├── .gitignore
├── project
    ├── 2 归一化
    │   ├── index.html
    │   └── script.js
    ├── 6 过拟合
    │   ├── index.html
    │   ├── data.js
    │   └── script.js
    ├── 0 准备
    │   ├── index2.html
    │   ├── script.js
    │   ├── node.js
    │   └── index.html
    ├── 1 线性回归
    │   ├── index.html
    │   └── script.js
    ├── 10 加载已有模型-商标识别
    │   ├── index.html
    │   ├── utils.js
    │   └── script.js
    ├── 8 预训练模型-图片分类
    │   ├── index.html
    │   ├── script.js
    │   └── imagenet_classes.js
    ├── 9 迁移学习-商标识别
    │   ├── index.html
    │   ├── utils.js
    │   ├── data.js
    │   └── script.js
    ├── 11 预训练模型-语音识别
    │   ├── index.html
    │   └── script.js
    ├── 3 逻辑回归
    │   ├── index.html
    │   ├── data.js
    │   └── script.js
    ├── 4 XOR
    │   ├── index.html
    │   ├── data.js
    │   └── script.js
    ├── 7 手写数字识别
    │   ├── index.html
    │   ├── script.js
    │   └── data.js
    ├── 5 多分类
    │   ├── index.html
    │   ├── script.js
    │   └── data.js
    ├── 12 迁移学习-声控数据采集
    │   ├── index.html
    │   └── script.js
    ├── 13 加载已有数据-声控轮播图
    │   ├── index.html
    │   └── script.js
    └── 14 python与js模型互转
    │   └── note.md
├── package.json
└── README.md


/.gitignore:
--------------------------------------------------------------------------------
1 | node_modules
2 | dist
3 | .cache
4 | data


--------------------------------------------------------------------------------
/project/2 归一化/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>


--------------------------------------------------------------------------------
/project/6 过拟合/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>


--------------------------------------------------------------------------------
/project/0 准备/index2.html:
--------------------------------------------------------------------------------
1 | <script src="./script.js"></script>


--------------------------------------------------------------------------------
/project/1 线性回归/index.html:
--------------------------------------------------------------------------------
1 | <script src="./script.js"></script>


--------------------------------------------------------------------------------
/project/0 准备/script.js:
--------------------------------------------------------------------------------
1 | import * as tf from "@tensorflow/tfjs"
2 | 
3 | const a = tf.tensor('1')
4 | a.print()


--------------------------------------------------------------------------------
/project/0 准备/node.js:
--------------------------------------------------------------------------------
1 | const tf = require("@tensorflow/tfjs-node")
2 | 
3 | const a = tf.tensor([1, 2])
4 | a.print()


--------------------------------------------------------------------------------
/project/10 加载已有模型-商标识别/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <input type="file" onchange="predict(this.files[0])">
3 | <br>


--------------------------------------------------------------------------------
/project/8 预训练模型-图片分类/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <input type="file" onchange="predict(this.files[0])">
3 | <h2 id="output"></h2>


--------------------------------------------------------------------------------
/project/9 迁移学习-商标识别/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <input type="file" onchange="predict(this.files[0])">
3 | <button onclick="download()">下载模型</button>
4 | <h2 id="output"></h2>


--------------------------------------------------------------------------------
/project/11 预训练模型-语音识别/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <style>
3 |     #result>div {
4 |         float: left;
5 |         padding: 20px;
6 |     }
7 | </style>
8 | <div id="result"></div>


--------------------------------------------------------------------------------
/project/3 逻辑回归/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <form onsubmit="predict(this);return false;">
3 |     x: <input type="text" name="x"> 
4 |     y: <input type="text" name="y">
5 |     <button type="submit">预测</button>
6 | </form>


--------------------------------------------------------------------------------
/project/4 XOR/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <form action="" onsubmit="predict(this);return false;">
3 |     x: <input type="text" name="x">
4 |     y: <input type="text" name="y">
5 |     <button type="submit">预测</button>
6 | </form>


