{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# 初学入门\n", "\n", "[![](https://gitee.com/mindspore/docs/raw/r1.3/resource/_static/logo_source.png)](https://gitee.com/mindspore/docs/blob/r1.3/tutorials/source_zh_cn/quick_start.ipynb) [![](https://gitee.com/mindspore/docs/raw/r1.3/resource/_static/logo_notebook.png)](https://obs.dualstack.cn-north-4.myhuaweicloud.com/mindspore-website/notebook/r1.3/tutorials/zh_cn/mindspore_quick_start.ipynb) [![](https://gitee.com/mindspore/docs/raw/r1.3/resource/_static/logo_modelarts.png)](https://authoring-modelarts-cnnorth4.huaweicloud.com/console/lab?share-url-b64=aHR0cHM6Ly9vYnMuZHVhbHN0YWNrLmNuLW5vcnRoLTQubXlodWF3ZWljbG91ZC5jb20vbWluZHNwb3JlLXdlYnNpdGUvbm90ZWJvb2svbW9kZWxhcnRzL3F1aWNrX3N0YXJ0L21pbmRzcG9yZV9xdWlja19zdGFydC5pcHluYg==&imageid=65f636a0-56cf-49df-b941-7d2a07ba8c8c)\n", "\n", "本节贯穿MindSpore的基础功能，实现深度学习中的常见任务，请参考各节链接进行更加深入的学习。\n", "\n", "## 配置运行信息\n", "\n", "MindSpore通过`context.set_context`来配置运行需要的信息，如运行模式、后端信息、硬件等信息。\n", "\n", "导入`context`模块，配置运行需要的信息。" ] }, { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [], "source": [ "import os\n", "import argparse\n", "from mindspore import context\n", "\n", "parser = argparse.ArgumentParser(description='MindSpore LeNet Example')\n", "parser.add_argument('--device_target', type=str, default=\"CPU\", choices=['Ascend', 'GPU', 'CPU'])\n", "\n", "args = parser.parse_known_args()[0]\n", "context.set_context(mode=context.GRAPH_MODE, device_target=args.device_target)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "在样例中，我们配置样例运行使用图模式。根据实际情况配置硬件信息，譬如代码运行在Ascend AI处理器上，则`--device_target`选择`Ascend`，代码运行在CPU、GPU同理。详细参数说明，请参见[context.set_context](https://www.mindspore.cn/docs/api/zh-CN/r1.3/api_python/mindspore.context.html)接口说明。" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 下载数据集\n", "\n", "我们示例中用到的MNIST数据集是由10类28∗28的灰度图片组成，训练数据集包含60000张图片，测试数据集包含10000张图片。\n", "\n", "你可以从[MNIST数据集下载页面](http://yann.lecun.com/exdb/mnist/)下载，并按下方目录结构放置，如运行环境为Linux，还可以在Jupyter Notebook中直接运行如下命令完成下载和放置：" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "./datasets/MNIST_Data\n", "├── test\n", "│   ├── t10k-images-idx3-ubyte\n", "│   └── t10k-labels-idx1-ubyte\n", "└── train\n", " ├── train-images-idx3-ubyte\n", " └── train-labels-idx1-ubyte\n", "\n", "2 directories, 4 files\n" ] } ], "source": [ "!mkdir -p ./datasets/MNIST_Data/train ./datasets/MNIST_Data/test\n", "!wget -NP ./datasets/MNIST_Data/train https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/train-labels-idx1-ubyte --no-check-certificate\n", "!wget -NP ./datasets/MNIST_Data/train https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/train-images-idx3-ubyte --no-check-certificate\n", "!wget -NP ./datasets/MNIST_Data/test https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/t10k-labels-idx1-ubyte --no-check-certificate\n", "!wget -NP ./datasets/MNIST_Data/test https://mindspore-website.obs.myhuaweicloud.com/notebook/datasets/mnist/t10k-images-idx3-ubyte --no-check-certificate\n", "!tree ./datasets/MNIST_Data" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 数据处理\n", "\n", "数据集对于模型训练非常重要，好的数据集可以有效提高训练精度和效率。\n", "MindSpore提供了用于数据处理的API模块 `mindspore.dataset` ，用于存储样本和标签。