{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "a93bbf66",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Parameter containing:\n",
      "tensor([[-0.1953,  0.0349,  0.2066,  0.3050,  0.0570, -0.1622, -0.1848, -0.1546,\n",
      "         -0.1174,  0.1902],\n",
      "        [-0.2510,  0.1219, -0.1402, -0.0613, -0.0890,  0.0510, -0.0082, -0.2142,\n",
      "          0.1112,  0.0488],\n",
      "        [-0.1486, -0.2754, -0.0817,  0.2407, -0.2888,  0.3057,  0.1441,  0.1292,\n",
      "          0.0201, -0.1883],\n",
      "        [-0.1317, -0.2687, -0.2205,  0.1629,  0.0945, -0.0216, -0.0291, -0.2245,\n",
      "          0.2112,  0.2221],\n",
      "        [-0.1276,  0.2041,  0.2358, -0.0377, -0.2722, -0.0177, -0.0213, -0.0972,\n",
      "         -0.0971,  0.0606]], requires_grad=True)\n",
      "Parameter containing:\n",
      "tensor([ 0.2709,  0.0198, -0.1025, -0.0345,  0.1773], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "import torch.nn as nn\n",
    "class MyModel(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        self.fc = nn.Linear(10, 5)\n",
    "\n",
    "    def forward(self, x):\n",
    "        return self.fc(x)\n",
    "\n",
    "# 实例化模型时，fc 就已经被初始化了\n",
    "model = MyModel()\n",
    "print(model.fc.weight)\n",
    "print(model.fc.bias)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "b11a6f0e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Parameter containing:\n",
      "tensor([[ 0.4859,  0.4045, -0.3092,  0.0537, -0.3359, -0.1761,  0.4106, -0.6187,\n",
      "         -0.2196, -0.0027],\n",
      "        [ 0.3794,  0.5629, -0.4486, -0.1120, -0.3996,  0.0114, -0.4191, -0.1235,\n",
      "          0.2704, -0.4082],\n",
      "        [-0.6163,  0.3071, -0.2106, -0.4641, -0.2065, -0.5145,  0.0373, -0.2939,\n",
      "          0.5346,  0.2736],\n",
      "        [ 0.5658, -0.4418, -0.3231,  0.2414, -0.1687, -0.1383,  0.0326,  0.1102,\n",
      "          0.4646, -0.4835],\n",
      "        [-0.3686,  0.5512, -0.0530,  0.4921, -0.5184, -0.4362,  0.1822,  0.4276,\n",
      "         -0.4707, -0.2698]], requires_grad=True)\n",
      "Parameter containing:\n",
      "tensor([0., 0., 0., 0., 0.], requires_grad=True)\n"
     ]
    }
   ],
   "source": [
    "import torch.nn as nn\n",
    "\n",
    "# 创建一个线性层\n",
    "layer = nn.Linear(10, 5)\n",
    "\n",
    "# 最简单的用法：直接用 init 覆盖默认初始化\n",
    "nn.init.xavier_uniform_(layer.weight)  # 均匀分布初始化权重\n",
    "nn.init.zeros_(layer.bias)             # 偏置设为0\n",
    "\n",
    "print(layer.weight)\n",
    "print(layer.bias)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "74cdfef7",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "False\n",
      "True\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "\n",
    "# 1. 普通张量：默认不记录梯度\n",
    "x = torch.tensor([2.0, 3.0])\n",
    "print(x.requires_grad)  # False\n",
    "\n",
    "#PyTorch最核心的优势就是Autograd自动微分引擎，不用手动求导，代码一行开启自动计算梯度。\n",
    "# 2. 开启梯度记录（模型参数专用）只有设置requires_grad=True的张量，才会记录运算流程、支持求梯度。\n",
    "w = torch.tensor([1.5], requires_grad=True)\n",
    "print(w.requires_grad)  # True\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "b0cd775e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "w的梯度： tensor([-18.])\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "# 权重参数，开启梯度\n",
    "w = torch.tensor([2.0], requires_grad=True)\n",
    "# 正向运算\n",
    "y = w * 3\n",
    "# 模拟损失函数\n",
    "loss = (y - 9) ** 2\n",
    "# 反向传播，自动求导\n",
    "loss.backward()\n",
    "# 打印w的梯度\n",
    "print(\"w的梯度：\", w.grad)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "7ac1ab87",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0.27272727 0.40677966]\n",
      " [1.         0.        ]\n",
      " [0.         1.        ]]\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "# 假设这是\"视频时长\"和\"粉丝数\"两个特征，数量级差异巨大\n",
    "data = np.array([[60, 500000],    # 60秒，50万粉丝\n",
    "                 [180, 20000],    # 3分钟，2万粉丝\n",
    "                 [15, 1200000]])  # 15秒，120万粉丝\n",
    "\n",
    "# Min-Max 归一化：每个特征独立压缩到 [0, 1]\n",
    "data_norm = (data - data.min(axis=0)) / (data.max(axis=0) - data.min(axis=0))\n",
    "print(data_norm)\n",
    "# [[0.273, 0.407],\n",
    "#  [1.000, 0.000],\n",
    "#  [0.000, 1.000]]\n",
    "# 现在两个特征在同一个尺度上了\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5c538338",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[1, 0, 0, 0],\n",
      "        [0, 1, 0, 0],\n",
      "        [0, 0, 1, 0],\n",
      "        [0, 0, 0, 1]])\n"
     ]
    }
   ],
   "source": [
    "import torch.nn.functional as F\n",
    "import torch\n",
    "# 4个类别 \n",
    "labels = torch.tensor([0, 1,2,3])\n",
    "one_hot = F.one_hot(labels, num_classes=4)\n",
    "\n",
    "print(one_hot)\n",
    "# tensor([[1, 0, 0, 0],\n",
    "#         [0, 1, 0, 0],\n",
    "#         [0, 0, 1, 0],\n",
    "#         [0, 0, 0, 1]])\n"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "shuhe_xwdev (3.11.14)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}