First, the tensor operations
1.1 the amount of stitching and segmentation
1.1.1 torch.cat()
torch.cat(tensors, dim=0, out=None)
Function: tensor spliced by dimension dim
- tensors: tensor sequence
- dim: To splicing dimension
t = torch.ones((2, 3))
t_0 = torch.cat([t, t], dim=0)
t_1 = torch.cat([t, t, t], dim=1)
print("t_0:{} shape:{}\nt_1:{} shape:{}".format(t_0, t_0.shape, t_1, t_1.shape))
1.1.2 torch.stack()
torch.stack(tensors, dim=0, out=None)
Function: Dimension newly created spliced on dim
- tensors: tensor sequence
- dim: To splicing dimension
t = torch.ones((2, 3))
t_stack = torch.stack([t, t], dim=2)
print("\nt_stack:{} shape:{}".format(t_stack, t_stack.shape))
note:
- cat () will not expand tensor dimension, and stack () will
- When the dim designated as 0, then the shift will have dimensions, such as the original (2, 3) t, t1 = torch.stack ([t, t], dim = 0), t1 dimension (2,2, 3)
1.1.3 torch.chunk()
torch.chunk(input, chunks, dim=0)
Function: tensor be cut by an average dimension dim points
Returns: tensor List
Note: If not divisible, the last one less than the other tensor tensor
- input: to be sliced tensor
- chunks: To cut the number of copies of the minutes
- dim: to be sliced dimension
a = torch.ones((2, 7)) # 7
print(a)
list_of_tensors = torch.chunk(a, dim=1, chunks=3) # 3
for idx, t in enumerate(list_of_tensors):
print("第{}个张量:{}, shape is {}".format(idx + 1, t, t.shape))
1.1.4 torch.split()
torch.split(tensor,split_size_or_sections, dim=0)
Function: The tensor dimension dim divided by cutting
Return Value: List tensor
- tensor: to be sliced tensor
- split_size_or_sections: is int, each represents a length; is list, the list element by slicing
- dim: to be sliced dimension
t = torch.ones((2, 5))
# list_of_tensors = torch.split(t, 2, dim=1)
# for idx, t in enumerate(list_of_tensors):
# print("第{}个张量:{}, shape is {}".format(idx + 1, t, t.shape))
list_of_tensors = torch.split(t, [2, 1, 2], dim=1)
for idx, t in enumerate(list_of_tensors):
print("第{}个张量:{}, shape is {}".format(idx+1, t, t.shape))
1.2 volume index
1.2.1 torch.index_select()
torch.index_select(input, dim, index, out=None)
Features: in the dimension dim, according to index index data
Returns: the index according to index data registration tensor
- input: tensor to be indexed
- dim: to index dimensions
- index: The index number to the data
1.2.2 torch.masked_select()
torch.masked_select(input, mask, out=None)
Function: indexed by mask of True
Returns: a one-dimensional tensor
- input: tensor to be indexed
- Boolean tensor and input the same shape: mask
1.3 amount conversion
1.3.1 torch.reshape()
torch.reshape(input, shape)
Function: the transformation tensor shape
Caution: When tensor are continuous in memory, the new shared tensor input data memory
- input: tensor to be transformed
- shape: the new shape tensor
1.3.2 torch.transpose()
torch.transpose(input, dim0, dim1)
Function: exchange tensor in two dimensions
- input: tensor to be transformed
- dim0: to be exchanged dimensions
- dim1: to be exchanged dimensions
1.3.3 torch.t()
torch.t(input)
Function: 2-dimensional tensor transpose matrix, the equivalent torch.transpose (input, 0, 1)
1.3.4 torch.squeeze()
torch.squeeze(input, dim=None, out=None)
Function: the dimension of a compressed length (axis)
- dim: If it is None, remove all of the length of the shaft 1; if the specified dimension, and only when the axis length is 1, may be removed;
1.3.5 torch.unsqueeze()
torch.unsqueeze(input, dim, out=None)
Function: According to dim extension dimension
- dim: Extended Dimensions