Import strategy

.coveragerc

@@ -1,13 +1,13 @@
 [run]
-source = src
+source = quanda
 omit =
     /tests/*
-    src/utils/explanations.py
-    src/utils/transforms.py
-    src/utils/datasets/transformed/sample.py
-    src/utils/cache.py
-    src/utils/datasets/activation_dataset.py
-    src/utils/datasets/indexed_subset.py
-    src/explainers/functional.py
+    quanda/utils/explanations.py
+    quanda/utils/transforms.py
+    quanda/utils/datasets/transformed/sample.py
+    quanda/utils/cache.py
+    quanda/utils/datasets/activation_dataset.py
+    quanda/utils/datasets/indexed_subset.py
+    quanda/explainers/functional.py
 [report]
 ignore_errors = True

Makefile

@@ -7,7 +7,7 @@ style:
 	black .
 	python -m flake8 . --pytest-parametrize-names-type=csv
 	python -m isort .
-	python -m mypy src --check-untyped-defs
+	python -m mypy quanda --check-untyped-defs
 	rm -f .coverage
 	rm -f .coverage.*
 	find . | grep -E "(__pycache__|\.pyc|\.pyo)" | xargs rm -rf

README.md

@@ -32,12 +32,10 @@ Excerpts from `tutorials/usage_testing.py`:
 <summary><b><big>Step 1. Import library components</big></b></summary>
 
 ```python
-from src.explainers.wrappers.captum_influence import captum_similarity_explain, CaptumSimilarity
-from src.metrics.localization.class_detection import ClassDetectionMetric
-from src.metrics.randomization.model_randomization import (
-    ModelRandomizationMetric,
-)
-from src.metrics.unnamed.top_k_overlap import TopKOverlapMetric
+from quanda.explainers.wrappers import captum_similarity_explain, CaptumSimilarity
+from quanda.metrics.localization import ClassDetectionMetric
+from quanda.metrics.randomization import ModelRandomizationMetric
+from quanda.metrics.unnamed.top_k_overlap import TopKOverlapMetric
 ```
 </details>
 

pyproject.toml

@@ -53,6 +53,9 @@ exclude = '''
 testpaths = ["tests"]
 python_files = "test_*.py"
 
+[tool.setuptools]
+py-modules = []
+
 [project.optional-dependencies]
 dev = [ # Install wtih pip install .[dev] or pip install -e '.[dev]' in zsh
     "coverage>=7.2.3",

quanda/__init__.py

@@ -0,0 +1,3 @@
+from quanda import explainers, metrics, toy_benchmarks, utils
+
+__all__ = ["explainers", "metrics", "toy_benchmarks", "utils"]

quanda/explainers/__init__.py

@@ -0,0 +1,23 @@
+from quanda.explainers import utils, wrappers
+from quanda.explainers.aggregators import (
+    AbsSumAggregator,
+    BaseAggregator,
+    SumAggregator,
+    aggr_types,
+)
+from quanda.explainers.base import BaseExplainer
+from quanda.explainers.functional import ExplainFunc, ExplainFuncMini
+from quanda.explainers.random import RandomExplainer
+
+__all__ = [
+    "BaseExplainer",
+    "RandomExplainer",
+    "ExplainFunc",
+    "ExplainFuncMini",
+    "utils",
+    "wrappers",
+    "BaseAggregator",
+    "SumAggregator",
+    "AbsSumAggregator",
+    "aggr_types",
+]

src/explainers/base.py → quanda/explainers/base.py

@@ -3,8 +3,8 @@
 
 import torch
 
-from src.utils.common import cache_result
-from src.utils.validation import validate_1d_tensor_or_int_list
+from quanda.utils.common import cache_result
+from quanda.utils.validation import validate_1d_tensor_or_int_list
 
 
 class BaseExplainer(ABC):

src/explainers/random.py → quanda/explainers/random.py

@@ -2,8 +2,8 @@
 
 import torch
 
-from src.explainers.base import BaseExplainer
-from src.utils.common import cache_result
+from quanda.explainers import BaseExplainer
+from quanda.utils.common import cache_result
 
