Document read_dataset and process_input
This commit is contained in:
parent
f8c1a54be3
commit
ccaa8484c7
|
@ -36,7 +36,7 @@ def encode_sequence(sequence) -> List[int]:
|
|||
return encoded_sequence
|
||||
|
||||
|
||||
def parse_data(filepath) -> List[bytes]:
|
||||
def read_fastq(filepath) -> List[bytes]:
|
||||
"""
|
||||
Parse a FASTQ file and generate a List of serialized Examples
|
||||
"""
|
||||
|
@ -54,7 +54,7 @@ def create_dataset(filepath) -> None:
|
|||
"""
|
||||
Create a training and test dataset with a 70/30 split respectively
|
||||
"""
|
||||
data = parse_data(filepath)
|
||||
data = read_fastq(filepath)
|
||||
train_test_split = 0.7
|
||||
with TFRecordWriter(TRAIN_DATASET) as train, TFRecordWriter(TEST_DATASET) as test:
|
||||
for element in data:
|
||||
|
@ -64,7 +64,10 @@ def create_dataset(filepath) -> None:
|
|||
test.write(element)
|
||||
|
||||
|
||||
def process_input(byte_string):
|
||||
def process_input(byte_string) -> Example:
|
||||
"""
|
||||
Parse a byte-string into an Example object
|
||||
"""
|
||||
schema = {
|
||||
"sequence": FixedLenFeature(shape=[], dtype=int64),
|
||||
"A_counts": FixedLenFeature(shape=[], dtype=float32),
|
||||
|
@ -75,13 +78,12 @@ def process_input(byte_string):
|
|||
return parse_single_example(byte_string, features=schema)
|
||||
|
||||
|
||||
def read_dataset():
|
||||
def read_dataset() -> TFRecordDataset:
|
||||
"""
|
||||
Read TFRecords files and generate a dataset
|
||||
"""
|
||||
data_input = TFRecordDataset(filenames=[TRAIN_DATASET, TEST_DATASET])
|
||||
dataset = data_input.map(map_func=process_input)
|
||||
shuffled_dataset = dataset.shuffle(buffer_size=10000, reshuffle_each_iteration=True)
|
||||
batched_dataset = shuffled_dataset.batch(batch_size=BATCH_SIZE).repeat(count=EPOCHS)
|
||||
return batched_dataset
|
||||
|
||||
|
||||
create_dataset("data/curesim-HVR.fastq")
|
||||
dataset = read_dataset()
|
||||
|
|
Loading…
Reference in New Issue