linting

lab2/text_recognizer/datasets/emnist_lines_dataset.py

@@ -16,7 +16,7 @@
 ESSENTIALS_FILENAME = Path(__file__).parents[0].resolve() / "emnist_lines_essentials.json"
 
 
-class EmnistLinesDataset(Dataset):
+class EmnistLinesDataset(Dataset):  # pylint: disable=too-many-instance-attributes
     """
     EmnistLinesDataset class.
 

lab2/text_recognizer/networks/line_cnn_all_conv.py

@@ -81,7 +81,7 @@ def line_cnn_for_transformer(
 
     # Because of MaxPooling, everything is divided by 2
     new_height = image_height // 2
-    new_width = image_width // 2
+    # new_width = image_width // 2
     new_window_width = window_width // 2
     new_window_stride = window_stride // 2
 

lab3/text_recognizer/datasets/emnist_lines_dataset.py

@@ -16,7 +16,7 @@
 ESSENTIALS_FILENAME = Path(__file__).parents[0].resolve() / "emnist_lines_essentials.json"
 
 
-class EmnistLinesDataset(Dataset):
+class EmnistLinesDataset(Dataset):  # pylint: disable=too-many-instance-attributes
     """
     EmnistLinesDataset class.
 

lab3/text_recognizer/models/line_model_transformer.py

@@ -15,7 +15,7 @@
     sparse_categorical_crossentropy_ignoring_padding,
 )
 from text_recognizer.networks import line_cnn_for_transformer
-from text_recognizer.networks.transformer import run_transformer_inference
+from text_recognizer.networks.transformer.main import run_transformer_inference
 
 
 class LineModelTransformer(Model):
@@ -81,6 +81,8 @@ def predict_on_image(self, image: np.ndarray) -> Tuple[str, float]:
         """Predict on a single image."""
         if image.dtype == np.uint8:
             image = (image / 255).astype(np.float32)
-        pred = run_inference(self.model, image, self.data.max_length, self.data.start_label, self.data.end_label)
-        string = "".join([dataset.mapping[ind] for ind in pred.numpy()]).strip()
-        return pred, 1.0  # NOTE: conference is always given as 1.0 for now...
+        pred = run_transformer_inference(
+            self.network, image, self.data.max_length, self.data.start_label, self.data.end_label
+        )
+        string_ = "".join([self.data.mapping[ind] for ind in pred.numpy()]).strip()
+        return string_, 1.0  # NOTE: conference is always given as 1.0 for now...

lab3/text_recognizer/models/util.py

@@ -1,3 +1,4 @@
+"""Utility functions."""
 from tensorflow.python.keras import backend as K
 from tensorflow.python.framework import ops
 from tensorflow.python.ops import array_ops
@@ -42,7 +43,7 @@ def sparse_categorical_accuracy_ignoring_padding(y_true, y_pred, padding_label):
     return math_ops.cast(math_ops.equal(y_true_masked, y_pred_masked), K.floatx())
 
 
-class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):
+class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):  # pylint: disable=abstract-method
     """Learning rate scheduler from the Attention is all you need paper."""
 
     def __init__(self, d_model, warmup_steps=4000):

lab3/text_recognizer/networks/line_cnn_all_conv.py

@@ -81,7 +81,7 @@ def line_cnn_for_transformer(
 
     # Because of MaxPooling, everything is divided by 2
     new_height = image_height // 2
-    new_width = image_width // 2
+    # new_width = image_width // 2
     new_window_width = window_width // 2
     new_window_stride = window_stride // 2
 

lab3/text_recognizer/networks/transformer/attention.py

@@ -1,10 +1,12 @@
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+"""MultiHeadAttention class and supporting functions."""
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 
 import tensorflow as tf
 
 
 def scaled_dot_product_attention(query, key, value, mask):
-    """Calculate the attention weights. """
+    """Calculate the attention weights."""
     matmul_qk = tf.matmul(query, key, transpose_b=True)
 
     # scale matmul_qk
@@ -45,7 +47,7 @@ def split_heads(self, inputs, batch_size):
         inputs = tf.reshape(inputs, shape=(batch_size, -1, self.num_heads, self.depth))
         return tf.transpose(inputs, perm=[0, 2, 1, 3])
 
