core\refac: #98 nullable sum to one reg

- nullable sum to one regularizer scaling factor

Author: blotero

core/pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 description = "Framework for handling image segmentation in the context of multiple annotators"
 name = "seg_tgce"
-version = "0.3.14"
+version = "0.3.16"
 readme = "README.md"
 authors = [{ name = "Brandon Lotero", email = "blotero@gmail.com" }]
 maintainers = [{ name = "Brandon Lotero", email = "blotero@gmail.com" }]
@@ -15,7 +15,7 @@ Issues = "https://github.com/blotero/seg_tgce/issues"
 
 [tool.poetry]
 name = "seg_tgce"
-version = "0.3.14"
+version = "0.3.16"
 authors = ["Brandon Lotero <blotero@gmail.com>"]
 description = "A package for the SEG TGCE project"
 readme = "README.md"

…re/seg_tgce/data/crowd_seg/__retrieve.py core/seg_tgce/data/crowd_seg/_retrieve.pycore/seg_tgce/data/crowd_seg/__retrieve.py renamed to core/seg_tgce/data/crowd_seg/_retrieve.py core/seg_tgce/data/crowd_seg/__retrieve.py renamed to core/seg_tgce/data/crowd_seg/_retrieve.py

@@ -13,8 +13,8 @@
 
 _TARGET_DIR = "__data__/crowd_seg"
 _BUCKET_NAME = "crowd-seg-data"
-PATCHES_OBJECT_NAME = "patches_refined.zip"
-MASKS_OBJECT_NAME = "masks_refined.zip"
+PATCHES_OBJECT_NAME = "patches_refined_v2.zip"
+MASKS_OBJECT_NAME = "masks_refined_v2.zip"
 
 
 s3 = boto3.client("s3", config=Config(signature_version=UNSIGNED))

core/seg_tgce/data/crowd_seg/tfds_builder.py

@@ -9,7 +9,7 @@
 from matplotlib import pyplot as plt
 from matplotlib.colors import ListedColormap, to_rgb
 
-from seg_tgce.data.crowd_seg.__retrieve import (
+from seg_tgce.data.crowd_seg._retrieve import (
     _BUCKET_NAME,
     MASKS_OBJECT_NAME,
     PATCHES_OBJECT_NAME,
@@ -20,12 +20,9 @@
 
 
 CLASSES_DEFINITION = {
-    0: "Ignore",
-    1: "Other",
-    2: "Tumor",
-    3: "Stroma",
-    4: "B. Inflammation",  # Benign Inflammation
-    5: "Necrosis",
+    0: "Other",
+    1: "Tumor",
+    2: "Stroma",
 }
 
 REAL_SCORERS = [
@@ -151,8 +148,9 @@ def map_sample(sample: SampleData, image_size: tuple[int, int]) -> ProcessedSamp
 class CrowdSegDataset(tfds.core.GeneratorBasedBuilder):
     """DatasetBuilder for crowd segmentation dataset."""
 
-    VERSION = tfds.core.Version("1.0.0")
+    VERSION = tfds.core.Version("1.1.0")
     RELEASE_NOTES = {
+        "1.1.0": "Use further refined patches and masks",
         "1.0.0": "Initial release.",
     }
 
@@ -184,8 +182,8 @@ def _info(self) -> tfds.core.DatasetInfo:
                 }
             ),
             supervised_keys=("image", "masks"),
-            homepage="https://github.com/your-repo/crowd-seg",
-            citation="""@article{your-citation}""",
+            homepage="https://github.com/blotero/seg_tgce",
+            # citation="""@article{your-citation}""",
         )
 
     def _split_generators(
@@ -559,7 +557,7 @@ def print_sample_info(data: dict) -> None:
     print("-" * 50)
 
 
-def main_processed() -> None:
+def main() -> None:
     target_size = (128, 128)
     batch_size = 16
 
@@ -568,26 +566,15 @@ def main_processed() -> None:
     )
     print(f"Dataset train length: {len(train)}")
 
-    fig = visualize_sample(
+    _ = visualize_sample(
         validation,
         target_size,
         batch_index=1,
         sample_indexes=[0, 1, 3, 4],
         sparse_labelers=False,
     )
-    fig.savefig(
-        "/home/brandon/unal/maestria/master_thesis/Cap1/Figures/multiannotator-segmentation.png"
-    )
-
-
-def main_raw():
-    train, validation, test = get_crowd_seg_dataset_tfds(
-        image_size=(64, 64),
-    )
-    for i, data in enumerate(train):
-        print(f"\nSample {i}:")
-        print_sample_info(data)
+    plt.show()
 
 
 if __name__ == "__main__":
-    main_processed()
+    main()

core/seg_tgce/experiments/histology/baseline.py

@@ -11,8 +11,8 @@
 
 from ..utils import handle_training
 
-TARGET_SHAPE = (256, 256)
-BATCH_SIZE = 32
+TARGET_SHAPE = (128, 128)
+BATCH_SIZE = 16
 TRAIN_EPOCHS = 20
 TUNER_EPOCHS = 2
 MAX_TRIALS = 5

core/seg_tgce/experiments/histology/features.py

@@ -24,9 +24,9 @@
 OBJECTIVE = "val_segmentation_output_dice_coefficient"
 
