OOD files added

trust/utils/.ipynb_checkpoints/dermamnist-checkpoint.py

@@ -0,0 +1,204 @@
+import numpy as np
+import os
+import torch
+import torchvision
+from sklearn import datasets
+from torchvision import datasets, transforms
+import PIL.Image as Image
+from .utils import *
+np.random.seed(42)
+torch.manual_seed(42)
+from torchvision.datasets import cifar
+from torch.utils.data import Dataset, Subset, ConcatDataset, DataLoader
+
+class DermaDataset(Dataset):
+    def __init__(self, data, root='/mnt/data2/akshit/data/', transform=None):
+        self.root = root
+        self.transform = transform
+        self.images = data['images']
+        self.targets = data['labels'].flatten()
+
+
+    def __len__(self):
+        return len(self.targets)
+
+    def __getitem__(self, idx):
+        img = Image.fromarray(np.uint8(self.images[idx])).convert('RGB')
+        if self.transform:
+            img = self.transform(img)
+        label = self.targets[idx]
+        return img, label
+
+
+def getOODtargets(targets, sel_cls_idx, ood_cls_id):
+
+    ood_targets = []
+    targets_list = list(targets)
+    for i in range(len(targets_list)):
+        if(targets_list[i] in list(sel_cls_idx)):
+            ood_targets.append(targets_list[i])
+        else:
+            ood_targets.append(ood_cls_id)
+    print("num ood samples: ", ood_targets.count(ood_cls_id))
+    return torch.Tensor(ood_targets)
+
+def create_ood_data(fullset, testset, split_cfg, num_cls, augVal):
+
+    np.random.seed(42)
+    train_idx = []
+    val_idx = []
+    lake_idx = []
+    test_idx = [] 
+    selected_classes = np.array(list(range(split_cfg['num_cls_idc'])))
+    for i in range(num_cls): #all_classes
+        full_idx_class = list(torch.where(torch.Tensor(fullset.targets) == i)[0].cpu().numpy())
+        if(i in selected_classes):
+            test_idx_class = list(torch.where(torch.Tensor(testset.targets) == i)[0].cpu().numpy())
+            test_idx += test_idx_class
+            class_train_idx = list(np.random.choice(np.array(full_idx_class), size=split_cfg['per_idc_train'], replace=True))
+            train_idx += class_train_idx
+            remain_idx = list(set(full_idx_class) - set(class_train_idx))
+            class_val_idx = list(np.random.choice(np.array(remain_idx), size=split_cfg['per_idc_val'], replace=True))
+            remain_idx = list(set(remain_idx) - set(class_val_idx))
+            class_lake_idx = list(np.random.choice(np.array(remain_idx), size=split_cfg['per_idc_lake'], replace=True))
+        else:
+            class_train_idx = list(np.random.choice(np.array(full_idx_class), size=split_cfg['per_ood_train'], replace=False)) #always 0
+            remain_idx = list(set(full_idx_class) - set(class_train_idx))
+            class_val_idx = list(np.random.choice(np.array(remain_idx), size=split_cfg['per_ood_val'], replace=False)) #Only for CG ood val has samples
+            remain_idx = list(set(remain_idx) - set(class_val_idx))
+            class_lake_idx = list(np.random.choice(np.array(remain_idx), size=split_cfg['per_ood_lake'], replace=False)) #many ood samples in lake
+
+        if(augVal and (i in selected_classes)): #augment with samples only from the imbalanced classes
+            train_idx += class_val_idx
+        val_idx += class_val_idx
+        lake_idx += class_lake_idx
+
+    train_set = SubsetWithTargets(fullset, train_idx, torch.Tensor(fullset.targets)[train_idx])
+    val_set = SubsetWithTargets(fullset, val_idx, torch.Tensor(fullset.targets)[val_idx])
+    lake_set = SubsetWithTargets(fullset, lake_idx, getOODtargets(torch.Tensor(fullset.targets)[lake_idx], selected_classes, split_cfg['num_cls_idc']))
+    test_set = SubsetWithTargets(testset, test_idx, torch.Tensor(testset.targets)[test_idx])
+
+    return train_set, val_set, test_set, lake_set, selected_classes
+
+############
