new Changes

Almost Completed!

.gitignore

@@ -9,6 +9,7 @@ wandb/
 protoruns/
 **/__pycache__/
 **/**/__pycache__/
+**/**/**/__pycache__/
 logs/
 *.npy
 centroids
@@ -17,3 +18,5 @@ model/
 OfficeHomeDataset_10072016.zip
 dalib/domainbed/__pycache__/*
 degaa
+**/tensorboard/
+san/

common/vision/datasets/Concatenate/__init__.py

@@ -0,0 +1,54 @@
+import bisect
+import warnings
+
+from  torch.utils.data import Dataset, IterableDataset
+# from torch.utils.data.dataset import ConcatDataset
+
+
+class ConcatenateDataset(Dataset):
+    r"""Dataset as a concatenation of multiple datasets.
+
+    This class is useful to assemble different existing datasets.
+
+    Arguments:
+        datasets (sequence): List of datasets to be concatenated
+    """
+
+    @staticmethod
+    def cumsum(sequence):
+        r, s = [], 0
+        for e in sequence:
+            l = len(e)
+            r.append(l + s)
+            s += l
+        return r
+
+    def __init__(self, datasets) -> None:
+        super(ConcatenateDataset, self).__init__()
+        # Cannot verify that datasets is Sized
+        assert len(datasets) > 0, 'datasets should not be an empty iterable'  # type: ignore
+        self.datasets = list(datasets)
+        for d in self.datasets:
+            assert not isinstance(d, IterableDataset), "ConcatenateDataset does not support IterableDataset"
+        self.cumulative_sizes = self.cumsum(self.datasets)
+
+    def __len__(self):
+        return self.cumulative_sizes[-1]
+
+    def __getitem__(self, idx):
+        if idx < 0:
+            if -idx > len(self):
+                raise ValueError("absolute value of index should not exceed dataset length")
+            idx = len(self) + idx
+        dataset_idx = bisect.bisect_right(self.cumulative_sizes, idx)
+        if dataset_idx == 0:
+            sample_idx = idx
+        else:
+            sample_idx = idx - self.cumulative_sizes[dataset_idx - 1]
+        return self.datasets[dataset_idx][sample_idx], self.datasets[dataset_idx].domain_index
+
+    @property
+    def cummulative_sizes(self):
+        warnings.warn("cummulative_sizes attribute is renamed to "
+                      "cumulative_sizes", DeprecationWarning, stacklevel=2)
+        return self.cumulative_sizes

dalib/domainbed/algorithms_proto.py

@@ -44,6 +44,7 @@ class Proto(nn.Module):
     """
 
     def __init__(self, input_shape, num_classes, num_domains, hparams, use_relu=True):
+    # def __init__(self, hparams, use_relu=True):
         super(Proto, self).__init__()
 
         self.hparams = hparams
@@ -60,19 +61,14 @@ def __init__(self, input_shape, num_classes, num_domains, hparams, use_relu=True
         #     weight_decay=self.hparams['weight_decay']
         # )
 
-        # initializing constants
-        self.nd = num_domains
-        self.nc = num_classes
-
         # initializing architecture parameters
-        featurizer = networks.Featurizer(input_shape, self.hparams)
+        featurizer = networks.Featurizer(hparams["input_shape"], self.hparams)
         self.ft_output_size = featurizer.n_outputs
         self.proto_size = int(self.ft_output_size * 0.25)
         self.feat_size = int(self.ft_output_size)
 
         # initializing hyperparameters
         self.proto_frac = hparams["proto_train_frac"]
-        self.epochs = hparams["n_steps"]
         self.proto_epochs = hparams["n_steps_proto"]
 
         # self.kernel_type = "gaussian"

dalib/domainbed/hparams_registry.py

@@ -27,6 +27,7 @@ def _hparams(algorithm, dataset, random_state):
     hparams["dataset"] = (dataset, dataset)
     hparams["domains_per_iter"] = (4, 4)
     hparams["data_parallel"] = (True, True)
+    hparams["input_shape"] = ((3, 224, 224,), (3, 224, 224,))
 
     if dataset in RESNET_DATASETS:
         hparams["lr"] = (1e-4, 10 ** random_state.uniform(-5.5, -3.5))
@@ -44,6 +45,7 @@ def _hparams(algorithm, dataset, random_state):
         hparams["weight_decay"] = (1e-4, 10 ** random_state.uniform(-4, -4))
 
     hparams["class_balanced"] = (False, False)
+    hparams["num_proto_extraction_points"] = (500, 500)
 
     if algorithm in ["DANN", "CDANN"]:
 

scripts/adapt.sh

@@ -0,0 +1 @@
+CUDA_VISIBLE_DEVICES=1 python src/degaa.py --output_dir ./adapt/run1 --tensorboard --source Ar,Pr --target Cl,Rw --batch_size 32

scripts/embeddings.sh

@@ -0,0 +1,3 @@
+# CUDA_VISIBLE_DEVICES=0 python src/embeddings_old.py --output_dir ./protoruns/run4 --batch_size 48 --num_proto_steps 100000 --tensorboard --resume 60000
+
+CUDA_VISIBLE_DEVICES=1 python src/embeddings_old.py --output_dir ./protoruns/run5 --batch_size 32 --hparams '{"mixup":1}' --num_proto_steps 1000000 --tensorboard --resume 400100 --checkpoint_freq 10000

scripts/source_only.sh

@@ -2,10 +2,10 @@
 # python image_source_final.py --output san --gpu_id 0 --dset office --max_epoch 50 --s 0 --net vit
 
 #office-home
-python image_source_final.py --output weights --gpu_id 0 --dset office-home --max_epoch 50 --s 0 --net resnet50
-python image_source_final.py --output weights --gpu_id 0 --dset office-home --max_epoch 50 --s 1 --net resnet50
-python image_source_final.py --output weights --gpu_id 1 --dset office-home --max_epoch 50 --s 2 --net resnet50
-python image_source_final.py --output weights --gpu_id 0 --dset office-home --max_epoch 50 --s 3 --net resnet50
+python image_source_final.py --output weights --gpu_id 2 --batch_size 128 --dataset OfficeHome --max_epoch 50 --source Ar,Pr --target Cl,Rw --wandb 0
+# python image_source_final.py --output weights --gpu_id 0 --dataset OfficeHome --max_epoch 50 --source "Ar,Pr" --target "Cl,Rw"
+# python image_source_final.py --output weights --gpu_id 1 --dataset OfficeHome --max_epoch 50 --source "Ar,Pr" --target "Cl,Rw"
+# python image_source_final.py --output weights --gpu_id 0 --dataset OfficeHome --max_epoch 50 --source "Ar,Pr" --target "Cl,Rw"
 
 #pacs
 # python image_source_final.py --output san --gpu_id 0 --dset pacs --max_epoch 50 --s 0 --net vit