 DEFAULT_HPARAMS = {
-    "initial_learning_rate": 1e-3,
-    "q": 0.5,
-    "noise_tolerance": 0.5,
+    "initial_learning_rate": 6.5e-4,
+    "q": 0.66,
+    "noise_tolerance": 0.62,
     "a": 0.5,
     "b": 0.5,
     "lambda_reg_weight": 0.1,
@@ -52,9 +52,9 @@ def build_model_from_trial(trial: HpTunerTrial | None) -> Model:
         )
 
     return build_features_model_from_hparams(
-        learning_rate=trial.suggest_float("learning_rate", 1e-5, 1e-2, log=True),
-        q=trial.suggest_float("q", 0.1, 0.9, step=0.01),
-        noise_tolerance=trial.suggest_float("noise_tolerance", 0.1, 0.9, step=0.01),
+        learning_rate=DEFAULT_HPARAMS["initial_learning_rate"],
+        q=DEFAULT_HPARAMS["q"],
+        noise_tolerance=DEFAULT_HPARAMS["noise_tolerance"],
         a=trial.suggest_float("a", 0.1, 10.0, step=0.1),
         b=trial.suggest_float("b", 0.1, 0.99, step=0.01),
         lambda_reg_weight=trial.suggest_float("lambda_reg_weight", 0.0, 10.0, step=0.1),

core/seg_tgce/experiments/histology/pixel.py

@@ -23,9 +23,9 @@
 OBJECTIVE = "val_segmentation_output_dice_coefficient"
 
 DEFAULT_HPARAMS = {
-    "initial_learning_rate": 1e-3,
-    "q": 0.7,
-    "noise_tolerance": 0.5,
+    "initial_learning_rate": 6.5e-4,
+    "q": 0.66,
+    "noise_tolerance": 0.62,
     "a": 0.2,
     "b": 0.7,
     "lambda_reg_weight": 0.1,
@@ -51,9 +51,9 @@ def build_model_from_trial(trial: HpTunerTrial | None) -> Model:
         )
 
     return build_pixel_model_from_hparams(
-        learning_rate=trial.suggest_float("learning_rate", 1e-5, 1e-2, log=True),
-        q=trial.suggest_float("q", 0.1, 0.9, step=0.01),
-        noise_tolerance=trial.suggest_float("noise_tolerance", 0.1, 0.9, step=0.01),
+        learning_rate=DEFAULT_HPARAMS["initial_learning_rate"],
+        q=DEFAULT_HPARAMS["q"],
+        noise_tolerance=DEFAULT_HPARAMS["noise_tolerance"],
         a=trial.suggest_float("a", 0.1, 10.0, step=0.1),
         b=trial.suggest_float("b", 0.1, 0.99, step=0.01),
         lambda_reg_weight=trial.suggest_float("lambda_reg_weight", 0.0, 10.0, step=0.1),

core/seg_tgce/experiments/histology/scalar.py

@@ -23,9 +23,9 @@
 OBJECTIVE = "val_segmentation_output_dice_coefficient"
 
 DEFAULT_HPARAMS = {
-    "initial_learning_rate": 1e-3,
-    "q": 0.5,
-    "noise_tolerance": 0.5,
+    "initial_learning_rate": 6.5e-4,
+    "q": 0.66,
+    "noise_tolerance": 0.62,
     "a": 0.3,
     "b": 0.7,
     "lambda_reg_weight": 0.1,
@@ -51,9 +51,9 @@ def build_model_from_trial(trial: HpTunerTrial | None) -> Model:
         )
 
     return build_scalar_model_from_hparams(
-        learning_rate=trial.suggest_float("learning_rate", 1e-5, 1e-2, log=True),
-        q=trial.suggest_float("q", 0.1, 0.9, step=0.01),
-        noise_tolerance=trial.suggest_float("noise_tolerance", 0.1, 0.9, step=0.01),
+        learning_rate=DEFAULT_HPARAMS["initial_learning_rate"],
+        q=DEFAULT_HPARAMS["q"],
+        noise_tolerance=DEFAULT_HPARAMS["noise_tolerance"],
         a=trial.suggest_float("a", 0.1, 10.0, step=0.1),
         b=trial.suggest_float("b", 0.1, 0.99, step=0.01),
         lambda_reg_weight=trial.suggest_float("lambda_reg_weight", 0.0, 10.0, step=0.1),
@@ -93,6 +93,7 @@ def build_model_from_trial(trial: HpTunerTrial | None) -> Model:
         tuner_epochs=TUNER_EPOCHS,
         objective=OBJECTIVE,
         tuner_max_trials=MAX_TRIALS,
+        study_name=STUDY_NAME,
     )
 
