work on trainin and dashboard statistics

.gitattributes

@@ -26,3 +26,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zstandard filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 *ubyte* filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text

app.py

@@ -2,51 +2,163 @@ import os
 import torch
 import gradio as gr
 import torchvision
+from PIL import Image
 from utils import *
 import torch.nn as nn
 import torch.nn.functional as F
 import torch.optim as optim
 from huggingface_hub import Repository, upload_file
+from torch.utils.data import Dataset
+import numpy as np
+from collections import Counter 
 
 
 
-n_epochs = 3
-batch_size_train = 64
+
+n_epochs = 10
+batch_size_train = 128
 batch_size_test = 1000
 learning_rate = 0.01
 momentum = 0.5
 log_interval = 10
 random_seed = 1
-
+TRAIN_CUTOFF = 5
+WHAT_TO_DO=WHAT_TO_DO.format(num_samples=TRAIN_CUTOFF)
+METRIC_PATH = './metrics.json'
 REPOSITORY_DIR = "data"
 LOCAL_DIR = 'data_local'
 os.makedirs(LOCAL_DIR,exist_ok=True)
 
 
+
+
 HF_TOKEN = os.getenv("HF_TOKEN")
 
 HF_DATASET ="mnist-adversarial-dataset"
+DATASET_REPO_URL = f"https://huggingface.co/datasets/chrisjay/{HF_DATASET}"
+repo = Repository(
+    local_dir="data_mnist", clone_from=DATASET_REPO_URL, use_auth_token=HF_TOKEN
+)
+repo.git_pull()
 
 torch.backends.cudnn.enabled = False
 torch.manual_seed(random_seed)
 
-train_loader = torch.utils.data.DataLoader(
-  torchvision.datasets.MNIST('files/', train=True, download=True,
-                             transform=torchvision.transforms.Compose([
+
+class MNISTAdversarial_Dataset(Dataset):
+
+    def __init__(self,data_dir,transform):
+        repo.git_pull()
+        self.data_dir = os.path.join(data_dir,'data')
+        self.transform = transform
+        files = [f.name for f in os.scandir(self.data_dir)]
+        self.images = []
+        self.numbers = []
+        for f in files:
+            self.FOLDER = os.path.join(os.path.join(self.data_dir,f))
+            
+            metadata_path = os.path.join(self.FOLDER,'metadata.jsonl')
+
+            image_path =os.path.join(self.FOLDER,'image.png')
+            if os.path.exists(image_path) and os.path.exists(metadata_path): 
+                img = Image.open(image_path)
+                self.images.append(img)
+                metadata = read_json_lines(metadata_path)
+                self.numbers.append(metadata[0]['correct_number'])
+        assert len(self.images)==len(self.numbers), f"Length of images and numbers must be the same. Got {len(self.images)} for images and {len(self.numbers)} for numbers."
+    def __len__(self):
+        return len(self.images)
+    
+    def __getitem__(self,idx):
+        img, label = self.images[idx], self.numbers[idx]
+        img = self.transform(img)
+        return img, label
+
+class MNISTCorrupted_By_Digit(Dataset):
+    def __init__(self,transform,digit,limit=30):
+        self.transform = transform
+        self.digit = digit
+        corrupted_dir="./mnist_c"
+        files = [f.name for f in os.scandir(corrupted_dir)]
+        images = [np.load(os.path.join(os.path.join(corrupted_dir,f),'test_images.npy')) for f in files]
+        labels = [np.load(os.path.join(os.path.join(corrupted_dir,f),'test_labels.npy')) for f in files]
+        self.data = np.vstack(images)
+        self.labels = np.hstack(labels)
+        
+        assert (self.data.shape[0] == self.labels.shape[0])
+        
+        mask = self.labels == self.digit
+
+        data_masked  = self.data[mask] 
+        # Just to be on the safe side, ensure limit is more than the minimum
+        limit = min(limit,data_masked.shape[0])
+
+        self.data_for_use = data_masked[:limit]
+        self.labels_for_use = self.labels[mask][:limit]
+        assert (self.data_for_use.shape[0] == self.labels_for_use.shape[0])
+
+    def __len__(self):
+        return len(self.data_for_use)
+    def __getitem__(self,idx):
+        if torch.is_tensor(idx):
+            idx = idx.tolist()
+
+        image = self.data_for_use[idx]
+        label = self.labels_for_use[idx]
+        if self.transform:
+            image_pil = torchvision.transforms.ToPILImage()(image) # Need to transform to PIL before using default transforms
+            image = self.transform(image_pil) 
+
+        return image, label
+
+
+
+
+
+class MNISTCorrupted(Dataset):
+    def __init__(self,transform):
+        self.transform = transform
+        corrupted_dir="./mnist_c"
+        files = [f.name for f in os.scandir(corrupted_dir)]
+        images = [np.load(os.path.join(os.path.join(corrupted_dir,f),'test_images.npy')) for f in files]
+        labels = [np.load(os.path.join(os.path.join(corrupted_dir,f),'test_labels.npy')) for f in files]
+        self.data = np.vstack(images)
+        self.labels = np.hstack(labels)
+
+        assert (self.data.shape[0] == self.labels.shape[0])
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, idx):
+        if torch.is_tensor(idx):
+            idx = idx.tolist()
+
+        image = self.data[idx]
+        label = self.labels[idx]
+        if self.transform:
+            image_pil = torchvision.transforms.ToPILImage()(image) # Need to transform to PIL before using default transforms
+            image = self.transform(image_pil) 
+
+        return image, label
+
+
+
+TRAIN_TRANSFORM = torchvision.transforms.Compose([
                                torchvision.transforms.ToTensor(),
                                torchvision.transforms.Normalize(
                                  (0.1307,), (0.3081,))
-                             ])),
+                             ])
+
+'''
+train_loader = torch.utils.data.DataLoader(
+  torchvision.datasets.MNIST('files/', train=True, download=True,
+                             transform=TRAIN_TRANSFORM),
   batch_size=batch_size_train, shuffle=True)
+'''
 
