quadra 2.5.0__py3-none-any.whl → 2.6.0__py3-none-any.whl

quadra/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "2.5.0"
+__version__ = "2.6.0"
 
 
 def get_version():

quadra/configs/experiment/base/classification/classification.yaml

@@ -57,7 +57,6 @@ trainer:
 scheduler:
   patience: 20
   factor: 0.9
-  verbose: False
   threshold: 0.01
 
 callbacks:

quadra/configs/scheduler/default.yaml

@@ -1,5 +1,4 @@
 _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
 patience: 20
 factor: 0.5
-verbose: False
 threshold: 0.01

quadra/configs/scheduler/rop.yaml

@@ -1,5 +1,4 @@
 _target_: torch.optim.lr_scheduler.ReduceLROnPlateau
 patience: 20
 factor: 0.995
-verbose: False
 threshold: 0.01

quadra/tasks/anomaly.py

@@ -244,7 +244,7 @@ class AnomalibDetection(Generic[AnomalyDataModuleT], LightningTask[AnomalyDataMo
         ):
             threshold = torch.tensor(100.0)
         else:
-            threshold = self.module.image_metrics.F1Score.threshold
+            threshold = self.module.image_metrics.F1Score.threshold  # type: ignore[union-attr,assignment]
 
         # The output of the prediction is a normalized score so the cumulative histogram is displayed with the
         # normalized scores
@@ -328,6 +328,22 @@ class AnomalibDetection(Generic[AnomalyDataModuleT], LightningTask[AnomalyDataMo
                 else:
                     utils.upload_file_tensorboard(a, tensorboard_logger)
 
+    def execute(self):
+        """Execute the experiment and all the steps."""
+        self.prepare()
+        self.train()
+        # When training in fp16 mixed precision, export function casts model weights from fp32 to fp16,
+        # for this reason, predictions logits could slightly change and predictions could be inconsistent between
+        # test and generated report.
+        # Performing export before test allows to have consistent results in test metrics and generated report.
+        if self.config.export is not None and len(self.config.export.types) > 0:
+            self.export()
+        if self.run_test:
+            self.test()
+        if self.report:
+            self.generate_report()
+        self.finalize()
+
 
 class AnomalibEvaluation(Evaluation[AnomalyDataModule]):
     """Evaluation task for Anomalib.
@@ -445,12 +461,22 @@ class AnomalibEvaluation(Evaluation[AnomalyDataModule]):
         training_threshold = self.model_data[f"{self.training_threshold_type}_threshold"]
         optimal_threshold = self.metadata["threshold"]
 
-        normalized_optimal_threshold = cast(float, normalize_anomaly_score(optimal_threshold, training_threshold))
-
         os.makedirs(os.path.join(self.report_path, "predictions"), exist_ok=True)
         os.makedirs(os.path.join(self.report_path, "heatmaps"), exist_ok=True)
 
         anomaly_scores = self.metadata["anomaly_scores"].cpu().numpy()
+
+        # The reason I have to expand dims and cast the optimal threshold to anomaly_scores dtype is because
+        # of internal roundings performed differently by numpy and python
+        # Particularly the normalized_optimal_threshold computed directly using float values might be higher than the
+        # actual value obtained by the anomaly_scores
+        normalized_optimal_threshold = cast(
+            np.ndarray,
+            normalize_anomaly_score(
+                np.expand_dims(np.array(optimal_threshold, dtype=anomaly_scores.dtype), -1), training_threshold
+            ),
+        ).item()
+
         anomaly_scores = normalize_anomaly_score(anomaly_scores, training_threshold)
 
         if not isinstance(anomaly_scores, np.ndarray):

quadra/utils/anomaly.py

@@ -5,6 +5,8 @@
 
 from __future__ import annotations
 
+from typing import cast
+
 try:
     from typing import Any, TypeAlias
 except ImportError:
@@ -43,6 +45,39 @@ def normalize_anomaly_score(raw_score: MapOrValue, threshold: float) -> MapOrVal
     else:
         normalized_score = 200.0 - ((raw_score / threshold) * 100.0)
 