     vis_callback = ScalarVisualizationCallback(

core/seg_tgce/loss/tgce.py

@@ -40,7 +40,7 @@ def __init__(  # pylint: disable=too-many-arguments
         b: float = 0.7,
         lambda_reg_weight: float = 0.1,
         lambda_entropy_weight: float = 0.1,
-        lambda_sum_weight: float = 0.1,
+        lambda_sum_weight: float | None = None,
         epsilon: float = 1e-8,
     ) -> None:
         self.q = q
@@ -98,8 +98,11 @@ def call(
             + (1 - valid_lambda_r) * tf.math.log1p(1 - valid_lambda_r)
         )
 
-        lambda_sum = self.lambda_sum_weight * tf.reduce_mean(
-            tf.square(tf.reduce_sum(valid_lambda_r, axis=-1) - 1.0)
+        lambda_sum = (
+            self.lambda_sum_weight
+            * tf.reduce_mean(tf.square(tf.reduce_sum(valid_lambda_r, axis=-1) - 1.0))
+            if self.lambda_sum_weight is not None
+            else 0.0
         )
 
         total_loss = (
@@ -153,7 +156,7 @@ def __init__(  # pylint: disable=too-many-arguments
         b: float = 0.7,
         lambda_reg_weight: float = 0.1,
         lambda_entropy_weight: float = 0.1,
-        lambda_sum_weight: float = 0.1,
+        lambda_sum_weight: float | None = None,
         epsilon: float = 1e-8,
     ) -> None:
         self.a = a
@@ -217,8 +220,11 @@ def call(
             + (1 - valid_lambda_r) * tf.math.log1p(1 - valid_lambda_r)
         )
 
-        lambda_sum = self.lambda_sum_weight * tf.reduce_mean(
-            tf.square(tf.reduce_sum(valid_lambda_r, axis=-1) - 1.0)
+        lambda_sum = (
+            self.lambda_sum_weight
+            * tf.reduce_mean(tf.square(tf.reduce_sum(valid_lambda_r, axis=-1) - 1.0))
+            if self.lambda_sum_weight is not None
+            else 0.0
         )
 
         total_loss = (
@@ -271,7 +277,7 @@ def __init__(  # pylint: disable=too-many-arguments
         b: float = 0.7,
         lambda_reg_weight: float = 0.1,
         lambda_entropy_weight: float = 0.1,
-        lambda_sum_weight: float = 0.1,
+        lambda_sum_weight: float | None = None,
         epsilon: float = 1e-8,
     ) -> None:
         self.a = a
@@ -330,8 +336,11 @@ def call(
             + (1 - valid_lambda_r) * tf.math.log1p(1 - valid_lambda_r)
         )
 
-        lambda_sum = self.lambda_sum_weight * tf.reduce_mean(
-            tf.square(tf.reduce_sum(valid_lambda_r, axis=-1) - 1.0)
+        lambda_sum = (
+            self.lambda_sum_weight
+            * tf.reduce_mean(tf.square(tf.reduce_sum(valid_lambda_r, axis=-1) - 1.0))
+            if self.lambda_sum_weight is not None
+            else 0.0
         )
 
         total_loss = (

core/seg_tgce/models/builders.py

@@ -63,7 +63,7 @@ def build_scalar_model_from_hparams(
     b: float,
     lambda_reg_weight: float,
     lambda_entropy_weight: float,
-    lambda_sum_weight: float,
+    lambda_sum_weight: float | None,
     num_classes: int,
     target_shape: tuple,
     n_scorers: int,
@@ -116,7 +116,7 @@ def build_features_model_from_hparams(
     b: float,
     lambda_reg_weight: float,
     lambda_entropy_weight: float,
-    lambda_sum_weight: float,
+    lambda_sum_weight: float | None,
     num_classes: int,
     target_shape: tuple,
     n_scorers: int,
@@ -184,7 +184,7 @@ def build_pixel_model_from_hparams(
     b: float,
     lambda_reg_weight: float,
     lambda_entropy_weight: float,
-    lambda_sum_weight: float,
+    lambda_sum_weight: float | None,
     num_classes: int,
     target_shape: tuple,
     n_scorers: int,