--------------------------------------------------------------------------------
/project/7 手写数字识别/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <canvas width="300" height="300" style="border: 2px solid #666;"></canvas>
3 | <button onclick="window.clear();" style="margin: 4px;">清除</button>
4 | <button onclick="window.predict();" style="margin: 4px;">预测</button>
5 | <h2 id="output"></h2>


--------------------------------------------------------------------------------
/project/5 多分类/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <form action="" onsubmit="predict(this); return false;">
3 |   花萼长度：<input type="text" name="a"><br>
4 |   花萼宽度：<input type="text" name="b"><br>
5 |   花瓣长度：<input type="text" name="c"><br>
6 |   花瓣宽度：<input type="text" name="d"><br>
7 |   <button type="submit">预测</button>
8 | </form>


--------------------------------------------------------------------------------
/project/12 迁移学习-声控数据采集/index.html:
--------------------------------------------------------------------------------
1 | <script src="script.js"></script>
2 | <button onclick="collect(this)">上一张</button>
3 | <button onclick="collect(this)">下一张</button>
4 | <button onclick="collect(this)">背景噪音</button>
5 | <button onclick="save()">保存</button>
6 | <pre id="count"></pre>
7 | <button onclick="train()">训练</button>
8 | <br><br> 监听开关：
9 | <input type="checkbox" onchange="toggle(this.checked)">


--------------------------------------------------------------------------------
/project/0 准备/index.html:
--------------------------------------------------------------------------------
 1 | <!DOCTYPE html>
 2 | <html lang="en">
 3 | 
 4 | <head>
 5 |     <meta charset="UTF-8">
 6 |     <meta name="viewport" content="width=device-width, initial-scale=1.0">
 7 |     <meta http-equiv="X-UA-Compatible" content="ie=edge">
 8 |     <title>tensorflow</title>
 9 | </head>
10 | 
11 | <body>
12 | 
13 | </body>
14 | <script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs@1.0.0/dist/tf.min.js"></script>
15 | <script>
16 |     const a = tf.tensor('1')
17 |     a.print()
18 | </script>
19 | 
20 | </html>


--------------------------------------------------------------------------------
/package.json:
--------------------------------------------------------------------------------
 1 | {
 2 |     "name": "my",
 3 |     "version": "1.0.0",
 4 |     "description": "",
 5 |     "main": "index.js",
 6 |     "dependencies": {
 7 |         "@tensorflow-models/speech-commands": "^0.4.2",
 8 |         "@tensorflow/tfjs-node": "^1.6.0",
 9 |         "@tensorflow/tfjs-vis": "^1.4.0"
10 |     },
11 |     "devDependencies": {},
12 |     "scripts": {
13 |         "test": "echo \"Error: no test specified\" && exit 1"
14 |     },
15 |     "author": "",
16 |     "license": "ISC",
17 |     "browserslist": [
18 |         "last 1 Chrome version"
19 |     ]
20 | }


--------------------------------------------------------------------------------
/project/10 加载已有模型-商标识别/utils.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | 
 3 | export function img2x(imgEl) {
 4 |     return tf.tidy(() => {
 5 |         const input = tf.browser.fromPixels(imgEl)
 6 |             .toFloat()
 7 |             .sub(255 / 2)
 8 |             .div(255 / 2)
 9 |             .reshape([1, 224, 224, 3]);
10 |         return input;
11 |     });
12 | }
13 | 
14 | export function file2img(f) {
15 |     return new Promise(resolve => {
16 |         const reader = new FileReader();
17 |         reader.readAsDataURL(f);
18 |         reader.onload = (e) => {
19 |             const img = document.createElement('img');
20 |             img.src = e.target.result;
21 |             img.width = 224;
22 |             img.height = 224;
23 |             img.onload = () => resolve(img);
24 |         };
25 |     });
26 | }