 
 class RandomExplainer(BaseExplainer):

quanda/explainers/wrappers/__init__.py

@@ -0,0 +1,25 @@
+from quanda.explainers.wrappers.captum_influence import (
+    CaptumArnoldi,
+    CaptumInfluence,
+    CaptumSimilarity,
+    CaptumTracInCP,
+    captum_arnoldi_explain,
+    captum_arnoldi_self_influence,
+    captum_similarity_explain,
+    captum_similarity_self_influence,
+    captum_tracincp_explain,
+    captum_tracincp_self_influence,
+)
+
+__all__ = [
+    "CaptumInfluence",
+    "CaptumSimilarity",
+    "captum_similarity_explain",
+    "captum_similarity_self_influence",
+    "CaptumArnoldi",
+    "captum_arnoldi_explain",
+    "captum_arnoldi_self_influence",
+    "CaptumTracInCP",
+    "captum_tracincp_explain",
+    "captum_tracincp_self_influence",
+]

src/explainers/wrappers/captum_influence.py → quanda/explainers/wrappers/captum_influence.py

@@ -11,14 +11,14 @@
     ArnoldiInfluenceFunction,
 )
 
-from src.explainers.base import BaseExplainer
-from src.explainers.utils import (
+from quanda.explainers.base import BaseExplainer
+from quanda.explainers.utils import (
     explain_fn_from_explainer,
     self_influence_fn_from_explainer,
 )
-from src.utils.common import get_load_state_dict_func
-from src.utils.functions.similarities import cosine_similarity
-from src.utils.validation import validate_checkpoints_load_func
+from quanda.utils.common import get_load_state_dict_func
+from quanda.utils.functions import cosine_similarity
+from quanda.utils.validation import validate_checkpoints_load_func
 
 
 class CaptumInfluence(BaseExplainer, ABC):

quanda/metrics/__init__.py

@@ -0,0 +1,16 @@
+from quanda.metrics import localization, randomization, unnamed
+from quanda.metrics.aggr_strategies import (
+    GlobalAggrStrategy,
+    GlobalSelfInfluenceStrategy,
+)
+from quanda.metrics.base import GlobalMetric, Metric
+
+__all__ = [
+    "Metric",
+    "GlobalMetric",
+    "GlobalAggrStrategy",
+    "GlobalSelfInfluenceStrategy",
+    "randomization",
+    "localization",
+    "unnamed",
+]

src/metrics/aggr_strategies.py → quanda/metrics/aggr_strategies.py

@@ -4,8 +4,7 @@
 
 import torch
 
-from src.explainers.aggregators import BaseAggregator
-from src.explainers.base import BaseExplainer
+from quanda.explainers import BaseAggregator, BaseExplainer
 
 
 class GlobalSelfInfluenceStrategy:

src/metrics/base.py → quanda/metrics/base.py

@@ -3,8 +3,8 @@
 
 import torch
 
-from src.explainers.aggregators import aggr_types
-from src.metrics.aggr_strategies import (
+from quanda.explainers import aggr_types
+from quanda.metrics.aggr_strategies import (
     GlobalAggrStrategy,
     GlobalSelfInfluenceStrategy,
 )

quanda/metrics/localization/__init__.py

@@ -0,0 +1,9 @@
+from quanda.metrics.localization.class_detection import ClassDetectionMetric
+from quanda.metrics.localization.mislabeling_detection import (
+    MislabelingDetectionMetric,
+)
+from quanda.metrics.localization.subclass_detection import (
+    SubclassDetectionMetric,
+)
+
+__all__ = ["ClassDetectionMetric", "SubclassDetectionMetric", "MislabelingDetectionMetric"]

src/metrics/localization/class_detection.py → quanda/metrics/localization/class_detection.py