-test_loader = torch.utils.data.DataLoader(
-  torchvision.datasets.MNIST('files/', train=False, download=True,
-                             transform=torchvision.transforms.Compose([
-                               torchvision.transforms.ToTensor(),
-                               torchvision.transforms.Normalize(
-                                 (0.1307,), (0.3081,))
-                             ])),
-  batch_size=batch_size_test, shuffle=True)
+test_loader = torch.utils.data.DataLoader(MNISTCorrupted(TRAIN_TRANSFORM),
+                                            batch_size=batch_size_test, shuffle=False)
 
 
 # Source: https://nextjournal.com/gkoehler/pytorch-mnist
@@ -69,7 +181,7 @@ class MNIST_Model(nn.Module):
         return F.log_softmax(x)
 
 
-def train(epochs,network,optimizer):
+def train(epochs,network,optimizer,train_loader):
     
     train_losses=[]
     network.train()
@@ -102,10 +214,11 @@ def test():
             correct += pred.eq(target.data.view_as(pred)).sum()
             test_loss /= len(test_loader.dataset)
         test_losses.append(test_loss)
+        acc = 100. * correct / len(test_loader.dataset)
+        acc = acc.item()
         test_metric = '〽Current test metric - Avg. loss: `{:.4f}`, Accuracy: `{}/{}` (`{:.0f}%`)\n'.format(
-        test_loss, correct, len(test_loader.dataset),
-        100. * correct / len(test_loader.dataset))
-        return test_metric
+        test_loss, correct, len(test_loader.dataset),acc  )
+        return test_metric,acc
 
 
 
@@ -156,15 +269,41 @@ def image_classifier(inp):
 