--------------------------------------------------------------------------------
/project/9 迁移学习-商标识别/utils.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | 
 3 | export function img2x(imgEl) {
 4 |     return tf.tidy(() => {
 5 |         const input = tf.browser.fromPixels(imgEl)
 6 |             .toFloat()
 7 |             .sub(255 / 2)
 8 |             .div(255 / 2)
 9 |             .reshape([1, 224, 224, 3]);
10 |         return input;
11 |     });
12 | }
13 | 
14 | export function file2image(file) {
15 |     return new Promise(resolve => {
16 |         const reader = new FileReader();
17 |         reader.readAsDataURL(f);
18 |         reader.onload = (e) => {
19 |             const img = document.createElement('img');
20 |             img.src = e.target.result;
21 |             img.width = 224;
22 |             img.height = 224;
23 |             img.onload = () => resolve(img);
24 |         };
25 |     })
26 | }


--------------------------------------------------------------------------------
/project/4 XOR/data.js:
--------------------------------------------------------------------------------
 1 | /**
 2 |  * @param {*} numSamples 随机生成点的数量
 3 |  */
 4 | export function getData(numSamples) {
 5 |   let points = [];
 6 | 
 7 |   function genGauss(cx, cy, label) {
 8 |     for (let i = 0; i < numSamples / 2; i++) {
 9 |       let x = normalRandom(cx);
10 |       let y = normalRandom(cy);
11 |       points.push({
12 |         x,
13 |         y,
14 |         label
15 |       });
16 |     }
17 |   }
18 | 
19 |   // 这里产生四类点
20 |   genGauss(2, 2, 0);
21 |   genGauss(-2, -2, 0);
22 |   genGauss(-2, 2, 1);
23 |   genGauss(2, -2, 1);
24 |   return points;
25 | }
26 | 
27 | /**
28 |  * @description 生成正态分布随机点
29 |  * @param mean 均值
30 |  * @param variance 方差，值越大点分布越分散，值越小点分布越密集
31 |  */
32 | function normalRandom(mean = 0, variance = 1) {
33 |   let v1, v2, s;
34 |   do {
35 |     v1 = 2 * Math.random() - 1;
36 |     v2 = 2 * Math.random() - 1;
37 |     s = v1 * v1 + v2 * v2;
38 |   } while (s > 1);
39 | 
40 |   let result = Math.sqrt(-2 * Math.log(s) / s) * v1;
41 |   return mean + Math.sqrt(variance) * result;
42 | }


--------------------------------------------------------------------------------
/project/3 逻辑回归/data.js:
--------------------------------------------------------------------------------
 1 | /**
 2 |  * @description 二分类数据集生成函数源码
 3 |  * @param {*} numSamples 随机生成点的数量
 4 |  */
 5 | export function getData(numSamples) {
 6 |   let points = [];
 7 | 
 8 |   function genGauss(cx, cy, label) {
 9 |     for (let i = 0; i < numSamples / 2; i++) {
10 |       let x = normalRandom(cx);
11 |       let y = normalRandom(cy);
12 |       points.push({
13 |         x,
14 |         y,
15 |         label
16 |       });
17 |     }
18 |   }
19 | 
20 |   genGauss(2, 2, 1); //生成以(2,2)为中心，label为1的点
21 |   genGauss(-2, -2, 0); //生成以(-2,-2)为中心，label为0的点
22 |   return points;
23 | }
24 | 
25 | /**
26 |  * @description 生成正态分布随机点
27 |  * @param mean 均值
28 |  * @param variance 方差，值越大点分布越分散，值越小点分布越密集
29 |  */
30 | function normalRandom(mean = 0, variance = 1) {
31 |   let v1, v2, s;
32 |   do {
33 |     v1 = 2 * Math.random() - 1;
34 |     v2 = 2 * Math.random() - 1;
35 |     s = v1 * v1 + v2 * v2;
36 |   } while (s > 1);
37 | 
38 |   let result = Math.sqrt(-2 * Math.log(s) / s) * v1;
39 |   return mean + Math.sqrt(variance) * result;
40 | }


