Merge pull request #86 from KevinMusgrave/dev

v0.0.79

format_code.sh

@@ -1,4 +1,4 @@
-black . --exclude examples
-isort . --profile black --skip-glob examples
+black src tests
+isort src tests --profile black
 nbqa black examples
 nbqa isort examples --profile black

run_linter.sh

@@ -1,2 +1,2 @@
-flake8 . --count --show-source --statistics
+flake8 src tests --count --show-source --statistics
 nbqa flake8 examples --count --show-source --statistics

setup.py

@@ -9,10 +9,10 @@
     long_description = fh.read()
 
 
-extras_require_detection = ["albumentations"]
+extras_require_detection = ["albumentations >= 1.2.1"]
 extras_require_ignite = ["pytorch-ignite == 0.4.9"]
 extras_require_lightning = ["pytorch-lightning"]
-extras_require_record_keeper = ["record-keeper >= 0.9.31"]
+extras_require_record_keeper = ["record-keeper >= 0.9.32"]
 extras_require_timm = ["timm"]
 extras_require_docs = [
     "mkdocs-material",
@@ -44,8 +44,8 @@
         "numpy",
         "torch",
         "torchvision",
-        "torchmetrics",
-        "pytorch-metric-learning >= 1.3.1.dev0",
+        "torchmetrics >= 0.9.3",
+        "pytorch-metric-learning >= 1.5.2",
     ],
     extras_require={
         "detection": extras_require_detection,

src/pytorch_adapt/__init__.py

@@ -1 +1 @@
-__version__ = "0.0.78"
+__version__ = "0.0.79"

src/pytorch_adapt/adapters/__init__.py

@@ -2,7 +2,7 @@
 from .adda import ADDA
 from .aligner import RTN, Aligner
 from .base_adapter import BaseAdapter
-from .classifier import Classifier, Finetuner
+from .classifier import Classifier, Finetuner, MultiLabelClassifier
 from .dann import CDANNE, DANN, DANNE, GVB, GVBE
 from .gan import CDAN, CDANE, GAN, GANE, VADA, DomainConfusion
 from .mcd import MCD

src/pytorch_adapt/adapters/adabn.py

@@ -14,8 +14,6 @@ class AdaBN(BaseAdapter):
     |models|```["G", "C"]```|
     """
 
-    hook_cls = AdaBNHook
-
     def __init__(self, *args, inference_fn=None, **kwargs):
         """
         Arguments:
@@ -27,6 +25,10 @@ def __init__(self, *args, inference_fn=None, **kwargs):
     def init_hook(self, hook_kwargs):
         self.hook = self.hook_cls(**hook_kwargs)
 
+    @property
+    def hook_cls(self):
+        return AdaBNHook
+
     def get_key_enforcer(self) -> KeyEnforcer:
         return KeyEnforcer(models=["G", "C"], optimizers=[])
 

src/pytorch_adapt/adapters/adda.py

@@ -20,8 +20,6 @@ class ADDA(BaseAdapter):
     The target model ("T") is created during initialization by deep-copying the G model.
     """
 
-    hook_cls = ADDAHook
-
     def __init__(self, *args, inference_fn=None, **kwargs):
         """
         Arguments:
@@ -55,3 +53,7 @@ def init_hook(self, hook_kwargs):
     def init_containers_and_check_keys(self, containers):
         containers["models"]["T"] = copy.deepcopy(containers["models"]["G"])
         super().init_containers_and_check_keys(containers)
+
+    @property
+    def hook_cls(self):
+        return ADDAHook

src/pytorch_adapt/adapters/aligner.py

@@ -16,12 +16,14 @@ class Aligner(BaseGCAdapter):
     |optimizers|```["G", "C"]```|
     """
 
-    hook_cls = AlignerPlusCHook
-
     def init_hook(self, hook_kwargs):
         opts = with_opt(list(self.optimizers.keys()))
         self.hook = self.hook_cls(opts, **hook_kwargs)
 
+    @property
+    def hook_cls(self):
+        return AlignerPlusCHook
+
 
 class RTN(Aligner):
     """
@@ -34,8 +36,6 @@ class RTN(Aligner):
     |misc|```["feature_combiner"]```|
     """
 
-    hook_cls = RTNHook
-
     def __init__(self, *args, inference_fn=None, **kwargs):
         """
         Arguments:
@@ -44,6 +44,10 @@ def __init__(self, *args, inference_fn=None, **kwargs):
         inference_fn = c_f.default(inference_fn, rtn_fn)
         super().__init__(*args, inference_fn=inference_fn, **kwargs)
 