@@ -2,7 +2,7 @@
 
 import torch
 
-from src.metrics.base import Metric
+from quanda.metrics.base import Metric
 
 
 class ClassDetectionMetric(Metric):

src/metrics/localization/mislabeling_detection.py → quanda/metrics/localization/mislabeling_detection.py

@@ -2,7 +2,7 @@
 
 import torch
 
-from src.metrics.base import GlobalMetric
+from quanda.metrics.base import GlobalMetric
 
 
 class MislabelingDetectionMetric(GlobalMetric):

src/metrics/localization/subclass_detection.py → quanda/metrics/localization/subclass_detection.py

@@ -1,6 +1,6 @@
 import torch
 
-from src.metrics.localization.class_detection import ClassDetectionMetric
+from quanda.metrics.localization import ClassDetectionMetric
 
 
 class SubclassDetectionMetric(ClassDetectionMetric):

quanda/metrics/randomization/__init__.py

@@ -0,0 +1,5 @@
+from quanda.metrics.randomization.model_randomization import (
+    ModelRandomizationMetric,
+)
+
+__all__ = ["ModelRandomizationMetric"]

src/metrics/randomization/model_randomization.py → quanda/metrics/randomization/model_randomization.py

@@ -3,12 +3,9 @@
 
 import torch
 
-from src.metrics.base import Metric
-from src.utils.common import get_parent_module_from_name
-from src.utils.functions.correlations import (
-    CorrelationFnLiterals,
-    correlation_functions,
-)
+from quanda.metrics.base import Metric
+from quanda.utils.common import get_parent_module_from_name
+from quanda.utils.functions import CorrelationFnLiterals, correlation_functions
 
 
 class ModelRandomizationMetric(Metric):

quanda/metrics/unnamed/__init__.py

@@ -0,0 +1,4 @@
+from quanda.metrics.unnamed.dataset_cleaning import DatasetCleaningMetric
+from quanda.metrics.unnamed.top_k_overlap import TopKOverlapMetric
+
+__all__ = ["DatasetCleaningMetric", "TopKOverlapMetric"]

src/metrics/unnamed/dataset_cleaning.py → quanda/metrics/unnamed/dataset_cleaning.py

@@ -4,9 +4,9 @@
 import lightning as L
 import torch
 
-from src.metrics.base import GlobalMetric
-from src.utils.common import class_accuracy
-from src.utils.training.trainer import BaseTrainer
+from quanda.metrics.base import GlobalMetric
+from quanda.utils.common import class_accuracy
+from quanda.utils.training import BaseTrainer
 
 
 class DatasetCleaningMetric(GlobalMetric):

src/metrics/unnamed/top_k_overlap.py → quanda/metrics/unnamed/top_k_overlap.py

@@ -1,6 +1,6 @@
 import torch
 
-from src.metrics.base import Metric
+from quanda.metrics.base import Metric
 
 
 class TopKOverlapMetric(Metric):

quanda/toy_benchmarks/__init__.py

@@ -0,0 +1,4 @@
+from quanda.toy_benchmarks import localization, randomization, unnamed
+from quanda.toy_benchmarks.base import ToyBenchmark
+
+__all__ = ["ToyBenchmark", "randomization", "localization", "unnamed"]

quanda/toy_benchmarks/localization/__init__.py

@@ -0,0 +1,9 @@
+from quanda.toy_benchmarks.localization.class_detection import ClassDetection
+from quanda.toy_benchmarks.localization.mislabeling_detection import (
+    MislabelingDetection,
+)
+from quanda.toy_benchmarks.localization.subclass_detection import (
+    SubclassDetection,
+)
+
+__all__ = ["ClassDetection", "SubclassDetection", "MislabelingDetection"]

src/toy_benchmarks/localization/class_detection.py → quanda/toy_benchmarks/localization/class_detection.py