+# OOD TYPE 1
+############
+
+def load_dataset_custom_1(datadir, feature, split_cfg, augVal=False, dataAug=True):
+
+    num_cls = 8
+    path = '/mnt/data2/akshit/'
+    download_path = '/mnt/data2/akshit/data/cifar10'    
+    train_data = np.load(f'{path}data/dermamnist/dm_train.npz', allow_pickle=True)
+    val_data = np.load(f'{path}data/dermamnist/dm_val.npz', allow_pickle=True)
+    test_data = np.load(f'{path}data/dermamnist/dm_test_balanced.npz', allow_pickle=True)
+    ptrain={
+    'images': np.concatenate((train_data['images'],val_data['images'])),
+    'labels': np.concatenate((train_data['labels'],val_data['labels']))
+    }
+
+    # Define the number of classes in our modified CIFAR10, which is 6. We also define our ID classes
+    cifar_training_transform = transforms.Compose([transforms.RandomCrop(28), transforms.ToTensor(), transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))])
+    cifar_test_transform = transforms.Compose([transforms.Resize(28), transforms.ToTensor(), transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))])
+    cifar_label_transform = lambda x: 7
+
+    # Get the dataset objects from PyTorch. Here, CIFAR10 is downloaded, and the transform is applied when points 
+    # are retrieved.
+    cifar10_full_train = cifar.CIFAR10(download_path, train=True, download=False, transform=cifar_training_transform, target_transform=cifar_label_transform)
+    cifar10_test = cifar.CIFAR10(download_path, train=False, download=False, transform=cifar_test_transform, target_transform=cifar_label_transform)
+
+
+    derma_full_train = DermaDataset(data=ptrain, transform=cifar_training_transform)
+    derma_test = DermaDataset(data=test_data, transform=cifar_test_transform)
+
+    fullset = ConcatDataset([derma_full_train, cifar10_full_train])
+    fullset.targets = np.append(derma_full_train.targets, [7 for i in range(50000)])
+    test_set = derma_test
+
+    if(feature=="ood"):
+        train_set, val_set, test_set, lake_set, ood_cls_idx = create_ood_data(fullset, test_set, split_cfg, num_cls, augVal)
+        print("Custom dataset stats: Train size: ", len(train_set), "Val size: ", len(val_set), "Lake size: ", len(lake_set), "Test set: ", len(test_set))
+        return train_set, val_set, test_set, lake_set, ood_cls_idx, split_cfg['num_cls_idc']
+
+#############
+## Taking examples from different classes instead of oversampling
+#############    
+def create_ood_data_2(fullset, testset, split_cfg, num_cls, augVal):
+
+    np.random.seed(42)
+    train_idx = []
+    val_idx = []
+    lake_idx = []
+    test_idx = [] 
+    selected_classes = np.array(list(range(split_cfg['num_cls_idc'])))
+    for i in range(num_cls): #all_classes
+        full_idx_class = list(torch.where(torch.Tensor(fullset.targets) == i)[0].cpu().numpy())
+        if(i in selected_classes):
+            test_idx_class = list(torch.where(torch.Tensor(testset.targets) == i)[0].cpu().numpy())
+            test_idx += test_idx_class
+            class_train_idx = list(np.random.choice(np.array(full_idx_class), size=split_cfg['per_idc_train'], replace=False))
+            train_idx += class_train_idx
+            remain_idx = list(set(full_idx_class) - set(class_train_idx))
+
+            class_val_idx = list(np.random.choice(np.array(remain_idx), size=split_cfg['per_idc_val'], replace=False))
+            remain_idx = list(set(remain_idx) - set(class_val_idx))
+            ## Taking examples from different classes instead of oversampling
+            if len(remain_idx)>=split_cfg['per_idc_lake']:
+                class_lake_idx = list(np.random.choice(np.array(remain_idx), size=split_cfg['per_idc_lake'], replace=False))
+            elif len(remain_idx)<split_cfg['per_idc_lake']:
+                class_lake_idx = list(np.random.choice(np.array(remain_idx), size=len(remain_idx), replace=False)) 
+