-    def call(self, inputs):
+    def call(self, inputs):  # pylint: disable=arguments-differ
         query, key, value, mask = inputs["query"], inputs["key"], inputs["value"], inputs.get("mask", None)
         batch_size = tf.shape(query)[0]
 

lab3/text_recognizer/networks/transformer/decoder.py

@@ -1,4 +1,6 @@
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+"""Transformer decoder layer."""
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 import tensorflow as tf
 
 from text_recognizer.networks.transformer.positional_encoding import PositionalEncoding

lab3/text_recognizer/networks/transformer/main.py

@@ -1,5 +1,6 @@
 """Transformer Keras model."""
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 from typing import Tuple
 
 import tensorflow as tf
@@ -15,7 +16,7 @@ def run_transformer_inference(
 ) -> tf.Tensor:
     output = tf.expand_dims([start_label], 0)
 
-    for i in range(max_length):
+    for _ in range(max_length):
         predictions = model(inputs=[image, output], training=False)
 
         # select the last word from the seq_len dimension
@@ -26,12 +27,13 @@ def run_transformer_inference(
             break
 
         # concatenated the predicted_id to the output which is given to the decoder as input
+        # pylint: disable=unexpected-keyword-arg,no-value-for-parameter
         output = tf.concat([output, predicted_id], axis=-1)
 
     return tf.squeeze(output, axis=0)
 
 
-def transformer(
+def transformer(  # pylint: disable=too-many-arguments
     image_shape: Tuple[int, ...],
     window_width: int,
     window_stride: int,

lab3/text_recognizer/networks/transformer/masking.py

@@ -1,4 +1,6 @@
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+"""Masking code for transformers."""
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 import tensorflow as tf
 
 

lab3/text_recognizer/networks/transformer/positional_encoding.py

@@ -1,18 +1,22 @@
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+"""Positional encoding for transformer."""
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 import tensorflow as tf
 
 
 MAX_POSITION = 1000
 
 
 class PositionalEncoding(tf.keras.layers.Layer):
+    """Positional encoding for transformer."""
+
     def __init__(self, d_model, position=None):
         super(PositionalEncoding, self).__init__()
         if position is None:
             position = MAX_POSITION
         self.pos_encoding = self.positional_encoding(position, d_model)
 
-    def get_angles(self, position, i, d_model):
+    def get_angles(self, position, i, d_model):  # pylint: disable=no-self-use
         angles = 1 / tf.pow(10000, (2 * (i // 2)) / tf.cast(d_model, tf.float32))
         return position * angles
 
@@ -27,9 +31,10 @@ def positional_encoding(self, position, d_model):
         # apply cos to odd index in the array
         cosines = tf.math.cos(angle_rads[:, 1::2])
 
+        # pylint: disable=unexpected-keyword-arg,no-value-for-parameter
         pos_encoding = tf.concat([sines, cosines], axis=-1)
         pos_encoding = pos_encoding[tf.newaxis, ...]
         return tf.cast(pos_encoding, tf.float32)
 
-    def call(self, inputs):
+    def call(self, inputs):  # pylint: disable=arguments-differ
         return inputs + self.pos_encoding[:, : tf.shape(inputs)[1], :]

lab4/text_recognizer/datasets/emnist_lines_dataset.py

@@ -16,7 +16,7 @@
 ESSENTIALS_FILENAME = Path(__file__).parents[0].resolve() / "emnist_lines_essentials.json"
 
 
-class EmnistLinesDataset(Dataset):
+class EmnistLinesDataset(Dataset):  # pylint: disable=too-many-instance-attributes
     """
     EmnistLinesDataset class.
 

lab4/text_recognizer/datasets/iam_lines_dataset.py

@@ -11,7 +11,7 @@
 
 from text_recognizer import util
 from text_recognizer.datasets.dataset import Dataset, _parse_args
-from text_recognizer.datasets.emnist_lines_dataset import EmnistLinesDataset
+from text_recognizer.datasets.emnist_lines_dataset import add_start_and_end_labels, EmnistLinesDataset
 
