{
"cells": [
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [],
"source": [
"%matplotlib inline\n",
"%reload_ext autoreload\n",
"%autoreload 1\n",
"%aimport d2lzh_pytorch\n",
"import torch\n",
"import torch.utils.data as Data\n",
"from IPython import display\n",
"from matplotlib import pyplot as plt\n",
"import numpy as np\n",
"import random\n",
"\n",
"num_inputs = 2\n",
"num_examples = 1000\n",
"true_w = [2, -3.4]\n",
"true_b = 4.2\n",
"features = torch.randn(num_examples, num_inputs, dtype=torch.float32)\n",
"labels = true_w[0] * features[:, 0] + true_w[1] * features[:, 1] + true_b\n",
"labels += torch.tensor(np.random.normal(0, 0.01, size=labels.size()), dtype=torch.float32)\n",
"batch_size = 10\n",
"# 将训练数据的特征和标签组合\n",
"dataset = Data.TensorDataset(features, labels)\n",
"# 随机读取小批量\n",
"data_iter = Data.DataLoader(dataset, batch_size, shuffle=True)"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"tensor([[-1.1650e+00, 3.9608e-01],\n",
" [-8.4998e-01, -1.5327e+00],\n",
" [-8.0021e-02, 5.4315e-01],\n",
" [-1.1686e+00, 1.3427e-03],\n",
" [ 5.8616e-01, 5.0903e-01],\n",
" [-1.2848e+00, 1.2465e+00],\n",
" [ 1.1104e-01, 1.7384e+00],\n",
" [-3.6663e-01, 2.8090e-01],\n",
" [ 2.9642e-02, 2.5131e-01],\n",
" [-1.2431e+00, 1.6079e+00]]) tensor([ 0.5214, 7.7054, 2.1807, 1.8651, 3.6499, -2.6117, -1.4787, 2.5015,\n",
" 3.4148, -3.7473])\n"
]
}
],
"source": [
"for X, y in data_iter:\n",
" print(X, y)\n",
" break"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {
"tags": [
"定义模型"
]
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Sequential(\n",
" (0): Linear(in_features=2, out_features=1, bias=True)\n",
")\n"
]
}
],
"source": [
"import torch.nn as nn\n",
"net = nn.Sequential(nn.Linear(num_inputs, 1))\n",
"print(net) # 使用print可以打印出网络的结构"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Parameter containing:\n",
"tensor([[0.0303, 0.3476]], requires_grad=True)\n",
"Parameter containing:\n",
"tensor([0.2868], requires_grad=True)\n"
]
}
],
"source": [
"for param in net.parameters():\n",
" print(param)"
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {
"tags": [
"初始化模型参数"
]
},
"outputs": [
{
"data": {
"text/plain": [
"Parameter containing:\n",
"tensor([0.], requires_grad=True)"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"from torch.nn import init\n",
"# net[0]这样根据下标访问子模块的写法只有当net是个ModuleList或者Sequential实例时才可以\n",
"init.normal_(net[0].weight, mean=0, std=0.01)\n",
"init.constant_(net[0].bias, val=0) # 也可以直接修改bias的data: net[0].bias.data.fill_(0)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {
"tags": [
"定义损失函数"
]
},
"outputs": [],
"source": [
"loss = nn.MSELoss()"
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {
"tags": [
"定义优化算法"
]
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"SGD (\n",
"Parameter Group 0\n",
" dampening: 0\n",
" lr: 0.03\n",
" momentum: 0\n",
" nesterov: False\n",
" weight_decay: 0\n",
")\n"
]
}
],
"source": [
"import torch.optim as optim\n",
"\n",
"optimizer = optim.SGD(net.parameters(), lr=0.03)\n",
"print(optimizer)"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {
"tags": [
"训练模型"
]
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"epoch 1, loss: 0.000241\n",
"epoch 2, loss: 0.000118\n",
"epoch 3, loss: 0.000060\n"
]
}
],
"source": [
"# 在使用Gluon训练模型时，我们通过调用optim实例的step函数来迭代模型参数。\n",
"#按照小批量随机梯度下降的定义，我们在step函数中指明批量大小，从而对批量中样本梯度求平均。\n",
"num_epochs = 3\n",
"for epoch in range(1, num_epochs + 1):\n",
" for X, y in data_iter:\n",
" output = net(X)\n",
" l = loss(output, y.view(-1, 1))\n",
" optimizer.zero_grad() # 梯度清零，等价于net.zero_grad()\n",
" l.backward()\n",
" optimizer.step()\n",
" print('epoch %d, loss: %f' % (epoch, l.item()))"
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[2, -3.4] Parameter containing:\n",
"tensor([[ 1.9996, -3.3990]], requires_grad=True)\n",
"4.2 Parameter containing:\n",
"tensor([4.1998], requires_grad=True)\n"
]
}
],
"source": [
"dense = net[0]\n",
"print(true_w, dense.weight)\n",
"print(true_b, dense.bias)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
}
],
"metadata": {
"celltoolbar": "Tags",
"kernelspec": {
"display_name": "Python 3",
"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": "3.6.10"
}
},
"nbformat": 4,
"nbformat_minor": 4
}