206 lines
4.1 KiB
Markdown
206 lines
4.1 KiB
Markdown
好的 😄，我给你整理一个 **PyTorch `tensor` 使用大全**，覆盖创建、操作、转换、计算等常用方法。为了系统化，我会按类别来整理，并给简单示例。
|
||
|
||
---
|
||
|
||
# 🟢 1. 创建 Tensor
|
||
|
||
```python
|
||
import torch
|
||
|
||
# 直接从数据创建
|
||
a = torch.tensor([1, 2, 3], dtype=torch.float32)
|
||
|
||
# 全 0 / 全 1 / 全某数
|
||
b = torch.zeros(3, 4)
|
||
c = torch.ones(2, 5) * 7
|
||
|
||
# 随机数
|
||
d = torch.rand(3, 3) # 0~1 均匀分布
|
||
e = torch.randn(3, 3) # 标准正态分布
|
||
|
||
# 类似 numpy 的 arange / linspace
|
||
f = torch.arange(0, 10, 2) # 0,2,4,6,8
|
||
g = torch.linspace(0, 1, steps=5) # 0,0.25,0.5,0.75,1
|
||
|
||
# 空 tensor（占位）
|
||
h = torch.empty(2, 3)
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 2. Tensor 属性
|
||
|
||
```python
|
||
x = torch.randn(2, 3)
|
||
|
||
x.shape # 返回 (2,3)
|
||
x.size() # 等同 shape
|
||
x.dtype # 数据类型，如 torch.float32
|
||
x.device # 运行设备，如 cpu 或 cuda:0
|
||
x.numel() # 元素总数
|
||
x.requires_grad_(True) # 设置是否需要梯度
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 3. Tensor 运算
|
||
|
||
### 3.1 基本算术
|
||
|
||
```python
|
||
a = torch.tensor([1,2,3])
|
||
b = torch.tensor([4,5,6])
|
||
|
||
c = a + b
|
||
d = a - b
|
||
e = a * b
|
||
f = a / b
|
||
g = a ** 2
|
||
```
|
||
|
||
### 3.2 矩阵运算
|
||
|
||
```python
|
||
A = torch.rand(2,3)
|
||
B = torch.rand(3,2)
|
||
|
||
C = torch.matmul(A, B) # 矩阵乘法
|
||
D = A @ B # 等价写法
|
||
E = A.T # 转置
|
||
```
|
||
|
||
### 3.3 统计函数
|
||
|
||
```python
|
||
x = torch.tensor([[1,2],[3,4]], dtype=torch.float32)
|
||
|
||
x.sum() # 所有元素求和
|
||
x.mean() # 平均值
|
||
x.max() # 最大值
|
||
x.min() # 最小值
|
||
x.argmax() # 最大值索引
|
||
x.argmin() # 最小值索引
|
||
x.std() # 标准差
|
||
x.var() # 方差
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 4. Tensor 索引与切片
|
||
|
||
```python
|
||
x = torch.tensor([[1,2,3],[4,5,6]])
|
||
|
||
x[0] # 第一行
|
||
x[:,1] # 第二列
|
||
x[0,2] # 第一行第三列
|
||
x[1,:2] # 第二行前两列
|
||
x[-1,-1] # 最后一个元素
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 5. Tensor 改变形状
|
||
|
||
```python
|
||
x = torch.arange(12)
|
||
|
||
y = x.view(3,4) # 改变形状 (reshape)
|
||
z = x.reshape(3,4) # reshape 等同 view，但更安全
|
||
w = x.unsqueeze(0) # 增加维度
|
||
v = x.squeeze() # 删除维度为1的维
|
||
t = x.transpose(0,0) # 交换维度
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 6. Tensor 拼接与拆分
|
||
|
||
```python
|
||
a = torch.randn(2,3)
|
||
b = torch.randn(2,3)
|
||
|
||
# 拼接
|
||
c = torch.cat([a,b], dim=0) # 纵向拼接
|
||
d = torch.cat([a,b], dim=1) # 横向拼接
|
||
|
||
# 堆叠（增加新维度）
|
||
e = torch.stack([a,b], dim=0)
|
||
|
||
# 拆分
|
||
f1, f2 = torch.chunk(c, 2, dim=0) # 按行拆成两块
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 7. Tensor 类型转换
|
||
|
||
```python
|
||
x = torch.tensor([1,2,3], dtype=torch.int32)
|
||
|
||
x.float() # 转 float
|
||
x.double() # 转 double
|
||
x.long() # 转 int64
|
||
x.bool() # 转 bool
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 8. Tensor 与 NumPy 互转
|
||
|
||
```python
|
||
import numpy as np
|
||
|
||
a = torch.tensor([1,2,3])
|
||
b = a.numpy() # Tensor → numpy array
|
||
|
||
c = np.array([4,5,6])
|
||
d = torch.from_numpy(c) # numpy → Tensor
|
||
```
|
||
|
||
⚠️ 注意：共享内存，如果修改其中一个，另一个也会变。
|
||
|
||
---
|
||
|
||
# 🟢 9. GPU 操作
|
||
|
||
```python
|
||
x = torch.randn(2,3)
|
||
|
||
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
||
x = x.to(device)
|
||
|
||
y = torch.tensor([1,2,3], device=device) # 直接在 GPU 上创建
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 10. 自动求导（梯度）
|
||
|
||
```python
|
||
x = torch.tensor([2.0,3.0], requires_grad=True)
|
||
y = x ** 2 + 3*x
|
||
z = y.sum()
|
||
z.backward() # 自动求梯度
|
||
print(x.grad) # dx
|
||
```
|
||
|
||
---
|
||
|
||
# 🟢 11. 常用函数速查
|
||
|
||
| 类别 | 常用函数 |
|
||
| ---- | --------------------------------------------- |
|
||
| 线性代数 | `matmul`, `mm`, `bmm`, `einsum` |
|
||
| 激活函数 | `relu`, `sigmoid`, `tanh`, `softmax` |
|
||
| 聚合函数 | `sum`, `mean`, `max`, `min`, `prod`, `cumsum` |
|
||
| 比较函数 | `eq`, `ne`, `gt`, `lt`, `ge`, `le` |
|
||
| 随机函数 | `rand`, `randn`, `randint`, `randperm` |
|
||
| 其他 | `clone`, `detach`, `item`, `expand`, `repeat` |
|
||
|
||
---
|
||
|
||
如果你需要，我可以帮你整理一个 **可直接拷贝的 PyTorch Tensor 使用小抄（含示例和用法备注）**，适合放在屏幕边上快速查。
|
||
|
||
你希望我帮你整理吗？
|