+    # Ensures that the normalized scores are consistent with the raw scores
+    # For all the items whose prediction changes after normalization, force the normalized score to be
+    # consistent with the prediction made on the raw score by clipping the score:
+    #   - to 100.0 if the prediction was "anomaly" on the raw score and "good" on the normalized score
+    #   - to 99.99 if the prediction was "good" on the raw score and "anomaly" on the normalized score
+    score = raw_score
+    if isinstance(score, torch.Tensor):
+        score = score.cpu().numpy()
+    # Anomalib classify as anomaly if anomaly_score gte threshold
+    is_anomaly_mask = score >= threshold
+    is_not_anomaly_mask = np.bitwise_not(is_anomaly_mask)
+    if isinstance(normalized_score, torch.Tensor):
+        if normalized_score.dim() == 0:
+            normalized_score = (
+                normalized_score.clamp(min=100.0) if is_anomaly_mask else normalized_score.clamp(max=99.99)
+            )
+        else:
+            normalized_score[is_anomaly_mask] = normalized_score[is_anomaly_mask].clamp(min=100.0)
+            normalized_score[is_not_anomaly_mask] = normalized_score[is_not_anomaly_mask].clamp(max=99.99)
+    elif isinstance(normalized_score, np.ndarray) or np.isscalar(normalized_score):
+        if np.isscalar(normalized_score) or normalized_score.ndim == 0:  # type: ignore[union-attr]
+            normalized_score = (
+                np.clip(normalized_score, a_min=100.0, a_max=None)
+                if is_anomaly_mask
+                else np.clip(normalized_score, a_min=None, a_max=99.99)
+            )
+        else:
+            normalized_score = cast(np.ndarray, normalized_score)
+            normalized_score[is_anomaly_mask] = np.clip(normalized_score[is_anomaly_mask], a_min=100.0, a_max=None)
+            normalized_score[is_not_anomaly_mask] = np.clip(
+                normalized_score[is_not_anomaly_mask], a_min=None, a_max=99.99
+            )
+
     if isinstance(normalized_score, torch.Tensor):
         return torch.clamp(normalized_score, 0.0, 1000.0)
 

{quadra-2.5.0.dist-info → quadra-2.6.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: quadra
-Version: 2.5.0
+Version: 2.6.0
 Summary: Deep Learning experiment orchestration library
 License: Apache-2.0
 License-File: LICENSE
@@ -22,7 +22,7 @@ Provides-Extra: onnx
 Requires-Dist: albumentations (>=1.3,<1.4)
 Requires-Dist: anomalib-orobix (==0.7.0.dev150)
 Requires-Dist: boto3 (>=1.26,<1.27)
-Requires-Dist: grad-cam-orobix (==1.5.3.dev001)
+Requires-Dist: grad-cam-orobix (==1.5.3.dev1)
 Requires-Dist: h5py (>=3.8,<3.9)
 Requires-Dist: hydra_colorlog (>=1.2,<1.3)
 Requires-Dist: hydra_core (>=1.3,<1.4)
@@ -35,7 +35,7 @@ Requires-Dist: numpy (<2)
 Requires-Dist: nvitop (>=0.11,<0.12)
 Requires-Dist: onnx (==1.15.0) ; extra == "onnx"
 Requires-Dist: onnxconverter-common (>=1.14.0,<2.0.0) ; extra == "onnx"
-Requires-Dist: onnxruntime_gpu (==1.20.0) ; extra == "onnx"
+Requires-Dist: onnxruntime_gpu (==1.23.2) ; extra == "onnx"
 Requires-Dist: onnxsim (==0.4.28) ; extra == "onnx"
 Requires-Dist: opencv_python_headless (>=4.7.0,<4.8.0)
 Requires-Dist: pandas (<2.0)
@@ -44,17 +44,17 @@ Requires-Dist: pydantic (>=1.10.10)
 Requires-Dist: python_dotenv (>=0.21,<0.22)
 Requires-Dist: pytorch_lightning (>=2.4,<2.5)
 Requires-Dist: rich (>=13.2,<13.3)
-Requires-Dist: scikit_learn (>=1.2,<1.3)
+Requires-Dist: scikit_learn (>=1.6,<1.7)
 Requires-Dist: scikit_multilearn (>=0.2,<0.3)
 Requires-Dist: seaborn (>=0.12,<0.13)
 Requires-Dist: segmentation_models_pytorch-orobix (==0.3.3.dev1)
 Requires-Dist: tensorboard (>=2.11,<2.12)
 Requires-Dist: timm (==0.9.12)
-Requires-Dist: torch (==2.6.0)
+Requires-Dist: torch (==2.8.0)
 Requires-Dist: torchinfo (>=1.8,<1.9)
 Requires-Dist: torchmetrics (>=0.10,<0.11)
 Requires-Dist: torchsummary (>=1.5,<1.6)
-Requires-Dist: torchvision (>=0.21,<0.22)
+Requires-Dist: torchvision (==0.23)
 Requires-Dist: tripy (>=1.0,<1.1)
 Requires-Dist: typing_extensions (==4.11.0) ; python_version < "3.10"
 Requires-Dist: xxhash (>=3.2,<3.3)

