""" Day 113 - PyTorch 基础：Tensor 创建与运算 ========================================== 本文件演示 PyTorch Tensor 的各种创建方式和基本运算操作。 运行前请确保已安装 PyTorch: pip install torch """ import torch def main(): print("=" * 60) print("Day 113 - Tensor 基础操作") print("=" * 60) # ===================================================== # 1. Tensor 创建 # ===================================================== print("\n--- 1. Tensor 创建 ---\n") # 从 Python 列表创建 t1 = torch.tensor([1, 2, 3, 4, 5]) print(f"从列表创建: {t1}") print(f" 形状: {t1.shape}, 数据类型: {t1.dtype}, 维度: {t1.ndim}") # 二维 Tensor t2 = torch.tensor([[1, 2, 3], [4, 5, 6]]) print(f"\n二维 Tensor:\n{t2}") print(f" 形状: {t2.shape}, 元素总数: {t2.numel()}") # 指定数据类型 t3 = torch.tensor([1.0, 2.0, 3.0], dtype=torch.float32) t4 = torch.tensor([1, 2, 3], dtype=torch.float64) print(f"\nfloat32 类型: {t3}, dtype={t3.dtype}") print(f"float64 类型: {t4}, dtype={t4.dtype}") # 特殊创建方法 print("\n--- 特殊创建方法 ---") zeros = torch.zeros(2, 3) print(f"全零矩阵 (2x3):\n{zeros}") ones = torch.ones(3, 2) print(f"\n全一矩阵 (3x2):\n{ones}") rand_normal = torch.randn(2, 3) # 标准正态分布 print(f"\n标准正态随机数 (2x3):\n{rand_normal}") rand_uniform = torch.rand(2, 3) # [0, 1) 均匀分布 print(f"\n均匀分布随机数 (2x3):\n{rand_uniform}") arange = torch.arange(0, 10, 2) print(f"\narange(0, 10, 2): {arange}") linspace = torch.linspace(0, 1, 5) print(f"linspace(0, 1, 5): {linspace}") eye = torch.eye(3) print(f"\n单位矩阵 (3x3):\n{eye}") # 常量 Tensor full = torch.full((2, 3), fill_value=3.14) print(f"\n常量 Tensor (填充值 3.14):\n{full}") # ===================================================== # 2. Tensor 属性 # ===================================================== print("\n--- 2. Tensor 属性 ---\n") x = torch.tensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]) print(f"Tensor:\n{x}") print(f" 形状 (shape): {x.shape}") print(f" 数据类型 (dtype): {x.dtype}") print(f" 设备 (device): {x.device}") print(f" 维度数 (ndim): {x.ndim}") print(f" 元素总数 (numel): {x.numel()}") print(f" 是否需要梯度 (requires_grad): {x.requires_grad}") # ===================================================== # 3. 索引与切片 # ===================================================== print("\n--- 3. 索引与切片 ---\n") t = torch.arange(12).reshape(3, 4) print(f"原始 Tensor (3x4):\n{t}") print(f"\n t[0] = {t[0]} # 第0行") print(f" t[:, 1] = {t[:, 1]} # 第1列") print(f" t[1:3] =\n{t[1:3]} # 第1-2行") print(f" t[0, 1:3] = {t[0, 1:3]} # 第0行, 第1-2列") print(f" t[t > 5] = {t[t > 5]} # 条件索引") # ===================================================== # 4. 基本运算 # ===================================================== print("\n--- 4. 基本运算 ---\n") a = torch.tensor([[1.0, 2.0], [3.0, 4.0]]) b = torch.tensor([[5.0, 6.0], [7.0, 8.0]]) print(f"Tensor a:\n{a}") print(f"\nTensor b:\n{b}") # 加法 print(f"\n逐元素加法: a + b =\n{a + b}") print(f"函数加法: torch.add(a, b) =\n{torch.add(a, b)}") # 减法 print(f"\n逐元素减法: a - b =\n{a - b}") # 逐元素乘法（注意不是矩阵乘法） print(f"\n逐元素乘法: a * b =\n{a * b}") # 矩阵乘法 print(f"\n矩阵乘法: a @ b =\n{a @ b}") print(f"函数矩阵乘法: torch.mm(a, b) =\n{torch.mm(a, b)}") print(f"torch.matmul(a, b) =\n{torch.matmul(a, b)}") # 转置 print(f"\na 的转置:\n{a.t()}") print(f"a.transpose(0,1):\n{a.transpose(0, 1)}") # 广播机制 print("\n--- 广播机制 ---") c = torch.tensor([10.0, 20.0]) print(f"Tensor a:\n{a}") print(f"Tensor c: {c}") print(f"a + c (广播加法):\n{a + c}") # ===================================================== # 5. 