 def train_and_test():
     # Train for one epoch and test
-    train(1,network,optimizer)
-    test_metric = test()
-
-def flag(input_image,correct_result,train):
-    # take an image, the wrong result, the correct result.
-    # push to dataset. 
-    # get size of current dataset 
-   
-    # Write audio to file
+    train_dataset = MNISTAdversarial_Dataset('./data_mnist',TRAIN_TRANSFORM)
+
+    train_loader = torch.utils.data.DataLoader(train_dataset,batch_size=batch_size_test, shuffle=True
+                                             )
+    train(n_epochs,network,optimizer,train_loader)
+    test_metric,test_acc = test()
+
+    if os.path.exists(METRIC_PATH):
+        metric_dict = read_json(METRIC_PATH)
+        metric_dict['all'] = metric_dict['all'] if 'all' in metric_dict else [] + [test_acc]
+    else:
+        metric_dict={}
+        metric_dict['all'] = [test_acc] 
+
+    for i in range(10):
+        data_per_digit = MNISTCorrupted_By_Digit(TRAIN_TRANSFORM,i)
+        dataloader_per_digit = torch.utils.data.DataLoader(data_per_digit,batch_size=len(data_per_digit), shuffle=False)
+        data_per_digit, label_per_digit = iter(dataloader_per_digit).next()
+        output = network(data_per_digit)
+        pred = output.data.max(1, keepdim=True)[1]
+        correct = pred.eq(label_per_digit.data.view_as(pred)).sum()
+        acc = 100. * correct / len(data_per_digit)
+        acc=acc.item()
+        if os.path.exists(METRIC_PATH):
+            metric_dict[str(i)].append(acc) 
+        else:
+            metric_dict[str(i)] = [acc] 
+
+    dump_json(thing=metric_dict,file=METRIC_PATH)    
+    return test_metric
+
+def flag(input_image,correct_result,adversarial_number):
+
+    adversarial_number = 0 if None else adversarial_number
+
     metadata_name = get_unique_name()
     SAVE_FILE_DIR = os.path.join(LOCAL_DIR,metadata_name)
     os.makedirs(SAVE_FILE_DIR,exist_ok=True)
@@ -182,7 +321,7 @@ def flag(input_image,correct_result,train):
     
     dump_json(metadata,json_file_path)  
         
-    # Simply upload the audio file and metadata using the hub's upload_file
+    # Simply upload the image file and metadata using the hub's upload_file
     # Upload the image
     repo_image_path = os.path.join(REPOSITORY_DIR,os.path.join(metadata_name,'image.png'))
     
@@ -201,43 +340,84 @@ def flag(input_image,correct_result,train):
                 repo_type='dataset',
                 token=HF_TOKEN
             )        
+    adversarial_number+=1
+    output = f'<div> ✔ ({adversarial_number}) Successfully saved your adversarial data. </div>'
+    repo.git_pull()
+    length_of_dataset = len([f for f in os.scandir("./data_mnist/data")])
+    test_metric = f"<html> {DEFAULT_TEST_METRIC} </html>"
+    if length_of_dataset % TRAIN_CUTOFF ==0:
+        test_metric_ = train_and_test()
+        test_metric = f"<html> {test_metric_} </html>"
+        output = f'<div> ✔ ({adversarial_number}) Successfully saved your adversarial data and trained the model on adversarial data! </div>'
+    return output,test_metric,adversarial_number
+
+def get_number_dict(DATA_DIR):
+    files = [f.name for f in os.scandir(DATA_DIR)]
+    numbers = [read_json_lines(os.path.join(os.path.join(DATA_DIR,f),'metadata.jsonl'))[0]['correct_number'] for f in files]
+    numbers_count = Counter(numbers)
+    numbers_count_keys = list(numbers_count.keys())
+    numbers_count_values = [numbers_count[k] for k in numbers_count_keys]
+    return numbers_count_keys,numbers_count_values
+
+
+
+def get_statistics():
+    model_state_dict = 'model.pth'
+    optimizer_state_dict = 'optmizer.pth'
+
+    if os.path.exists(model_state_dict):
+        network_state_dict = torch.load(model_state_dict)
+        network.load_state_dict(network_state_dict)
+
+    if os.path.exists(optimizer_state_dict):
+        optimizer_state_dict = torch.load(optimizer_state_dict)
+        optimizer.load_state_dict(optimizer_state_dict)                      
+    repo.git_pull()
+    DATA_DIR = './data_mnist/data'
+    numbers_count_keys,numbers_count_values = get_number_dict(DATA_DIR)
+
+    
+    plt_digits = plot_bar(numbers_count_values,numbers_count_keys,'Number of adversarial samples',"Digit",f"Distribution of adversarial samples over digits")
     
-    output = f'<div> ✔ Successfully saved to flagged dataset. </div>'
-    train=True
-    if train:
-        output = f'<div> ✔ Successfully saved to flagged dataset. Training the model on adversarial data! </div>'
+    fig_d, ax_d = plt.subplots(figsize=(10,4),tight_layout=True)
 