+    @property
+    def hook_cls(self):
+        return RTNHook
+
     def get_key_enforcer(self) -> KeyEnforcer:
         ke = super().get_key_enforcer()
         ke.requirements["models"].append("residual_model")

src/pytorch_adapt/adapters/base_adapter.py

@@ -150,6 +150,11 @@ def init_hook(self):
         """
         pass
 
+    @property
+    @abstractmethod
+    def hook_cls(self):
+        pass
+
     def init_containers_and_check_keys(self, containers):
         """
         Called in ```__init__``` before

src/pytorch_adapt/adapters/classifier.py

@@ -1,5 +1,5 @@
 from ..containers import KeyEnforcer, MultipleContainers, Optimizers
-from ..hooks import ClassifierHook, FinetunerHook
+from ..hooks import ClassifierHook, FinetunerHook, MultiLabelClassifierHook
 from .base_adapter import BaseGCAdapter
 from .utils import default_optimizer_tuple, with_opt
 
@@ -14,12 +14,14 @@ class Classifier(BaseGCAdapter):
     |optimizers|```["G", "C"]```|
     """
 
-    hook_cls = ClassifierHook
-
     def init_hook(self, hook_kwargs):
         opts = with_opt(list(self.optimizers.keys()))
         self.hook = self.hook_cls(opts=opts, **hook_kwargs)
 
+    @property
+    def hook_cls(self):
+        return ClassifierHook
+
 
 class Finetuner(Classifier):
     """
@@ -31,7 +33,9 @@ class Finetuner(Classifier):
     |optimizers|```["C"]```|
     """
 
-    hook_cls = FinetunerHook
+    @property
+    def hook_cls(self):
+        return FinetunerHook
 
     def get_default_containers(self) -> MultipleContainers:
         optimizers = Optimizers(default_optimizer_tuple(), keys=["C"])
@@ -41,3 +45,9 @@ def get_key_enforcer(self) -> KeyEnforcer:
         ke = super().get_key_enforcer()
         ke.requirements["optimizers"].remove("G")
         return ke
+
+
+class MultiLabelClassifier(Classifier):
+    @property
+    def hook_cls(self):
+        return MultiLabelClassifierHook

src/pytorch_adapt/adapters/dann.py

@@ -15,19 +15,25 @@ class DANN(BaseGCDAdapter):
     |optimizers|```["G", "C", "D"]```|
     """
 
-    hook_cls = DANNHook
-
     def init_hook(self, hook_kwargs):
         opts = with_opt(list(self.optimizers.keys()))
         self.hook = self.hook_cls(opts=opts, **hook_kwargs)
 
+    @property
+    def hook_cls(self):
+        return DANNHook
+
 
 class DANNE(DANN):
-    hook_cls = DANNEHook
+    @property
+    def hook_cls(self):
+        return DANNEHook
 
 
 class CDANNE(DANN, CDAN):
-    hook_cls = CDANNEHook
+    @property
+    def hook_cls(self):
+        return CDANNEHook
 
 
 class GVB(DANN):
@@ -44,7 +50,9 @@ class GVB(DANN):
     with each bridge being a re-initialized copy of each model.
     """
 
-    hook_cls = GVBHook
+    @property
+    def hook_cls(self):
+        return GVBHook
 
     def init_containers_and_check_keys(self, containers):
         models = containers["models"]
@@ -55,4 +63,6 @@ def init_containers_and_check_keys(self, containers):
 
 
 class GVBE(GVB):
-    hook_cls = GVBEHook
+    @property
+    def hook_cls(self):
+        return GVBEHook

src/pytorch_adapt/adapters/gan.py

@@ -23,16 +23,20 @@ class GAN(BaseGCDAdapter):
     |optimizers|```["G", "C", "D"]```|
     """
 
-    hook_cls = GANHook
-
     def init_hook(self, hook_kwargs):
         g_opts = with_opt(["G", "C"])
         d_opts = with_opt(["D"])
         self.hook = self.hook_cls(d_opts=d_opts, g_opts=g_opts, **hook_kwargs)
 
+    @property
+    def hook_cls(self):
+        return GANHook
+
 
 class GANE(GAN):
-    hook_cls = GANEHook
+    @property
+    def hook_cls(self):
+        return GANEHook
 
 
 class CDAN(GAN):
@@ -46,7 +50,9 @@ class CDAN(GAN):
     |misc|```["feature_combiner"]```|
     """
 
-    hook_cls = CDANHook
+    @property
+    def hook_cls(self):
+        return CDANHook
 
     def get_key_enforcer(self) -> KeyEnforcer:
         ke = super().get_key_enforcer()
@@ -55,7 +61,9 @@ def get_key_enforcer(self) -> KeyEnforcer:
 
 
 class CDANE(CDAN):
-    hook_cls = CDANEHook
+    @property
+    def hook_cls(self):
+        return CDANEHook
 
 
 class DomainConfusion(GAN):
@@ -68,7 +76,9 @@ class DomainConfusion(GAN):
     |optimizers|```["G", "C", "D"]```|
     """
 