在加载数据集前，我们通常会对数据集进行一些处理，`mindspore.dataset`也集成了常见的数据处理方法。\n", "\n", "首先导入MindSpore中`mindspore.dataset`和其他相应的模块。" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [], "source": [ "import mindspore.dataset as ds\n", "import mindspore.dataset.transforms.c_transforms as C\n", "import mindspore.dataset.vision.c_transforms as CV\n", "from mindspore.dataset.vision import Inter\n", "from mindspore import dtype as mstype" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "数据集处理主要分为四个步骤：\n", "\n", "1. 定义函数`create_dataset`来创建数据集。\n", "2. 定义需要进行的数据增强和处理操作，为之后进行map映射做准备。\n", "3. 使用map映射函数，将数据操作应用到数据集。\n", "4. 进行数据shuffle、batch操作。" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "def create_dataset(data_path, batch_size=32, repeat_size=1,\n", " num_parallel_workers=1):\n", " # 定义数据集\n", " mnist_ds = ds.MnistDataset(data_path)\n", " resize_height, resize_width = 32, 32\n", " rescale = 1.0 / 255.0\n", " shift = 0.0\n", " rescale_nml = 1 / 0.3081\n", " shift_nml = -1 * 0.1307 / 0.3081\n", "\n", " # 定义所需要操作的map映射\n", " resize_op = CV.Resize((resize_height, resize_width), interpolation=Inter.LINEAR)\n", " rescale_nml_op = CV.Rescale(rescale_nml, shift_nml)\n", " rescale_op = CV.Rescale(rescale, shift)\n", " hwc2chw_op = CV.HWC2CHW()\n", " type_cast_op = C.TypeCast(mstype.int32)\n", " \n", " # 使用map映射函数，将数据操作应用到数据集\n", " mnist_ds = mnist_ds.map(operations=type_cast_op, input_columns=\"label\", num_parallel_workers=num_parallel_workers)\n", " mnist_ds = mnist_ds.map(operations=resize_op, input_columns=\"image\", num_parallel_workers=num_parallel_workers)\n", " mnist_ds = mnist_ds.map(operations=rescale_op, input_columns=\"image\", num_parallel_workers=num_parallel_workers)\n", " mnist_ds = mnist_ds.map(operations=rescale_nml_op, input_columns=\"image\", num_parallel_workers=num_parallel_workers)\n", " mnist_ds = mnist_ds.map(operations=hwc2chw_op, input_columns=\"image\", num_parallel_workers=num_parallel_workers)\n", " \n", " # 进行shuffle、batch操作\n", " buffer_size = 10000\n", " mnist_ds = mnist_ds.shuffle(buffer_size=buffer_size)\n", " mnist_ds = mnist_ds.batch(batch_size, drop_remainder=True)\n", "\n", " return mnist_ds" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", "其中，`batch_size`为每组包含的数据个数，现设置每组包含32个数据。\n", "\n", "> MindSpore支持进行多种数据处理和增强的操作，具体可以参考[数据处理](https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.3/pipeline.html)和[数据增强](https://www.mindspore.cn/docs/programming_guide/zh-CN/r1.3/augmentation.html)章节。\n", "\n", "## 创建模型\n", "\n", "使用MindSpore定义神经网络需要继承`mindspore.nn.Cell`。`Cell`是所有神经网络（如`Conv2d-relu-softmax`等）的基类。\n", "\n", "神经网络的各层需要预先在`__init__`方法中定义，然后通过定义`construct`方法来完成神经网络的前向构造。按照LeNet的网络结构，定义网络各层如下：\n" ] }, { "cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [], "source": [ "import mindspore.nn as nn\n", "from mindspore.common.initializer import Normal\n", "\n", "class LeNet5(nn.Cell):\n", " \"\"\"\n", " Lenet网络结构\n", " \"\"\"\n", " def __init__(self, num_class=10, num_channel=1):\n", " super(LeNet5, self).