 
 PROCESSED_DATA_DIRNAME = Dataset.data_dirname() / "processed" / "iam_lines"
@@ -43,6 +43,9 @@ def __init__(
         self.with_start_and_end_labels = with_start_and_end_labels
         self.mapping = EmnistLinesDataset().mapping
         self.inverse_mapping = {v: k for k, v in self.mapping.items()}
+        self.padding_label = self.inverse_mapping["_"]
+        self.start_label = self.inverse_mapping["<s>"]
+        self.end_label = self.inverse_mapping["<e>"]
         self.num_classes = len(self.mapping)
         self.input_shape = (28, 952)
         self.output_shape = (97, self.num_classes)

lab4/text_recognizer/models/line_model_transformer.py

@@ -15,7 +15,7 @@
     sparse_categorical_crossentropy_ignoring_padding,
 )
 from text_recognizer.networks import line_cnn_for_transformer
-from text_recognizer.networks.transformer import run_transformer_inference
+from text_recognizer.networks.transformer.main import run_transformer_inference
 
 
 class LineModelTransformer(Model):
@@ -81,6 +81,8 @@ def predict_on_image(self, image: np.ndarray) -> Tuple[str, float]:
         """Predict on a single image."""
         if image.dtype == np.uint8:
             image = (image / 255).astype(np.float32)
-        pred = run_inference(self.model, image, self.data.max_length, self.data.start_label, self.data.end_label)
-        string = "".join([dataset.mapping[ind] for ind in pred.numpy()]).strip()
-        return pred, 1.0  # NOTE: conference is always given as 1.0 for now...
+        pred = run_transformer_inference(
+            self.network, image, self.data.max_length, self.data.start_label, self.data.end_label
+        )
+        string_ = "".join([self.data.mapping[ind] for ind in pred.numpy()]).strip()
+        return string_, 1.0  # NOTE: conference is always given as 1.0 for now...

lab4/text_recognizer/models/util.py

@@ -1,3 +1,4 @@
+"""Utility functions."""
 from tensorflow.python.keras import backend as K
 from tensorflow.python.framework import ops
 from tensorflow.python.ops import array_ops
@@ -42,7 +43,7 @@ def sparse_categorical_accuracy_ignoring_padding(y_true, y_pred, padding_label):
     return math_ops.cast(math_ops.equal(y_true_masked, y_pred_masked), K.floatx())
 
 
-class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):
+class CustomSchedule(tf.keras.optimizers.schedules.LearningRateSchedule):  # pylint: disable=abstract-method
     """Learning rate scheduler from the Attention is all you need paper."""
 
     def __init__(self, d_model, warmup_steps=4000):

lab4/text_recognizer/networks/line_cnn_all_conv.py

@@ -81,7 +81,7 @@ def line_cnn_for_transformer(
 
     # Because of MaxPooling, everything is divided by 2
     new_height = image_height // 2
-    new_width = image_width // 2
+    # new_width = image_width // 2
     new_window_width = window_width // 2
     new_window_stride = window_stride // 2
 

lab4/text_recognizer/networks/transformer/attention.py

@@ -1,10 +1,12 @@
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+"""MultiHeadAttention class and supporting functions."""
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 
 import tensorflow as tf
 
 
 def scaled_dot_product_attention(query, key, value, mask):
-    """Calculate the attention weights. """
+    """Calculate the attention weights."""
     matmul_qk = tf.matmul(query, key, transpose_b=True)
 
     # scale matmul_qk
@@ -45,7 +47,7 @@ def split_heads(self, inputs, batch_size):
         inputs = tf.reshape(inputs, shape=(batch_size, -1, self.num_heads, self.depth))
         return tf.transpose(inputs, perm=[0, 2, 1, 3])
 
-    def call(self, inputs):
+    def call(self, inputs):  # pylint: disable=arguments-differ
         query, key, value, mask = inputs["query"], inputs["key"], inputs["value"], inputs.get("mask", None)
         batch_size = tf.shape(query)[0]
 

lab4/text_recognizer/networks/transformer/decoder.py

@@ -1,4 +1,6 @@
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+"""Transformer decoder layer."""
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 import tensorflow as tf
 
 from text_recognizer.networks.transformer.positional_encoding import PositionalEncoding

lab4/text_recognizer/networks/transformer/main.py

@@ -1,5 +1,6 @@
 """Transformer Keras model."""
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 from typing import Tuple
 
 import tensorflow as tf
@@ -15,7 +16,7 @@ def run_transformer_inference(
 ) -> tf.Tensor:
     output = tf.expand_dims([start_label], 0)
 
