Add learning rate and l2 regularizer constants

src/constants.py

@@ -3,3 +3,5 @@ TRAIN_DATASET = "data/train_data.tfrecords"
 TEST_DATASET = "data/test_data.tfrecords"
 EPOCHS = 1000
 BATCH_SIZE = 256
+LEARNING_RATE = 0.004
+L2 = 0.001

src/preprocessing.py

@@ -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()