+        else:
+            class_train_idx = list(np.random.choice(np.array(full_idx_class), size=split_cfg['per_ood_train'], replace=False)) #always 0
+            remain_idx = list(set(full_idx_class) - set(class_train_idx))
+            class_val_idx = list(np.random.choice(np.array(remain_idx), size=split_cfg['per_ood_val'], replace=False)) #Only for CG ood val has samples
+            remain_idx = list(set(remain_idx) - set(class_val_idx))
+            class_lake_idx = list(np.random.choice(np.array(remain_idx), size=split_cfg['per_ood_lake'], replace=False)) #many ood samples in lake
+
+        if(augVal and (i in selected_classes)): #augment with samples only from the imbalanced classes
+            train_idx += class_val_idx
+        val_idx += class_val_idx
+        lake_idx += class_lake_idx
+
+    train_set = SubsetWithTargets(fullset, train_idx, torch.Tensor(fullset.targets)[train_idx])
+    val_set = SubsetWithTargets(fullset, val_idx, torch.Tensor(fullset.targets)[val_idx])
+    lake_set = SubsetWithTargets(fullset, lake_idx, getOODtargets(torch.Tensor(fullset.targets)[lake_idx], selected_classes, split_cfg['num_cls_idc']))
+    test_set = SubsetWithTargets(testset, test_idx, torch.Tensor(testset.targets)[test_idx])
+
+    return train_set, val_set, test_set, lake_set, selected_classes    
+
+def load_dataset_custom_2(datadir, feature, split_cfg, augVal=False, dataAug=True):
+
+    num_cls = 8
+    path = '/mnt/data2/akshit/'
+    download_path = '/mnt/data2/akshit/data/cifar10'    
+    train_data = np.load(f'{path}data/dermamnist/dm_train.npz', allow_pickle=True)
+    val_data = np.load(f'{path}data/dermamnist/dm_val.npz', allow_pickle=True)
+    test_data = np.load(f'{path}data/dermamnist/dm_test_balanced.npz', allow_pickle=True)
+    ptrain={
+    'images': np.concatenate((train_data['images'],val_data['images'])),
+    'labels': np.concatenate((train_data['labels'],val_data['labels']))
+    }
+
+    # Define the number of classes in our modified CIFAR10, which is 6. We also define our ID classes
+    cifar_training_transform = transforms.Compose([transforms.RandomCrop(28), transforms.ToTensor(), transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))])
+    cifar_test_transform = transforms.Compose([transforms.Resize(28), transforms.ToTensor(), transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))])
+    cifar_label_transform = lambda x: 7
+
+    # Get the dataset objects from PyTorch. Here, CIFAR10 is downloaded, and the transform is applied when points 
+    # are retrieved.
+    cifar10_full_train = cifar.CIFAR10(download_path, train=True, download=False, transform=cifar_training_transform, target_transform=cifar_label_transform)
+    cifar10_test = cifar.CIFAR10(download_path, train=False, download=False, transform=cifar_test_transform, target_transform=cifar_label_transform)
+
+
+    derma_full_train = DermaDataset(data=ptrain, transform=cifar_training_transform)
+    derma_test = DermaDataset(data=test_data, transform=cifar_test_transform)
+
+    fullset = ConcatDataset([derma_full_train, cifar10_full_train])
+    fullset.targets = np.append(derma_full_train.targets, [7 for i in range(50000)])
+    test_set = derma_test
+
+    if(feature=="ood"):
+        train_set, val_set, test_set, lake_set, ood_cls_idx = create_ood_data_2(fullset, test_set, split_cfg, num_cls, augVal)
+        print("Custom dataset stats: Train size: ", len(train_set), "Val size: ", len(val_set), "Lake size: ", len(lake_set), "Test set: ", len(test_set))
+        return train_set, val_set, test_set, lake_set, ood_cls_idx, split_cfg['num_cls_idc']

trust/utils/.ipynb_checkpoints/medmnist-checkpoint.py

@@ -0,0 +1,146 @@
+import os
+import numpy as np
+from PIL import Image
+from torch.utils.data import Dataset
+from .medmnist_info import INFO
+
+
+class MedMNIST(Dataset):
+
+    flag = ...
