TensorFlow.js 示例

这个仓库包含了一组使用 TensorFlow.js 实现的示例。

每个示例目录都是独立的, 因此可以直接复制到其他项目中使用。

示例概览

<table> <tr> <th>示例名称</th> <th>演示链接</th> <th>输入数据类型</th> <th>任务类型</th> <th>模型类型</th> <th>训练</th> <th>推断</th> <th>API类型</th> <th>保存-加载操作</th> <tr> <td><a href="./abalone-node">abalone-node</a></td> <td></td> <td>数字</td> <td>从本地文件加载数据并在Node.js中进行训练</td> <td>多层感知机</td> <td>Node.js</td> <td>Node.js</td> <td>层</td> <td>保存到文件系统并在Node.js中加载</td> </tr> <tr> <td><a href="./addition-rnn">addition-rnn</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/addition-rnn/dist/index.html">🔗</a></td> <td>文本</td> <td>序列到序列</td> <td>RNN: SimpleRNN, GRU和LSTM</td> <td>浏览器</td> <td>浏览器</td> <td>层</td> <td></td> </tr> <tr> <td><a href="./addition-rnn-webworker">addition-rnn-webworker</a></td> <td></td> <td>文本</td> <td>序列到序列</td> <td>RNN: SimpleRNN, GRU和LSTM</td> <td>浏览器: Web Worker</td> <td>浏览器: Web Worker</td> <td>层</td> <td></td> </tr> <tr> <td><a href="./angular-predictive-prefetching">angular-predictive-prefetching</a></td> <td></td> <td>数字</td> <td>多类分类器</td> <td>DNN</td> <td></td> <td>浏览器: Service Worker</td> <td>层</td> <td></td> </tr> <tr> <td><a href="./baseball-node">baseball-node</a></td> <td></td> <td>数字</td> <td>多类分类</td> <td>多层感知机</td> <td>Node.js</td> <td>Node.js</td> <td>层</td> <td></td> </tr> <tr> <td><a href="./boston-housing">boston-housing</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/boston-housing/dist/index.html">🔗</a></td> <td>数字</td> <td>回归</td> <td>多层感知机</td> <td>浏览器</td> <td>浏览器</td> <td>层</td> <td></td> </tr> <tr> <td><a href="./cart-pole">cart-pole</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/cart-pole/dist/index.html">🔗</a></td> <td></td> <td>强化学习</td> <td>策略梯度</td> <td>浏览器</td> <td>浏览器</td> <td>层</td> <td>IndexedDB</td> </tr> <tr> <td><a href="./chrome-extension">chrome-extension</a></td> <td></td> <td>图像</td> <td>(部署TF.js在Chrome扩展)</td> <td>Convnet</td> <td></td> <td>浏览器</td> <td></td> <td></td> </tr> <tr> <td><a href="./custom-layer">custom-layer</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/custom-layer/dist/index.html">🔗</a></td> <td></td> <td>(定义一个自定义层子类型)</td> <td></td> <td></td> <td>浏览器</td> <td>层</td> <td></td> </tr> <tr> <td><a href="./data-csv">data-csv</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/data-csv/dist/index.html">🔗</a></td> <td></td> <td>从远程CSV构建tf.data.Dataset</td> <td></td> <td></td> <td></td> <td></td> <td></td> </tr> <tr> <td><a href="./data-generator">data-generator</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/data-generator/dist/index.html">🔗</a></td> <td></td> <td>使用生成器构建tf.data.Dataset</td> <td>回归</td> <td>浏览器</td> <td>浏览器</td> <td>层</td> <td></td> </tr> <tr> <td><a href="./date-conversion-attention">date-conversion-attention</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/date-conversion-attention/dist/index.html">🔗</a></td> <td>文本</td> <td>文本到文本转换</td> <td>注意力机制, RNN</td> <td>Node.js</td> <td>浏览器和Node.js</td> <td>层</td> <td>保存到文件系统并在浏览器中加载</td> </tr> <tr> <td><a