--------------------------------------------------------------------------------
/project/6 过拟合/data.js:
--------------------------------------------------------------------------------
 1 | export function getData(numSamples, variance) {
 2 |     let points = [];
 3 |   
 4 |     function genGauss(cx, cy, label) {
 5 |       for (let i = 0; i < numSamples / 2; i++) {
 6 |         let x = normalRandom(cx, variance);
 7 |         let y = normalRandom(cy, variance);
 8 |         points.push({ x, y, label });
 9 |       }
10 |     }
11 |   
12 |     genGauss(2, 2, 1);
13 |     genGauss(-2, -2, 0);
14 |     return points;
15 |   }
16 |   
17 |   /**
18 |    * Samples from a normal distribution. Uses the seedrandom library as the
19 |    * random generator.
20 |    *
21 |    * @param mean The mean. Default is 0.
22 |    * @param variance The variance. Default is 1.
23 |    */
24 |   function normalRandom(mean = 0, variance = 1) {
25 |     let v1, v2, s;
26 |     do {
27 |       v1 = 2 * Math.random() - 1;
28 |       v2 = 2 * Math.random() - 1;
29 |       s = v1 * v1 + v2 * v2;
30 |     } while (s > 1);
31 |   
32 |     let result = Math.sqrt(-2 * Math.log(s) / s) * v1;
33 |     return mean + Math.sqrt(variance) * result;
34 |   }


--------------------------------------------------------------------------------
/project/1 线性回归/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import * as tfvis from '@tensorflow/tfjs-vis';
 3 | 
 4 | window.onload = async () => {
 5 |   // x、y样本数据集
 6 |   const xs = [1, 2, 3, 4];
 7 |   const ys = [1, 3, 5, 7];
 8 | 
 9 |   // 使用 @tensorflow/tfjs-vis 库绘制散点图
10 |   tfvis.render.scatterplot(
11 |     { name: '线性回归训练集' },
12 |     { values: xs.map((x, i) => ({ x, y: ys[i] })) },
13 |     { xAxisDomain: [0, 5], yAxisDomain: [0, 8] }
14 |   );
15 | 
16 |   // 定义模型。添加含有1个神经元的1层网络
17 |   const model = tf.sequential();
18 |   model.add(tf.layers.dense({ units: 1, inputShape: [1] }));
19 |   model.compile({ loss: tf.losses.meanSquaredError, optimizer: tf.train.sgd(0.1) });
20 | 
21 |   const inputs = tf.tensor(xs);
22 |   const labels = tf.tensor(ys);
23 | 
24 |   // 训练模型
25 |   await model.fit(inputs, labels, {
26 |     batchSize: 4,
27 |     epochs: 200,
28 |     callbacks: tfvis.show.fitCallbacks(
29 |       { name: '训练过程' },
30 |       ['loss']
31 |     )
32 |   });
33 | 
34 |   // 预测
35 |   const output = model.predict(tf.tensor([5]));
36 |   console.log(`如果 x 为 5，那么预测 y 为 ${output.dataSync()[0]}`);
37 | };