{quadra-2.5.0.dist-info → quadra-2.6.0.dist-info}/RECORD

@@ -1,4 +1,4 @@
-quadra/__init__.py,sha256=my_GwHSTlCXOqpoWuPRTAF1_rKRUMylmeA4eWoMa96g,112
+quadra/__init__.py,sha256=ylxTD-7xItBBl8y254g9lUc-ehYh4LCjfEoDhXhBfUw,112
 quadra/callbacks/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 quadra/callbacks/anomalib.py,sha256=WLBEGhZA9HoP4Yh9UbbC2GzDOKYTkvU9EY1lkZcV7Fs,11971
 quadra/callbacks/lightning.py,sha256=qvtzDiv8ZUV7K11gKHKWCyo-a9XR_Jm_M-IEicTM1Yo,20242
@@ -59,7 +59,7 @@ quadra/configs/experiment/base/anomaly/fastflow.yaml,sha256=jN3TeJXPGT7_GOhu2Eac
 quadra/configs/experiment/base/anomaly/inference.yaml,sha256=aLS3U0yC0Yb17BD1NSl-nhjJ9fcV6jDmxVQrrwZoMYI,446
 quadra/configs/experiment/base/anomaly/padim.yaml,sha256=5trGY5kL7gKzRTQuWT-LHYVPf4-q9J2I_-196noIZd4,829
 quadra/configs/experiment/base/anomaly/patchcore.yaml,sha256=a795iOcdH6kSoeA7ufChjBGWYdBn0ztRVtLBk1vpI3Q,841
-quadra/configs/experiment/base/classification/classification.yaml,sha256=EYr_nGKvRe5XkBkX6T83g9Qg9dlwXnaTwEMvthBK7ZU,1307
+quadra/configs/experiment/base/classification/classification.yaml,sha256=MNWrzDPpTIBc5gfIyeZmEiLqeI-FPjXJt2APD4gY0gU,1290
 quadra/configs/experiment/base/classification/classification_evaluation.yaml,sha256=r_pWUr9MbZYXlJlKIZIRb3GQfuGkuFOMjSOiZrt5fdU,463
 quadra/configs/experiment/base/classification/multilabel_classification.yaml,sha256=I8g8TUMW1q6uoa_pT_V9swwsVzS4ouZxUN3Dz424bJQ,774
 quadra/configs/experiment/base/classification/sklearn_classification.yaml,sha256=Mun6mabSBtFoOo1zE96syfoVNBmonmSmw7HAfEx3D5s,519
@@ -148,8 +148,8 @@ quadra/configs/optimizer/adamw.yaml,sha256=bofKJKqkjhtCf07KrARdUxMpQO6ux8UKGG2c0
 quadra/configs/optimizer/default.yaml,sha256=_JNfHj0JnkggcHKorPNJ_dSAJdkifNapve8VlHRcIVg,69
 quadra/configs/optimizer/lars.yaml,sha256=NwBKZNLQ5lbPAiWNPgmX2W4ap0PW3h-LpWKF6Qz35Vs,138
 quadra/configs/optimizer/sgd.yaml,sha256=_JNfHj0JnkggcHKorPNJ_dSAJdkifNapve8VlHRcIVg,69
-quadra/configs/scheduler/default.yaml,sha256=RamJr7-wjTHw6xCV9pseRfh72yScMV7B0gGvwsZv9ko,109
-quadra/configs/scheduler/rop.yaml,sha256=8t1vR5IX8d21ZC9S8vXAeAs0OtzvOcemV2gWqbbfoAM,111
+quadra/configs/scheduler/default.yaml,sha256=vmLBhrgBLs9DDXC69CRIbTkcKiGT11vfrq-hJNrYWy4,94
+quadra/configs/scheduler/rop.yaml,sha256=Z--NTA5X60asZ3nm85dNucJL7h1J4cICtryA9Bc-KbI,96
 quadra/configs/scheduler/step.yaml,sha256=0pj1rENjeeke8XQUhdrWrKD7eaoQ2a9-H25Deww9w3o,67
 quadra/configs/scheduler/warmrestart.yaml,sha256=cD7pPYFwBgCqLTCnYkrOW24Fd-8CiLgs88rsF0NHoyw,90
 quadra/configs/scheduler/warmup.yaml,sha256=2vv928tgNOE7yelnBSL2DVslqS3vqGz39LVXnZCJkPU,198
@@ -248,7 +248,7 @@ quadra/schedulers/__init__.py,sha256=mQivr18c0j36hpV3Lm8nlyBVKFevWp8TtLuTfvI9kQc
 quadra/schedulers/base.py,sha256=T1EdrLOJ0i9MzWoLCkrNA0uypm7hJ-L6NFhjIXFB6NE,1462