href="./electron">electron</a></td> <td></td> <td>图像</td> <td>(部署TF.js在基于Electron的桌面应用程序)</td> <td>Convnet</td> <td></td> <td>Node.js</td> <td></td> <td></td> </tr> <tr> <td><a href="./fashion-mnist-vae">fashion-mnist-vae</a></td> <td></td> <td>图像</td> <td>生成性</td> <td>变分自编码器(VAE)</td> </table> <td>Node.js</td> <td>浏览器</td> <td>图层</td> <td>从tfjs-node导出已训练的模型，并在浏览器中加载</td> <tr> <td><a href="./interactive-visualizers">interactive-visualizers</a></td> <td></td> <td>图像</td> <td>多类分类、目标检测、分割</td> <td></td> <td></td> <td>浏览器</td> <td></td> <td></td> </tr> <tr> <td><a href="./iris">iris</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/iris/dist/index.html">🔗</a></td> <td>数值</td> <td>多类分类</td> <td>多层感知机</td> <td>浏览器</td> <td>浏览器</td> <td>图层</td> <td></td> </tr> <tr> <td><a href="./iris-fitDataset">iris-fitDataset</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/iris-fitDataset/dist/index.html">🔗</a></td> <td>数值</td> <td>多类分类</td> <td>多层感知机</td> <td>浏览器</td> <td>浏览器</td> <td>图层</td> <td></td> </tr> <tr> <td><a href="./jena-weather">jena-weather</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/jena-weather/dist/index.html">🔗</a></td> <td>序列</td> <td>序列到预测</td> <td>MLP和RNNs</td> <td>浏览器和Node</td> <td>浏览器</td> <td>图层</td> <td></td> </tr> <tr> <td><a href="./lstm-text-generation">lstm-text-generation</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/lstm-text-generation/dist/index.html">🔗</a></td> <td>文本</td> <td>序列预测</td> <td>RNN: LSTM</td> <td>浏览器</td> <td>浏览器</td> <td>图层</td> <td>IndexedDB</td> </tr> <tr> <td><a href="./mnist">mnist</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/mnist/dist/index.html">🔗</a></td> <td>图像</td> <td>多类分类</td> <td>卷积神经网络</td> <td>浏览器</td> <td>浏览器</td> <td>图层</td> <td></td> </tr> <tr> <td><a href="./mnist-acgan">mnist-acgan</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/mnist-acgan/dist/index.html">🔗</a></td> <td>图像</td> <td>生成对抗网络(GAN)</td> <td>卷积神经网络; GAN</td> <td>Node.js</td> <td>浏览器</td> <td>图层</td> <td>从Node.js保存到文件系统，并在浏览器中加载</td> </tr> <tr> <td><a href="./mnist-core">mnist-core</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/mnist-core/dist/index.html">🔗</a></td> <td>图像</td> <td>多类分类</td> <td>卷积神经网络</td> <td>浏览器</td> <td>浏览器</td> <td>核心(Ops)</td> <td></td> </tr> <tr> <td><a href="./mnist-node">mnist-node</a></td> <td></td> <td>图像</td> <td>多类分类</td> <td>卷积神经网络</td> <td>Node.js</td> <td>Node.js</td> <td>图层</td> <td>保存到文件系统</td> </tr> <tr> <td><a href="./mnist-transfer-cnn">mnist-transfer-cnn</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/mnist-transfer-cnn/dist/index.html">🔗</a></td> <td>图像</td> <td>多类分类(迁移学习)</td> <td>卷积神经网络</td> <td>浏览器</td> <td>浏览器</td> <td>图层</td> <td>加载预训练模型</td> </tr> <tr> <td><a