聚合运算 # ===================================================== print("\n--- 5. 聚合运算 ---\n") t = torch.tensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]]) print(f"Tensor:\n{t}") print(f"\n 总和 (sum): {t.sum().item()}") print(f" 均值 (mean): {t.mean().item():.4f}") print(f" 最大值 (max): {t.max().item()}") print(f" 最小值 (min): {t.min().item()}") print(f" 标准差 (std): {t.std().item():.4f}") # 按维度聚合 print(f"\n 按行求和 (dim=1): {t.sum(dim=1)}") print(f" 按列求和 (dim=0): {t.sum(dim=0)}") print(f" 按行求均值 (dim=1): {t.mean(dim=1)}") # ===================================================== # 6. 变形操作 # ===================================================== print("\n--- 6. 变形操作 ---\n") t = torch.arange(12) print(f"原始 (1D): {t}") print(f"reshape(3, 4):\n{t.reshape(3, 4)}") print(f"reshape(4, 3):\n{t.reshape(4, 3)}") print(f"view(2, 6):\n{t.view(2, 6)}") print(f"flatten(): {t.flatten()}") # 拼接 print("\n--- 拼接与拆分 ---") a = torch.tensor([[1, 2], [3, 4]]) b = torch.tensor([[5, 6], [7, 8]]) print(f"a:\n{a}") print(f"b:\n{b}") print(f"\ntorch.cat([a, b], dim=0) (纵向拼接):\n{torch.cat([a, b], dim=0)}") print(f"torch.cat([a, b], dim=1) (横向拼接):\n{torch.cat([a, b], dim=1)}") print(f"torch.stack([a, b]) (堆叠):\n{torch.stack([a, b])}") # ===================================================== # 7. 设备相关操作 # ===================================================== print("\n--- 7. 设备相关操作 ---\n") device = torch.device("cuda" if torch.cuda.is_available() else "cpu") print(f"当前设备: {device}") x = torch.rand(3, 3) print(f"Tensor 默认设备: {x.device}") # 如果有 GPU，演示迁移 if torch.cuda.is_available(): x_gpu = x.to(device) print(f"迁移到 GPU 后: {x_gpu.device}") print(f"GPU Tensor 示例:\n{x_gpu}") else: print("当前环境无 GPU，跳过 GPU 演示") # ===================================================== # 8. Tensor 与 NumPy 互转 # ===================================================== print("\n--- 8. Tensor 与 NumPy 互转 ---\n") import numpy as np # Tensor -> NumPy t = torch.tensor([1.0, 2.0, 3.0]) arr = t.numpy() print(f"Tensor: {t}") print(f"NumPy 数组: {arr}") print(f"类型: {type(arr)}") # NumPy -> Tensor arr2 = np.array([4.0, 5.0, 6.0]) t2 = torch.from_numpy(arr2) print(f"\nNumPy 数组: {arr2}") print(f"Tensor: {t2}") # 共享内存（重要！） print("\n⚠️ 注意：Tensor 和 NumPy 数组共享内存！") arr[0] = 999 print(f"修改 NumPy 数组后 Tensor: {t}") # t 也会变！ # ===================================================== # 9. 自动广播 # ===================================================== print("\n--- 9. 自动广播 (Broadcasting) ---\n") a = torch.zeros(3, 4, 5) b = torch.zeros(4, 1) # a + b 会自动广播 b 到 (3, 4, 5) c = a + b print(f"a shape: {a.shape}") print(f"b shape: {b.shape}") print(f"a + b shape: {c.shape} (自动广播)") a = torch.zeros(3, 1) b = torch.zeros(1, 4) c = a + b print(f"\na shape: {a.shape}") print(f"b shape: {b.shape}") print(f"a + b shape: {c.shape} (自动广播)") print("\n" + "=" * 60) print("Tensor 基础操作演示完成！") print("=" * 60) if __name__ == "__main__": main()