【資料圖】

深度學(xué)習(xí)--PyTorch維度變換、自動(dòng)拓展、合并與分割

一、維度變換

1.1 view/reshape 變換

?這兩個(gè)方法用法相同，就是變換變量的shape，變換前后的數(shù)據(jù)量相等。

a=torch.rand(4,1,28,28)a.view(4,28*28)#tensor([[0.9787, 0.6729, 0.4877,  ..., 0.8975, 0.3361, 0.9341],#        [0.4316, 0.8875, 0.2974,  ..., 0.3385, 0.5543, 0.5648],#        [0.3156, 0.1170, 0.7126,  ..., 0.4445, 0.7357, 0.4900],#        [0.7127, 0.9316, 0.7615,  ..., 0.7660, 0.5437, 0.1383]])a.reshape(4,28*28)#tensor([[0.9787, 0.6729, 0.4877,  ..., 0.8975, 0.3361, 0.9341],#        [0.4316, 0.8875, 0.2974,  ..., 0.3385, 0.5543, 0.5648],#        [0.3156, 0.1170, 0.7126,  ..., 0.4445, 0.7357, 0.4900],#        [0.7127, 0.9316, 0.7615,  ..., 0.7660, 0.5437, 0.1383]])

1.2 squeeze/unsqueeze 擠壓/拉伸 維度刪除/維度添加

a.shape#torch.Size([4, 1, 28, 28])#a.unsqueeze(x) 在x位置之后插入a.unsqueeze(1).shape#torch.Size([4, 1, 1, 28, 28])#a.squeeze(x) 刪除x維度a.squeeze(1).shape#torch.Size([4, 28, 28])a.squeeze(0).shape#torch.Size([4, 1, 28, 28]) 當(dāng)前維度非0，不會(huì)進(jìn)行壓縮

1.3 expand/reapeat 擴(kuò)展/重復(fù) 對(duì)一個(gè)維度內(nèi)部的大小進(jìn)行擴(kuò)展，增加數(shù)據(jù)

#b.expand(4,32,14,14)  對(duì)維度內(nèi)進(jìn)行拓展，用-1表示不對(duì)當(dāng)前層進(jìn)行操作a=torch.rand(4,32,14,14)b=torch.rand(1,32,1,1)b.shape#torch.Size([1, 32, 1, 1])b.expand(4,32,14,14).shape#torch.Size([4, 32, 14, 14])#b.repeat(次數(shù))  中間表示當(dāng)前層拷貝的次數(shù)b.repeat(4,1,14,14).shape#torch.Size([4, 32, 14, 14])

1.4 轉(zhuǎn)置操作 t/transpose/permute

#t()  只能對(duì)2D進(jìn)行操作a=torch.randn(3,4)a.t().shape#torch.Size([4, 3])#transpose(d1,d2) 不當(dāng)?shù)牟僮骺赡軐?duì)信息進(jìn)行破壞a=torch.randn(4,3,32,32)a1=a.transpose(1,3).contiguous().view(4,3*32*32).view(4,32,32,3)a1.shape#torch.Size([4, 32, 32, 3])#permute(d,d,d,d) 對(duì)維度下標(biāo)進(jìn)行操作a.permute(3,2,1,0).shape#torch.Size([32, 32, 3, 4])

二、Broadcast自動(dòng)擴(kuò)展

步驟為：

1. 在前面插入一個(gè)維度dim
2. 對(duì)新加入的維度進(jìn)行擴(kuò)張，擴(kuò)張成對(duì)應(yīng)的size

符合擴(kuò)展規(guī)則才能用。

三、合并與分割

3.1合并

• cat

#torch.cat([變量，變量]，dim) 在當(dāng)前維度上進(jìn)行操作a=torch.rand(4,32,8)b=torch.rand(5,32,8)torch.cat([a,b],dim=0).shape#torch.Size([9, 32, 8])

• stack

#torch.stack([變量，變量]，dim) 創(chuàng)建一個(gè)新的維度a=torch.rand(4,3,32,32)b=torch.rand(4,3,32,32)torch.stack([a,b],dim=2).shape#torch.Size([4, 3, 2, 32, 32])

3.2拆分

• split:按照長(zhǎng)度拆分

#a.split([b1,b2,b3...],dim=0)  對(duì)dim進(jìn)行拆分a=torch.rand(2,4,3,32,32)a1,a2=a.split(1,dim=0)print(a1.shape)print(a2.shape)#torch.Size([1, 4, 3, 32, 32])#torch.Size([1, 4, 3, 32, 32])

• chunk：按數(shù)量進(jìn)行拆分

#a.chunk(num,dim=0)  對(duì)當(dāng)前dim的size/num進(jìn)行拆分a=torch.rand(4,4,3,32,32)a1,a2=a.chunk(2,dim=0)print(a1.shape)print(a2.shape)#torch.Size([2, 4, 3, 32, 32])#torch.Size([2, 4, 3, 32, 32])

關(guān)鍵詞：