-    for i in range(max_length):
+    for _ in range(max_length):
         predictions = model(inputs=[image, output], training=False)
 
         # select the last word from the seq_len dimension
@@ -26,12 +27,13 @@ def run_transformer_inference(
             break
 
         # concatenated the predicted_id to the output which is given to the decoder as input
+        # pylint: disable=unexpected-keyword-arg,no-value-for-parameter
         output = tf.concat([output, predicted_id], axis=-1)
 
     return tf.squeeze(output, axis=0)
 
 
-def transformer(
+def transformer(  # pylint: disable=too-many-arguments
     image_shape: Tuple[int, ...],
     window_width: int,
     window_stride: int,

lab4/text_recognizer/networks/transformer/masking.py

@@ -1,4 +1,6 @@
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+"""Masking code for transformers."""
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 import tensorflow as tf
 
 

lab4/text_recognizer/networks/transformer/positional_encoding.py

@@ -1,18 +1,22 @@
-# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master/community/en/transformer_chatbot.ipynb
+"""Positional encoding for transformer."""
+# Code originally from https://colab.research.google.com/github/tensorflow/examples/blob/master
+#                      /community/en/transformer_chatbot.ipynb
 import tensorflow as tf
 
 
 MAX_POSITION = 1000
 
 
 class PositionalEncoding(tf.keras.layers.Layer):
+    """Positional encoding for transformer."""
+
     def __init__(self, d_model, position=None):
         super(PositionalEncoding, self).__init__()
         if position is None:
             position = MAX_POSITION
         self.pos_encoding = self.positional_encoding(position, d_model)
 
-    def get_angles(self, position, i, d_model):
+    def get_angles(self, position, i, d_model):  # pylint: disable=no-self-use
         angles = 1 / tf.pow(10000, (2 * (i // 2)) / tf.cast(d_model, tf.float32))
         return position * angles
 
@@ -27,9 +31,10 @@ def positional_encoding(self, position, d_model):
         # apply cos to odd index in the array
         cosines = tf.math.cos(angle_rads[:, 1::2])
 
+        # pylint: disable=unexpected-keyword-arg,no-value-for-parameter
         pos_encoding = tf.concat([sines, cosines], axis=-1)
         pos_encoding = pos_encoding[tf.newaxis, ...]
         return tf.cast(pos_encoding, tf.float32)
 
-    def call(self, inputs):
+    def call(self, inputs):  # pylint: disable=arguments-differ
         return inputs + self.pos_encoding[:, : tf.shape(inputs)[1], :]

lab5/text_recognizer/datasets/emnist_lines_dataset.py

@@ -16,7 +16,7 @@
 ESSENTIALS_FILENAME = Path(__file__).parents[0].resolve() / "emnist_lines_essentials.json"
 
 
-class EmnistLinesDataset(Dataset):
+class EmnistLinesDataset(Dataset):  # pylint: disable=too-many-instance-attributes
     """
     EmnistLinesDataset class.
 

lab5/text_recognizer/datasets/iam_lines_dataset.py

@@ -11,7 +11,7 @@
 
 from text_recognizer import util
 from text_recognizer.datasets.dataset import Dataset, _parse_args
-from text_recognizer.datasets.emnist_lines_dataset import EmnistLinesDataset
+from text_recognizer.datasets.emnist_lines_dataset import add_start_and_end_labels, EmnistLinesDataset
 
 
 PROCESSED_DATA_DIRNAME = Dataset.data_dirname() / "processed" / "iam_lines"
@@ -43,6 +43,9 @@ def __init__(
         self.with_start_and_end_labels = with_start_and_end_labels
         self.mapping = EmnistLinesDataset().mapping
         self.inverse_mapping = {v: k for k, v in self.mapping.items()}
+        self.padding_label = self.inverse_mapping["_"]
+        self.start_label = self.inverse_mapping["<s>"]
+        self.end_label = self.inverse_mapping["<e>"]
         self.num_classes = len(self.mapping)
         self.input_shape = (28, 952)
         self.output_shape = (97, self.num_classes)