-    return output,train
+    if os.path.exists(METRIC_PATH):
+        metric_dict = read_json(METRIC_PATH)
+        for i in range(10):
+            try:    
+                x_i = [i+1 for i in range(len(metric_dict[str(i)]))]
+                ax_d.plot(x_i, metric_dict[str(i)],label=str(i))
+            except Exception:
+                continue
+        dump_json(thing=metric_dict,file=METRIC_PATH)    
+    else:
+        metric_dict={}
 
+    fig_d.legend()
+    ax_d.set(xlabel='Adversarial train steps', ylabel='MNIST_C Test Accuracy',title="Test Accuracy over digits per train step")
 
+    done_html =  """<div style="color: green">
+                <p> ✅ Statistics loaded successfully!</p>
+                </div>
+               """ 
 
-def main():
-    TITLE = "# MNIST Adversarial: Try to fool this MNIST model"
-    description = """This project is about dynamic adversarial data collection (DADC). 
-    The basic idea is to collect “adversarial data” - the kind of data that is difficult for a model to predict correctly. 
-    This kind of data is presumably the most valuable for a model, so this can be helpful in low-resource settings where data is hard to collect and label.
-
-    ### What to do:
-    - Draw a number from 0-9.
-    - Click `Submit` and see the model's prediciton.
-    - If the model misclassifies it, Flag that example.
-    - This will add your (adversarial) example to a dataset on which the model will be trained later.
-    """
+    return plt_digits,fig_d,done_html    
 
-    MODEL_IS_WRONG = """
-    ---
 
-    > Did the model get it wrong? Choose the correct prediction below and flag it. When you flag it, the instance is saved to our dataset and the model is trained on it.
-    """
+
+
+def main():
     #block = gr.Blocks(css=BLOCK_CSS)
     block = gr.Blocks()
 
     with block:
         gr.Markdown(TITLE)
+        gr.Markdown(description)
 
         with gr.Tabs():
-            gr.Markdown(description)
             with gr.TabItem('MNIST'):
+                gr.Markdown(WHAT_TO_DO)
+                test_metric = gr.outputs.HTML(DEFAULT_TEST_METRIC)
                 with gr.Row():     
             
 
@@ -249,15 +429,32 @@ def main():
                 number_dropdown = gr.Dropdown(choices=[i for i in range(10)],type='value',default=None,label="What was the correct prediction?") 
 
                 flag_btn = gr.Button("Flag")
+
                 output_result = gr.outputs.HTML()
-                to_train = gr.Variable(value=False)
+                adversarial_number = gr.Variable(value=0)
+
+
                 submit.click(image_classifier,inputs = [image_input],outputs=[label_output])
-                flag_btn.click(flag,inputs=[image_input,number_dropdown,to_train],outputs=[output_result,to_train])
-                if to_train.value:
-                    import pdb;pdb.set_trace()
-                    train_and_test()        
+                flag_btn.click(flag,inputs=[image_input,number_dropdown,adversarial_number],outputs=[output_result,test_metric,adversarial_number])
                 
-            
+            with gr.TabItem('Dashboard') as dashboard:
+                notification = gr.HTML("""<div style="color: green">
+                                        <p> ⌛ Creating statistics... </p>
+                                        </div>
+                                    """)
+                _,numbers_count_values_ = get_number_dict('./data_mnist/data')
+
+                STATS_EXPLANATION_ = STATS_EXPLANATION.format(num_adv_samples = sum(numbers_count_values_))
+
+                gr.Markdown(STATS_EXPLANATION_)
+                stat_adv_image =gr.Plot(type="matplotlib")
+                gr.Markdown(DASHBOARD_EXPLANATION)
+                test_results=gr.Plot(type="matplotlib")
+
+            dashboard.select(get_statistics,inputs=[],outputs=[stat_adv_image,test_results,notification])
+
+
+
     block.launch()  
         
