glue_augmentation, params print with ''

ovotools/params/params.py

@@ -41,7 +41,7 @@ class AttrDict(OrderedDict):
             elif isinstance(v, list):
                 self[k] = [AttrDict(item) if isinstance(item, dict) else item for item in v]
 
-    def __str__(self):
+    def __repr__(self):
         def write_item(item, margin='\n'):
             if isinstance(item, dict):
                 s = '{'
@@ -62,7 +62,7 @@ class AttrDict(OrderedDict):
                     s += write_item(v, margin=margin + '    ')  + ","
                 s += ' ' + (']' if isinstance(item, list) else ')')
             else:
-                s = str(item)
+                s = repr(item)
             return s
         return write_item(self)
 
@@ -236,6 +236,6 @@ if __name__=='__main__':
     mm.save(fn, can_overwrite=True)
     print(m)
     print(mm)
-    assert str(m)==str(mm)
+    assert repr(m)==repr(mm)
     os.remove(fn + '.param.txt')
     os.remove(fn + '0.param.txt')

ovotools/pytorch/data/__init__.py

 from .threading_dataloader import BatchThreadingDataLoader, ThreadingDataLoader
 from .cached_dataset import CachedDataSet
+
+from .glue_augmentation import glue_augmentation
\ No newline at end of file

ovotools/pytorch/data/glue_augmentation.py

+import random
+
+def glue_augmentation(type='x', var_range=(0.1, 0.9), p=0.5, check_f = lambda i1,i2: True, **tensor_args):
+    '''
+    :param type: 'x' - glues 2 vertical slices
+    :param range: cut by x in this range
+    :param p: with probability p
+    :param check_f: function gets 2 data ingexes and returns if they can be glued
+    :param tensor_args: arguments with tensors (BxCxHxW) to be glued
+    :return: None. modifies  tensor_args inplace.
+    '''
+    assert type == 'x'
+    b = None
+    for _, t in tensor_args.items():
+        if b is None:
+            b = t.shape[0]
+        else:
+            assert b == t.shape[0]
+    assert b > 1, "Batch must be > 1 for glue_augmentation"
+    if random.random() < p:
+        for i1 in range(0, b-1, 2):
+            i2 = i1+1
+            if check_f(i1, i2):
+                x = int(random.uniform(var_range[0], var_range[1])*t.shape[3])
+                tmp = t[i1, :,:, :x].clone()
+                t[i1, :, :, :x] = t[i2, :,:, :x].clone()
+                t[i2, :, :, :x] = tmp
+
+
+if __name__=='__main__':
+    import torch
+    t = torch.stack([torch.zeros(1,2,4), torch.ones(1,2,4), torch.zeros(1,2,4), torch.ones(1,2,4)])
+    glue_augmentation(type='x', var_range=(0.5,0.5), p=1, t=t)
+    print(t)
+    assert( (t == torch.Tensor([[[[1., 1., 0., 0.],
+          [1., 1., 0., 0.]]],
+
+
+        [[[0., 0., 1., 1.],
+          [0., 0., 1., 1.]]],
+
+
+        [[[1., 1., 0., 0.],
+          [1., 1., 0., 0.]]],
+
+
+        [[[0., 0., 1., 1.],
+          [0., 0., 1., 1.]]]]) ).all() )