 quadra/schedulers/warmup.py,sha256=chzzrK7OqqlicBCxiF4CqMYNrWu6nflIbRE-C86Jrw0,4962
 quadra/tasks/__init__.py,sha256=tmAfMoH0k3UC7r2pNrgbBa1Pfc3tpLl3IObFF6Z0eRE,820
-quadra/tasks/anomaly.py,sha256=RHeiM1vZF1zsva37iYdiGx_HLgdAp8lXnmUzXja69YU,24638
+quadra/tasks/anomaly.py,sha256=40ciWnkle45Hl6SjY_4OHWdT2Mb1Qfoog60alZG-uQ0,25899
 quadra/tasks/base.py,sha256=piYlTFtvqH-4s4oEq4GczdAs_gL29UHAJGsOC5Sd3Bc,14187
 quadra/tasks/classification.py,sha256=_GQOPMGuOZ_uLA9jFhLEaJkW_Sid_WKHNn9ALXnGNmo,53407
 quadra/tasks/patch.py,sha256=nzo8o-ei7iF1Iarvd8-c08s0Rs_lPvVPDLAbkFMx-Qw,20251
@@ -258,7 +258,7 @@ quadra/trainers/README.md,sha256=XtpbUOxwvPpOUL7E5s2JHjRgwT-CRKTxsBeUSXrg9BU,248
 quadra/trainers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 quadra/trainers/classification.py,sha256=YeJ0z7Vk0-dsMTcoKBxSdSA0rxtilEcQTp-Zq9Xi1hw,7042
 quadra/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-quadra/utils/anomaly.py,sha256=49vFvT5-4SxczsEM2Akcut_M1DDwKlOVdGv36oLTgR0,4067
+quadra/utils/anomaly.py,sha256=uyzrFTz5QFTyIEbOT81A6OxXbHceGeiGP-JP3qr65D8,6044
 quadra/utils/classification.py,sha256=dKFuv4RywWhvhstOnEOnaf-6qcViUK0dTgah9m9mw2Q,24917
 quadra/utils/deprecation.py,sha256=zF_S-yqenaZxRBOudhXts0mX763WjEUWCnHd09TZnwY,852
 quadra/utils/evaluation.py,sha256=oooRJPu1AaHhOwvB1Y6SFjQ645OkgrDzKtUvwWq8oq4,19005
@@ -293,8 +293,8 @@ quadra/utils/validator.py,sha256=wmVXycB90VNyAbKBUVncFCxK4nsYiOWJIY3ISXwxYCY,463
 quadra/utils/visualization.py,sha256=yYm7lPziUOlybxigZ2qTycNewb67Q80H4hjQGWUh788,16094
 quadra/utils/vit_explainability.py,sha256=Gh6BHaDEzWxOjJp1aqvCxLt9Rb8TXd5uKXOAx7-acUk,13351
 hydra_plugins/quadra_searchpath_plugin.py,sha256=AAn4TzR87zUK7nwSsK-KoqALiPtfQ8FvX3fgZPTGIJ0,1189
-quadra-2.5.0.dist-info/METADATA,sha256=s-yGU-tHTS03MQtY5lXMvPv739qgVmz4ZvdDJs_J4FE,17632
-quadra-2.5.0.dist-info/WHEEL,sha256=zp0Cn7JsFoX2ATtOhtaFYIiE2rmFAD4OcMhtUki8W3U,88
-quadra-2.5.0.dist-info/entry_points.txt,sha256=sRYonBZyx-sAJeWcQNQoVQIU5lm02cnCQt6b15k0WHU,43
-quadra-2.5.0.dist-info/licenses/LICENSE,sha256=8cTbQtcWa02YJoSpMeV_gxj3jpMTkxvl-w3WJ5gV_QE,11342
-quadra-2.5.0.dist-info/RECORD,,
+quadra-2.6.0.dist-info/METADATA,sha256=eAbmWIkIwEc5poNJHwM_9gEu4gH6uXmGlDv2HOfO7iA,17624
+quadra-2.6.0.dist-info/WHEEL,sha256=3ny-bZhpXrU6vSQ1UPG34FoxZBp3lVcvK0LkgUz6VLk,88
+quadra-2.6.0.dist-info/entry_points.txt,sha256=sRYonBZyx-sAJeWcQNQoVQIU5lm02cnCQt6b15k0WHU,43
+quadra-2.6.0.dist-info/licenses/LICENSE,sha256=8cTbQtcWa02YJoSpMeV_gxj3jpMTkxvl-w3WJ5gV_QE,11342
+quadra-2.6.0.dist-info/RECORD,,

{quadra-2.5.0.dist-info → quadra-2.6.0.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: poetry-core 2.2.1
+Generator: poetry-core 2.3.0
 Root-Is-Purelib: true
 Tag: py3-none-any