@@ -3,8 +3,8 @@
 import torch
 from tqdm import tqdm
 
-from src.metrics.localization.class_detection import ClassDetectionMetric
-from src.toy_benchmarks.base import ToyBenchmark
+from quanda.metrics.localization import ClassDetectionMetric
+from quanda.toy_benchmarks.base import ToyBenchmark
 
 
 class ClassDetection(ToyBenchmark):

src/toy_benchmarks/localization/mislabeling_detection.py → quanda/toy_benchmarks/localization/mislabeling_detection.py

@@ -5,12 +5,14 @@
 import torch
 from tqdm import tqdm
 
-from src.metrics.localization.mislabeling_detection import (
+from quanda.metrics.localization.mislabeling_detection import (
     MislabelingDetectionMetric,
 )
-from src.toy_benchmarks.base import ToyBenchmark
-from src.utils.datasets.transformed.label_flipping import LabelFlippingDataset
-from src.utils.training.trainer import BaseTrainer
+from quanda.toy_benchmarks.base import ToyBenchmark
+from quanda.utils.datasets.transformed.label_flipping import (
+    LabelFlippingDataset,
+)
+from quanda.utils.training.trainer import BaseTrainer
 
 
 class MislabelingDetection(ToyBenchmark):
@@ -256,16 +258,16 @@ def evaluate(
             pbar = tqdm(expl_dl)
             n_batches = len(expl_dl)
 
-            for i, (input, labels) in enumerate(pbar):
+            for i, (inputs, labels) in enumerate(pbar):
                 pbar.set_description("Metric evaluation, batch %d/%d" % (i + 1, n_batches))
 
-                input, labels = input.to(device), labels.to(device)
+                inputs, labels = inputs.to(device), labels.to(device)
                 if use_predictions:
                     with torch.no_grad():
-                        targets = self.model(input).argmax(dim=-1)
+                        targets = self.model(inputs).argmax(dim=-1)
                 else:
                     targets = labels
-                explanations = explainer.explain(test=input, targets=targets)
+                explanations = explainer.explain(test=inputs, targets=targets)
                 metric.update(explanations)
         else:
             metric = MislabelingDetectionMetric.self_influence_based(

src/toy_benchmarks/localization/subclass_detection.py → quanda/toy_benchmarks/localization/subclass_detection.py

@@ -5,13 +5,13 @@
 import torch
 from tqdm import tqdm
 
-from src.metrics.localization.class_detection import ClassDetectionMetric
-from src.toy_benchmarks.base import ToyBenchmark
-from src.utils.datasets.transformed.label_grouping import (
+from quanda.metrics.localization.class_detection import ClassDetectionMetric
+from quanda.toy_benchmarks.base import ToyBenchmark
+from quanda.utils.datasets.transformed.label_grouping import (
     ClassToGroupLiterals,
     LabelGroupingDataset,
 )
-from src.utils.training.trainer import BaseTrainer
+from quanda.utils.training.trainer import BaseTrainer
 
 
 class SubclassDetection(ToyBenchmark):
@@ -231,19 +231,19 @@ def evaluate(
         pbar = tqdm(expl_dl)
         n_batches = len(expl_dl)
 
-        for i, (input, labels) in enumerate(pbar):
+        for i, (inputs, labels) in enumerate(pbar):
             pbar.set_description("Metric evaluation, batch %d/%d" % (i + 1, n_batches))
 
-            input, labels = input.to(device), labels.to(device)
+            inputs, labels = inputs.to(device), labels.to(device)
             grouped_labels = torch.tensor([self.class_to_group[i.item()] for i in labels], device=labels.device)
             if use_predictions:
                 with torch.no_grad():
-                    output = self.group_model(input)
+                    output = self.group_model(inputs)
                     targets = output.argmax(dim=-1)
             else:
                 targets = grouped_labels
             explanations = explainer.explain(
-                test=input,
+                test=inputs,
                 targets=targets,
             )
 

quanda/toy_benchmarks/randomization/__init__.py

@@ -0,0 +1,5 @@
+from quanda.toy_benchmarks.randomization.model_randomization import (
+    ModelRandomization,
+)
+
+__all__ = ["ModelRandomization"]