--------------------------------------------------------------------------------
/project/13 加载已有数据-声控轮播图/index.html:
--------------------------------------------------------------------------------
 1 | <script src="script.js"></script>
 2 | 监听开关：<input type="checkbox" onchange="toggle(this.checked)">
 3 | 
 4 | <style>
 5 |     .slider {
 6 |         width: 600px;
 7 |         overflow: hidden;
 8 |         margin: 10px auto;
 9 |     }
10 |     
11 |     .slider>div {
12 |         display: flex;
13 |         align-items: center;
14 |     }
15 | </style>
16 | <div class="slider">
17 |     <div>
18 |         <img src="https://cdn.pixabay.com/photo/2019/10/29/15/57/vancouver-4587302__480.jpg" alt="" width="600">
19 |         <img src="https://cdn.pixabay.com/photo/2019/10/31/07/14/coffee-4591159__480.jpg" alt="" width="600">
20 |         <img src="https://cdn.pixabay.com/photo/2019/11/01/11/08/landscape-4593909__480.jpg" alt="" width="600">
21 |         <img src="https://cdn.pixabay.com/photo/2019/11/02/21/45/maple-leaf-4597501__480.jpg" alt="" width="600">
22 |         <img src="https://cdn.pixabay.com/photo/2019/11/02/03/13/in-xinjiang-4595560__480.jpg" alt="" width="600">
23 |         <img src="https://cdn.pixabay.com/photo/2019/11/01/22/45/reschensee-4595385__480.jpg" alt="" width="600">
24 |     </div>
25 | </div>


--------------------------------------------------------------------------------
/project/9 迁移学习-商标识别/data.js:
--------------------------------------------------------------------------------
 1 | import {
 2 |     promises
 3 | } from "dns";
 4 | 
 5 | const IMAGE_SIZE = 224;
 6 | 
 7 | const loadImg = src => {
 8 |     return new Promise(resolve => {
 9 |         const img = new Image()
10 |         img.crossOrigin = 'anonymous'
11 |         img.src = src
12 |         img.width = IMAGE_SIZE;
13 |         img.height = IMAGE_SIZE;
14 |         img.onload = () => resolve(img)
15 |     })
16 | }
17 | 
18 | // 输入数据格式
19 | export const getInputs = async() => {
20 |     const imgList = [],
21 |         labelList = []
22 |     for (let i = 0; i < 30; i++) {
23 |         ['android', 'apple', 'windows'].forEach(label => {
24 |             const src = `http://127.0.0.1:8081/brand/train/${label}-${i}.jpg`;
25 |             const img = loadImg(src)
26 |             imgList.push(img)
27 |             labelList.push([
28 |                 label === 'android' ? 1 : 0,
29 |                 label === 'apple' ? 1 : 0,
30 |                 label === 'windows' ? 1 : 0,
31 |             ])
32 |         })
33 |     }
34 |     const inputs = await Promise.all(imgList)
35 |     return {
36 |         inputs,
37 |         labelList
38 |     }
39 | }


--------------------------------------------------------------------------------
/project/11 预训练模型-语音识别/script.js:
--------------------------------------------------------------------------------
 1 | import * as speechCommands from '@tensorflow-models/speech-commands';
 2 | const MODEL_PATH = 'http://127.0.0.1:8081/speech';
 3 | 
 4 | window.onload = async() => {
 5 |     const recognizer = speechCommands.create(
 6 |         'BROWSER_FFT', //傅里叶变换
 7 |         null,
 8 |         MODEL_PATH + '/model.json',
 9 |         MODEL_PATH + '/metadata.json'
10 |     )
11 |     await recognizer.ensureModelLoaded()
12 | 
13 | 
14 |     const labels = recognizer.wordLabels().slice(2)
15 |     const resultEl = document.querySelector('#result')
16 |     resultEl.innerHTML = labels.map(l => `
17 |     <div>${l}</div>
18 | `).join('');
19 | 
20 |     // 语音识别
21 |     recognizer.listen(resilt => {
22 |         const {
23 |             scores
24 |         } = result;
25 |         const maxValue = Math.max(...scores);
26 |         const index = scores.indexOf(maxValue) - 2;
27 |         resultEl.innerHTML = labels.map((l, i) => `
28 |         <div style="background: ${i === index && 'green'}">${l}</div>
29 |         `).join('');
30 |     }, {
31 |         overlapFactor: 0.3, //控制频率，值越大，识别的次数越多
32 |         probabilityThreshold: 0.9 //准确度
33 |     })
34 | }