href="./mobilenet">mobilenet</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/mobilenet/dist/index.html">🔗</a></td> <td>图像</td> <td>多类分类</td> <td>卷积神经网络</td> <td></td> <td>浏览器</td> <td>图层</td> <td>加载预训练模型</td> </tr> <tr> <td><a href="./polynomial-regression">polynomial-regression</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/polynomial-regression/dist/index.html">🔗</a></td> <td>数值</td> <td>回归</td> <td>浅层神经网络</td> <td>浏览器</td> <td>浏览器</td> <td>图层</td> <td></td> </tr> <tr> <td><a href="./polynomial-regression-core">polynomial-regression-core</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/polynomial-regression-core/dist/index.html">🔗</a></td> <td>数值</td> <td>回归</td> <td>浅层神经网络</td> <td>浏览器</td> <td>浏览器</td> <td>核心(Ops)</td> <td></td> </tr> <td><a href="./quantization">量化</a></td> <td></td> <td>各种</td> <td>演示训练后权重量化的效果</td> <td>各种</td> <td>Node.js</td> <td>Node.js</td> <td>层</td> <td></td> <td><a href="./sentiment">情感</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/sentiment/dist/index.html">🔗</a></td> <td>文本</td> <td>序列到二进制预测</td> <td>LSTM, 1D卷积网络</td> <td>Node.js或Python</td> <td>浏览器</td> <td>层</td> <td>从Keras和tfjs-node加载模型</td> <td><a href="./simple-object-detection">简单物体检测</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/simple-object-detection/dist/index.html">🔗</a></td> <td>图像</td> <td>物体检测</td> <td>卷积神经网络(迁移学习)</td> <td>Node.js</td> <td>浏览器</td> <td>层</td> <td>从tfjs-node导出训练好的模型,并在浏览器中加载</td> <td><a href="./snake-dqn">贪吃蛇-DQN</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/snake-dqn/index.html">🔗</a></td> <td></td> <td>强化学习</td> <td>深度Q网络(DQN)</td> <td>Node.js</td> <td>浏览器</td> <td>层</td> <td>从tfjs-node导出训练好的模型,并在浏览器中加载</td> <td><a href="./translation">翻译</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/translation/dist/index.html">🔗</a></td> <td>文本</td> <td>序列到序列</td> <td>LSTM编码器和解码器</td> <td>Node.js或Python</td> <td>浏览器</td> <td>层</td> <td>从Keras转换的模型加载</td> <td><a href="./tsne-mnist-canvas">tsne-mnist-画布</a></td> <td></td> <td></td> <td>降维和数据可视化</td> <td>tSNE</td> <td>浏览器</td> <td>浏览器</td> <td>核心(运算)</td> <td></td> <td><a href="./webcam-transfer-learning">网络摄像头迁移学习</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/webcam-transfer-learning/dist/index.html">🔗</a></td> <td>图像</td> <td>多类分类(迁移学习)</td> <td>卷积神经网络</td> <td>浏览器</td> <td>浏览器</td> <td>层</td> <td>加载预训练模型</td> <td><a href="./website-phishing">网站钓鱼</a></td> <td><a href="https://storage.googleapis.com/tfjs-examples/website-phishing/dist/index.html">🔗</a></td> <td>数值</td> <td>二元分类</td> <td>多层感知器</td> <td>浏览器</td> <td>浏览器</td> <td>层</td> <td></td>

依赖项

除了 getting_started 外，所有示例都需要安装以下依赖项。

Node.js 版本 8.9 或更高
NPM cli 或 Yarn

如何构建示例

转到目录

如果使用 yarn:

cd mnist-core
yarn
yarn watch

如果使用 npm:

cd mnist-core
npm install
npm run watch

详情

约定是每个示例都包含两个脚本:

yarn watch 或 npm run watch: 启动本地开发 HTTP 服务器,监视文件系统的更改,这样您就可以编辑代码(JS 或 HTML)并在刷新页面时立即看到更改。
yarn build 或 npm run build: 生成一个 dist/ 文件夹,其中包含构建制品,可用于部署。