      

data_mnist

@@ -0,0 +1 @@
+Subproject commit eb1e3cf9de597112c1da3b921ffcd07c8e4419c1

mnist_c/brightness/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0be26b927b2be43bc0b2430ccf6aa6048a5fcd8bcd087d97576972f009c1e8a9
+size 7840128

mnist_c/brightness/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/brightness/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:466c8f7365f37d14012ec66e2d1203ef07acae08240bb1af1863f09318293255
+size 47040128

mnist_c/brightness/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/canny_edges/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e86b158456cbaf0bf621092259f186a223b68e31128c18cceedb8a2f1b8baf1f
+size 7840128

mnist_c/canny_edges/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/canny_edges/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d6454058f0025bd15f34f9d14df4fdf2cca8f47b38661da5b650bee9c77aac5
+size 47040128

mnist_c/canny_edges/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/dotted_line/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02f5a8bd11e1bed6ca5f31418c6804a4f89ab5abdaba38f880e4743294516c6b
+size 7840128

mnist_c/dotted_line/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/dotted_line/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2d1ad4696a96af19f6bcd07a2ea9f2fb0990910397ef41b969d26080aef23cb
+size 47040128

mnist_c/dotted_line/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/fog/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f48ea89b4793c45bedda39bd27b42710f4f0cbf5bced559597c015157188488
+size 7840128

mnist_c/fog/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/fog/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c92a6df9c3703b3df29fdb272d829736e706150104a7f7c81f37387940120e9
+size 47040128

mnist_c/fog/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/glass_blur/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:300a4829b7fe6e4cfa5ad0e134cf19136e2c23e1e6bc29b6c574848b6a493388
+size 7840128

mnist_c/glass_blur/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/glass_blur/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:255b6832620840d2db4423212d02c903d4edf5b31b94e16ac130f2539c73275f
+size 47040128

mnist_c/glass_blur/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/identity/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c544e9053023b30ffd401a9825696f84af6fb1eb822bba5fcffc6992808357c3
+size 7840128

mnist_c/identity/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/identity/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f323590fab80abce68b33a35fb0a176bed275ab73b934621f7583bd1c7c1b1ba
+size 47040128

mnist_c/identity/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/impulse_noise/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63ad2d4c0860dca3814da6034943dda6e9fe267ba6e12136341c52f2c13ca3d5
+size 7840128

mnist_c/impulse_noise/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/impulse_noise/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5921469548b8d051544f29b984112d0680054bba619b7a69a1479cbd082b9563
+size 47040128

mnist_c/impulse_noise/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/motion_blur/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7767b48be95a63abcdf851a4baedc0491d5643a192072e4f6de0a14c75cb089e
+size 7840128

mnist_c/motion_blur/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/motion_blur/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f11f24aa857845517b700892d5dcbeeb8c2ee3a7b0028e27a6231c9aaa4f32db
+size 47040128

mnist_c/motion_blur/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/rotate/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8fb565f28d858733ec94d4e6d976d9f2cb3e6aa75deb9a7a11b0504b9ae64520
+size 7840128

mnist_c/rotate/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/rotate/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:505bcc0035d1f36fe6560e39cbb3888ed0130d429bca92019f29b0cfe5ab8724
+size 47040128

mnist_c/rotate/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/scale/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:282dfee12607c13ea76cc82985e3e5d78f36b44d56d6ad9f12a2298e20ea2200
+size 7840128

mnist_c/scale/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/scale/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f558258555337a45d82b572a6a2ed48c3117985213937ded3fc6a738832b4f1b
+size 47040128

mnist_c/scale/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/shear/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38412620554c8d775e38d6b9d771b4825605998179e88bd004d8d5d0a7afaa7d
+size 7840128

mnist_c/shear/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/shear/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:133295f31cf838b4905b8e05083570e8189e37839168f37e3e83831244e54579
+size 47040128

mnist_c/shear/train_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10226b938104d9231ead4e9a0627f17e66cd18baf31a28784f1b72147057decf
+size 480128

mnist_c/shot_noise/test_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2c7026fb754f234a1563711576266f179f47b6f5e4a8e43dfbbb182e1299d764
+size 7840128

mnist_c/shot_noise/test_labels.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:996074233ccae69ee65b8551233f608adc1bce84b6b440e3531478a847958149
+size 80128

mnist_c/shot_noise/train_images.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67dc136c87501002f971be5dabd14c8e56615af61c699a7707faf4185dfbfdf6
+size 47040128