+
+    def __init__(self,
+                 root,
+                 split='train',
+                 transform=None,
+                 target_transform=None,
+                 download=False):
+        ''' dataset
+        :param split: 'train', 'val' or 'test', select subset
+        :param transform: data transformation
+        :param target_transform: target transformation
+
+        '''
+
+        self.info = INFO[self.flag]
+        self.root = root
+
+        if download:
+            self.download()
+
+        if not os.path.exists(
+                os.path.join(self.root, "{}.npz".format(self.flag))):
+            raise RuntimeError('Dataset not found.' +
+                               ' You can use download=True to download it')
+
+        npz_file = np.load(os.path.join(self.root, "{}.npz".format(self.flag)))
+
+        self.split = split
+        self.transform = transform
+        self.target_transform = target_transform
+
+        if self.split == 'train':
+            self.data = npz_file['train_images']
+            self.targets = npz_file['train_labels']
+            self.targets = np.squeeze(self.targets)
+        elif self.split == 'val':
+            self.data = npz_file['val_images']
+            self.targets = npz_file['val_labels']
+            self.targets = np.squeeze(self.targets)
+        elif self.split == 'test':
+            self.data = npz_file['test_images']
+            self.targets = npz_file['test_labels']
+            self.targets = np.squeeze(self.targets)
+
+        if self.flag == 'octmnist' or self.flag == 'pneumoniamnist':
+            new_data = []
+            for i in self.data:
+                i = np.stack((i,)*3,axis=-1)
+                new_data.append(i)
+            self.data = np.array(new_data)
+
+    def __getitem__(self, index):
+        data, target = self.data[index], self.targets[index].astype(int)
+        data = Image.fromarray(np.uint8(data))
+
+        if self.transform is not None:
+            data = self.transform(data)
+
+        if self.target_transform is not None:
+            target = self.target_transform(target)
+
+        return data, target
+
+    def __len__(self):
+        return self.data.shape[0]
+
+    def __repr__(self):
+        '''Adapted from torchvision.
+        '''
+        _repr_indent = 4
+        head = "Dataset " + self.__class__.__name__
+
+        body = ["Number of datapoints: {}".format(self.__len__())]
+        body.append("Root location: {}".format(self.root))
+        body.append("Split: {}".format(self.split))
+        body.append("Task: {}".format(self.info["task"]))
+        body.append("Number of channels: {}".format(self.info["n_channels"]))
+        body.append("Meaning of labels: {}".format(self.info["label"]))
+        body.append("Number of samples: {}".format(self.info["n_samples"]))
+        body.append("Description: {}".format(self.info["description"]))
+        body.append("License: {}".format(self.info["license"]))
+
+        lines = [head] + [" " * _repr_indent + line for line in body]
+        return '\n'.join(lines)
+
+    def download(self):
+        try:
+            from torchvision.datasets.utils import download_url
+            download_url(url=self.info["url"],
+                         root=self.root,
+                         filename="{}.npz".format(self.flag),
+                         md5=self.info["MD5"])
+        except:
+            raise RuntimeError('Something went wrong when downloading! ' +
+                               'Go to the homepage to download manually. ' +
+                               'https://github.com/MedMNIST/MedMNIST')
+
+
+class PathMNIST(MedMNIST):
+    flag = "pathmnist"
+
+
+class OCTMNIST(MedMNIST):
+    flag = "octmnist"
+
+
+class PneumoniaMNIST(MedMNIST):
+    flag = "pneumoniamnist"
+
+
+class ChestMNIST(MedMNIST):
+    flag = "chestmnist"
+
+
+class DermaMNIST(MedMNIST):
+    flag = "dermamnist"
+
+
+class RetinaMNIST(MedMNIST):
+    flag = "retinamnist"
+
+
+class BreastMNIST(MedMNIST):
+    flag = "breastmnist"
+
+
+class OrganMNISTAxial(MedMNIST):
+    flag = "organmnist_axial"
+
+
+class OrganMNISTCoronal(MedMNIST):
+    flag = "organmnist_coronal"
+
+
+class OrganMNISTSagittal(MedMNIST):
+    flag = "organmnist_sagittal"