Pad and mask the sequences in each batch

src/hyperparameters.py

@@ -11,6 +11,7 @@ class Hyperparameters:
         learning_rate=0.004,
         l2_rate=0.001,
         log_directory="logs",
+        max_length=80,
     ):
         self.data_file = data_file
         self.label_file = label_file
@@ -22,3 +23,4 @@ class Hyperparameters:
         self.learning_rate = learning_rate
         self.l2_rate = l2_rate
         self.log_directory = log_directory
+        self.max_length = max_length

src/model.py

@@ -17,7 +17,8 @@ def build_model(hyperparams) -> Model:
     """
     model = Sequential(
         [
-            Input(shape=(None, len(BASES))),
+            Input(shape=(None, hyperparams.max_length, len(BASES))),
+            Masking(mask_value=-1),
             Conv1D(
                 filters=16,
                 kernel_size=5,

src/preprocessing.py

@@ -22,19 +22,6 @@ def align_sequences(sequence, label) -> Tuple[str, str]:
     return aligned_seq, aligned_label
 
 
-def generate_example(sequence, label) -> bytes:
-    """
-    Create a binary-string for each sequence containing the sequence and the bases' counts
-    """
-    aligned_seq, aligned_label = align_sequences(sequence, label)
-    schema = {
-        "sequence": Feature(int64_list=Int64List(value=encode_sequence(aligned_seq))),
-        "label": Feature(int64_list=Int64List(value=encode_sequence(aligned_label))),
-    }
-    example = Example(features=Features(feature=schema))
-    return example.SerializeToString()
-
-
 def encode_sequence(sequence) -> List[int]:
     """
     Encode the DNA sequence using the indices of the BASES constant
@@ -43,6 +30,30 @@ def encode_sequence(sequence) -> List[int]:
     return encoded_sequence
 
 
+def prepare_sequences(sequence, label):
+    """
+    Align and encode the sequences to obtain a fixed length output in order to perform batching
+    """
+    encoded_sequences = []
+    aligned_seq, aligned_label = align_sequences(sequence, label)
+    for item in [aligned_seq, aligned_label]:
+        encoded_sequences.append(encode_sequence(item))
+    return encoded_sequences[0], encoded_sequences[1]
+
+
+def generate_example(sequence, label) -> bytes:
+    """
+    Create a binary-string for each sequence containing the sequence and the bases' counts
+    """
+    processed_seq, processed_label = prepare_sequences(sequence, label)
+    schema = {
+        "sequence": Feature(int64_list=Int64List(value=processed_seq)),
+        "label": Feature(int64_list=Int64List(value=processed_label)),
+    }
+    example = Example(features=Features(feature=schema))
+    return example.SerializeToString()
+
+
 def read_fastq(hyperparams) -> List[bytes]:
     """
     Parses a data and a label FASTQ files and generates a List of serialized Examples
@@ -104,8 +115,13 @@ def read_dataset(filepath, hyperparams) -> TFRecordDataset:
     data_input = TFRecordDataset(filenames=filepath)
     dataset = data_input.map(map_func=process_input, num_parallel_calls=AUTOTUNE)
     shuffled_dataset = dataset.shuffle(buffer_size=10000, seed=42)
-    batched_dataset = shuffled_dataset.batch(batch_size=hyperparams.batch_size).repeat(
+    batched_dataset = shuffled_dataset.padded_batch(
-        count=hyperparams.epochs
+        batch_size=hyperparams.batch_size,
+        padded_shapes=(
+            [hyperparams.max_length, len(BASES)],
+            [hyperparams.max_length, len(BASES)],
+        ),
+        padding_values=-1.0,
     )
     return batched_dataset