub / github.com/DeepRec-AI/DeepRec / main

Function main

modelzoo/dssm/train.py:552–668  ·  view source on GitHub ↗
(tf_config=None, server=None)

Source from the content-addressed store, hash-verified

550
551
552def main(tf_config=None, server=None):
553 # check dataset and count data set size
554 print("Checking dataset...")
555 train_file = args.data_location + '/taobao_train_data'
556 test_file = args.data_location + '/taobao_test_data'
557 if args.parquet_dataset and not args.tf:
558 train_file += '.parquet'
559 test_file += '.parquet'
560 if (not os.path.exists(train_file)) or (not os.path.exists(test_file)):
561 print("Dataset does not exist in the given data_location.")
562 sys.exit()
563 no_of_training_examples = 0
564 no_of_test_examples = 0
565 if args.parquet_dataset and not args.tf:
566 import pyarrow.parquet as pq
567 no_of_training_examples = pq.read_table(train_file).num_rows
568 no_of_test_examples = pq.read_table(test_file).num_rows
569 else:
570 no_of_training_examples = sum(1 for line in open(train_file))
571 no_of_test_examples = sum(1 for line in open(test_file))
572 print("Numbers of training dataset is {}".format(no_of_training_examples))
573 print("Numbers of test dataset is {}".format(no_of_test_examples))
574
575 # set batch size, eporch & steps
576 batch_size = math.ceil(
577 args.batch_size / args.micro_batch
578 ) if args.micro_batch and not args.tf else args.batch_size
579
580 if args.steps == 0:
581 no_of_epochs = 100
582 train_steps = math.ceil(
583 (float(no_of_epochs) * no_of_training_examples) / batch_size)
584 else:
585 no_of_epochs = math.ceil(
586 (float(batch_size) * args.steps) / no_of_training_examples)
587 train_steps = args.steps
588 test_steps = math.ceil(float(no_of_test_examples) / batch_size)
589 print("The training steps is {}".format(train_steps))
590 print("The testing steps is {}".format(test_steps))
591
592 # set fixed random seed
593 tf.set_random_seed(args.seed)
594
595 # set directory path for checkpoint_dir
596 model_dir = os.path.join(args.output_dir,
597 'model_DSSM_' + str(int(time.time())))
598 checkpoint_dir = args.checkpoint if args.checkpoint else model_dir
599 print("Saving model checkpoints to " + checkpoint_dir)
600
601 # create data pipline of train & test dataset
602 train_dataset = build_model_input(train_file, batch_size, no_of_epochs)
603 test_dataset = build_model_input(test_file, batch_size, 1)
604
605 dataset_output_types = tf.data.get_output_types(train_dataset)
606 dataset_output_shapes = tf.data.get_output_shapes(test_dataset)
607 iterator = tf.data.Iterator.from_structure(dataset_output_types,
608 dataset_output_shapes)
609 next_element = iterator.get_next()

Callers 1

train.pyFile · 0.70

Calls 14

sumFunction · 0.85
DSSMClass · 0.85
exitMethod · 0.80
timeMethod · 0.80
from_structureMethod · 0.80
build_model_inputFunction · 0.70
build_feature_columnsFunction · 0.70
trainFunction · 0.70
evalFunction · 0.70
formatMethod · 0.45
joinMethod · 0.45
get_nextMethod · 0.45

Tested by

no test coverage detected