__init__()\n", " # 定义所需要的运算\n", " self.conv1 = nn.Conv2d(num_channel, 6, 5, pad_mode='valid')\n", " self.conv2 = nn.Conv2d(6, 16, 5, pad_mode='valid')\n", " self.fc1 = nn.Dense(16 * 5 * 5, 120, weight_init=Normal(0.02))\n", " self.fc2 = nn.Dense(120, 84, weight_init=Normal(0.02))\n", " self.fc3 = nn.Dense(84, num_class, weight_init=Normal(0.02))\n", " self.relu = nn.ReLU()\n", " self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)\n", " self.flatten = nn.Flatten()\n", "\n", " def construct(self, x):\n", " # 使用定义好的运算构建前向网络\n", " x = self.conv1(x)\n", " x = self.relu(x)\n", " x = self.max_pool2d(x)\n", " x = self.conv2(x)\n", " x = self.relu(x)\n", " x = self.max_pool2d(x)\n", " x = self.flatten(x)\n", " x = self.fc1(x)\n", " x = self.relu(x)\n", " x = self.fc2(x)\n", " x = self.relu(x)\n", " x = self.fc3(x)\n", " return x\n", "\n", "# 实例化网络\n", "net = LeNet5()" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 优化模型参数\n", "\n", "要训练神经网络模型，需要定义损失函数和优化器。\n", "\n", "MindSpore支持的损失函数有`SoftmaxCrossEntropyWithLogits`、`L1Loss`、`MSELoss`等。这里使用交叉熵损失函数`SoftmaxCrossEntropyWithLogits`。\n" ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [], "source": [ "# 定义损失函数\n", "net_loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction='mean')" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", ">阅读更多有关[在MindSpore中使用损失函数](https://www.mindspore.cn/tutorials/zh-CN/r1.3/optimization.html#损失函数)的信息。\n", "\n", "MindSpore支持的优化器有`Adam`、`AdamWeightDecay`、`Momentum`等。这里使用`Momentum`优化器为例。" ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [], "source": [ "# 定义优化器\n", "net_opt = nn.Momentum(net.trainable_params(), learning_rate=0.01, momentum=0.9)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", ">阅读更多有关[在MindSpore中使用优化器](https://www.mindspore.cn/tutorials/zh-CN/r1.3/optimization.html#优化器)的信息。\n", "\n", "## 训练及保存模型\n", "\n", "MindSpore提供了回调Callback机制，可以在训练过程中执行自定义逻辑，这里以使用框架提供的`ModelCheckpoint`为例。\n", "`ModelCheckpoint`可以保存网络模型和参数，以便进行后续的Fine-tuning（微调）操作。\n" ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [], "source": [ "from mindspore.train.callback import ModelCheckpoint, CheckpointConfig\n", "# 设置模型保存参数\n", "config_ck = CheckpointConfig(save_checkpoint_steps=1875, keep_checkpoint_max=10)\n", "# 应用模型保存参数\n", "ckpoint = ModelCheckpoint(prefix=\"checkpoint_lenet\", config=config_ck)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", "通过MindSpore提供的`model.train`接口可以方便地进行网络的训练，`LossMonitor`可以监控训练过程中`loss`值的变化。\n" ] }, { "cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [], "source": [ "# 导入模型训练需要的库\n", "from mindspore.nn import Accuracy\n", "from mindspore.train.callback import LossMonitor\n", "from mindspore import Model" ] }, { "cell_type": "code", "execution_count": 10, "metadata": {}, "outputs": [], "source": [ "def train_net(args, model, epoch_size, data_path, repeat_size, ckpoint_cb, sink_mode):\n", " \"\"\"定义训练的方法\"\"\"\n", " # 加载训练数据集\n", " ds_train = create_dataset(os.path.join(data_path, \"train\"), 32, repeat_size)\n", " model.train(epoch_size, ds_train, callbacks=[ckpoint_cb, LossMonitor(125)], dataset_sink_mode=sink_mode)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", "其中，`dataset_sink_mode`用于控制数据是否下沉，数据下沉是指数据通过通道直接传送到Device上，可以加快训练速度，`dataset_sink_mode`为True表示数据下沉，否则为非下沉。\n", "\n", "通过模型运行测试数据集得到的结果，验证模型的泛化能力。\n", "\n", "1. 使用`model.eval`接口读入测试数据集。\n", "2. 