-    hook_cls = DomainConfusionHook
+    @property
+    def hook_cls(self):
+        return DomainConfusionHook
 
 
 class VADA(GAN):
@@ -86,7 +96,9 @@ class VADA(GAN):
     automatically during initialization.
     """
 
-    hook_cls = VADAHook
+    @property
+    def hook_cls(self):
+        return VADAHook
 
     def init_containers_and_check_keys(self, containers):
         models = containers["models"]

src/pytorch_adapt/adapters/mcd.py

@@ -19,8 +19,6 @@ class MCD(BaseGCAdapter):
     classifiers is 2, so C should output ```[logits1, logits2]```.
     """
 
-    hook_cls = MCDHook
-
     def __init__(self, *args, inference_fn=None, **kwargs):
         inference_fn = c_f.default(inference_fn, mcd_fn)
         super().__init__(*args, inference_fn=inference_fn, **kwargs)
@@ -29,3 +27,7 @@ def init_hook(self, hook_kwargs):
         self.hook = self.hook_cls(
             g_opts=with_opt(["G"]), c_opts=with_opt(["C"]), **hook_kwargs
         )
+
+    @property
+    def hook_cls(self):
+        return MCDHook

src/pytorch_adapt/adapters/symnets.py

@@ -17,8 +17,6 @@ class SymNets(BaseGCAdapter):
     The C model must output a list of logits: ```[logits1, logits2]```.
     """
 
-    hook_cls = SymNetsHook
-
     def __init__(self, *args, inference_fn=None, **kwargs):
         inference_fn = c_f.default(inference_fn, symnets_fn)
         super().__init__(*args, inference_fn=inference_fn, **kwargs)
@@ -27,3 +25,7 @@ def init_hook(self, hook_kwargs):
         self.hook = self.hook_cls(
             g_opts=with_opt(["G"]), c_opts=with_opt(["C"]), **hook_kwargs
         )
+
+    @property
+    def hook_cls(self):
+        return SymNetsHook

src/pytorch_adapt/datasets/base_dataset.py

@@ -61,5 +61,6 @@ def download_dataset(self, root):
         download_url(self.url, root, filename=self.filename, md5=self.md5)
         filepath = os.path.join(root, self.filename)
         decompressor = tarfile.open if tarfile.is_tarfile(filepath) else zipfile.ZipFile
+        c_f.LOGGER.info("Extracting")
         with decompressor(filepath, "r") as f:
             f.extractall(path=root, members=c_f.extract_progress(f))

src/pytorch_adapt/datasets/utils.py

@@ -33,3 +33,14 @@ def maybe_download(fn, kwargs):
             fn(**kwargs)
         else:
             raise
+
+
+def num_classes(dataset_name):
+    return {
+        "mnist": 10,
+        "domainnet": 345,
+        "domainnet126": 126,
+        "office31": 31,
+        "officehome": 65,
+        "voc_multilabel": 20,
+    }[dataset_name]

src/pytorch_adapt/frameworks/ignite/__init__.py

@@ -1,4 +1,5 @@
 from .checkpoint_utils import CheckpointFnCreator
 from .ignite import Ignite
+from .ignite_multilabel_classification import IgniteMultiLabelClassification
 from .ignite_preds_as_features import IgnitePredsAsFeatures
 from .ignite_val_hook_wrapper import IgniteValHookWrapper

src/pytorch_adapt/frameworks/ignite/ignite.py

@@ -30,6 +30,7 @@ def __init__(
         with_pbars=True,
         device=None,
         auto_dist=True,
+        val_output_dict_fn=None,
     ):
         """
         Arguments:
@@ -77,6 +78,9 @@ def __init__(
         self.log_freq = log_freq
         self.with_pbars = with_pbars
         self.device = c_f.default(device, idist.device, {})
+        self.val_output_dict_fn = c_f.default(
+            val_output_dict_fn, f_utils.create_output_dict
+        )
         self.trainer_init()
         self.collector_init()
         self.dist_init_done = False
@@ -325,7 +329,7 @@ def evaluate_best_model(
     def get_collector_step(self, inference):
         def collector_step(engine, batch):
             batch = c_f.batch_to_device(batch, self.device)
-            return f_utils.collector_step(inference, batch, f_utils.create_output_dict)
+            return f_utils.collector_step(inference, batch, self.val_output_dict_fn)
 
         return collector_step