使用保存后的模型参数进行推理。" ] }, { "cell_type": "code", "execution_count": 11, "metadata": {}, "outputs": [], "source": [ "def test_net(network, model, data_path):\n", " \"\"\"定义验证的方法\"\"\"\n", " ds_eval = create_dataset(os.path.join(data_path, \"test\"))\n", " acc = model.eval(ds_eval, dataset_sink_mode=False)\n", " print(\"{}\".format(acc))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", "这里把`train_epoch`设置为1，对数据集进行1个迭代的训练。在`train_net`和 `test_net`方法中，我们加载了之前下载的训练数据集，`mnist_path`是MNIST数据集路径。" ] }, { "cell_type": "code", "execution_count": 12, "metadata": {}, "outputs": [], "source": [ "train_epoch = 1\n", "mnist_path = \"./datasets/MNIST_Data\"\n", "dataset_size = 1\n", "model = Model(net, net_loss, net_opt, metrics={\"Accuracy\": Accuracy()})\n", "train_net(args, model, train_epoch, mnist_path, dataset_size, ckpoint, False)\n", "test_net(net, model, mnist_path)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", "使用以下命令运行脚本：\n", "\n", "```bash\n", "python lenet.py --device_target=CPU\n", "```\n", "\n", "其中，\n", "\n", "`lenet.py`：可以把前面的代码都粘贴到lenet.py中（不包含“下载数据集”的代码）。一般情况下，可将import部分移到代码头部，类、函数和方法的定义放在之后，最后在main方法中将前面的操作串起来即可。\n", "\n", "`--device_target=CPU`：指定运行硬件平台，参数为`CPU`、`GPU`或者`Ascend`，根据你的实际运行硬件平台来指定。\n", "\n", "训练过程中会打印loss值，类似下图。loss值会波动，但总体来说loss值会逐步减小，精度逐步提高。每个人运行的loss值有一定随机性，不一定完全相同。\n", "训练过程中loss打印示例如下：\n", "\n", "```text\n", "epoch: 1 step: 125, loss is 2.3083377\n", "epoch: 1 step: 250, loss is 2.3019726\n", "...\n", "epoch: 1 step: 1500, loss is 0.028385757\n", "epoch: 1 step: 1625, loss is 0.0857362\n", "epoch: 1 step: 1750, loss is 0.05639569\n", "epoch: 1 step: 1875, loss is 0.12366105\n", "{'Accuracy': 0.9663477564102564}\n", "```\n", "\n", "可以在打印信息中看出模型精度数据，示例中精度数据达到96.6%，模型质量良好。随着网络迭代次数`train_epoch`增加，模型精度会进一步提高。\n", "\n", "## 加载模型\n" ] }, { "cell_type": "code", "execution_count": 13, "metadata": {}, "outputs": [], "source": [ "from mindspore.train.serialization import load_checkpoint, load_param_into_net\n", "# 加载已经保存的用于测试的模型\n", "param_dict = load_checkpoint(\"checkpoint_lenet-1_1875.ckpt\")\n", "# 加载参数到网络中\n", "load_param_into_net(net, param_dict)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "\n", ">阅读更多有关[MindSpore加载模型](https://www.mindspore.cn/tutorials/zh-CN/r1.3/save_load_model.html#id3)的信息。\n", "\n", "## 验证模型\n", "\n", "我们使用生成的模型进行单个图片数据的分类预测，具体步骤如下：" ] }, { "cell_type": "code", "execution_count": 14, "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Predicted: \"6\", Actual: \"6\"\n" ] } ], "source": [ "import numpy as np\n", "from mindspore import Tensor\n", "\n", "# 定义测试数据集，batch_size设置为1，则取出一张图片\n", "ds_test = create_dataset(os.path.join(mnist_path, \"test\"), batch_size=1).create_dict_iterator()\n", "data = next(ds_test)\n", "\n", "# images为测试图片，labels为测试图片的实际分类\n", "images = data[\"image\"].asnumpy()\n", "labels = data[\"label\"].asnumpy()\n", "\n", "# 使用函数model.predict预测image对应分类\n", "output = model.predict(Tensor(data['image']))\n", "predicted = np.argmax(output.asnumpy(), axis=1)\n", "\n", "# 输出预测分类与实际分类\n", "print(f'Predicted: \"{predicted[0]}\", Actual: \"{labels[0]}\"')" ] } ], "metadata": { "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": "3.7.5" } }, "nbformat": 4, "nbformat_minor": 4 }