--------------------------------------------------------------------------------
/project/10 加载已有模型-商标识别/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import {
 3 |     img2x,
 4 |     file2img
 5 | } from './utils';
 6 | 
 7 | const MODEL_PATH = 'http://127.0.0.1:8081';
 8 | const BRAND_CLASSES = ['android', 'apple', 'windows'];
 9 | 
10 | window.onload = async() => {
11 |     const mobilenet = await tf.loadLayersModel(MODEL_PATH + '/mobilenet/web_model/model.json');
12 |     mobilenet.summary();
13 |     const layer = mobilenet.getLayer('conv_pw_13_relu');
14 |     const truncatedMobilenet = tf.model({
15 |         inputs: mobilenet.inputs,
16 |         outputs: layer.output
17 |     });
18 | 
19 |     const model = await tf.loadLayersModel(MODEL_PATH + '/brand/web_model/model.json');
20 | 
21 |     window.predict = async(file) => {
22 |         const img = await file2img(file);
23 |         document.body.appendChild(img);
24 |         const pred = tf.tidy(() => {
25 |             const x = img2x(img);
26 |             const input = truncatedMobilenet.predict(x);
27 |             return model.predict(input);
28 |         });
29 | 
30 |         const index = pred.argMax(1).dataSync()[0];
31 |         setTimeout(() => {
32 |             alert(`预测结果：${BRAND_CLASSES[index]}`);
33 |         }, 0);
34 |     };
35 | };


--------------------------------------------------------------------------------
/project/2 归一化/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import * as tfvis from '@tensorflow/tfjs-vis';
 3 | 
 4 | window.onload = async () => {
 5 |   // 样本数据
 6 |   const heights = [150, 160, 170];
 7 |   const weights = [40, 50, 60];
 8 | 
 9 |   // 绘制散点图
10 |   tfvis.render.scatterplot({
11 |     name: '身高体重训练数据'
12 |   }, {
13 |     values: heights.map((x, i) => ({
14 |       x,
15 |       y: weights[i]
16 |     }))
17 |   }, {
18 |     xAxisDomain: [140, 180],
19 |     yAxisDomain: [30, 70]
20 |   });
21 | 
22 |   // 归一化数据
23 |   const inputs = tf.tensor(heights).sub(150).div(20); //压缩到0-1
24 |   const labels = tf.tensor(weights).sub(40).div(20); //压缩到0-1
25 | 
26 |   // 构建模型
27 |   const model = tf.sequential();
28 |   model.add(tf.layers.dense({
29 |     units: 1,
30 |     inputShape: [1]
31 |   }));
32 |   model.compile({
33 |     loss: tf.losses.meanSquaredError,
34 |     optimizer: tf.train.sgd(0.1)
35 |   });
36 | 
37 |   // 训练模型
38 |   await model.fit(inputs, labels, {
39 |     batchSize: 3,
40 |     epochs: 200,
41 |     callbacks: tfvis.show.fitCallbacks({
42 |       name: '训练过程'
43 |     }, ['loss'])
44 |   });
45 | 
46 |   // 预测 反归一化
47 |   const output = model.predict(tf.tensor([180]).sub(150).div(20));
48 |   console.log(`如果身高为 180cm，那么预测体重为 ${output.mul(20).add(40).dataSync()[0]}kg`);
49 | };


--------------------------------------------------------------------------------
/project/4 XOR/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import * as tfvis from '@tensorflow/tfjs-vis';
 3 | import { getData } from './data.js';
 4 | 
 5 | // 不能使用线性逻辑回归
 6 | window.onload = async () => {
 7 |   const data = getData(400);
 8 | 
 9 |   tfvis.render.scatterplot({
10 |     name: 'XOR 训练数据'
11 |   }, {
12 |     values: [
13 |       data.filter(p => p.label === 1),
14 |       data.filter(p => p.label === 0),
15 |     ]
16 |   });
17 | 
18 |   // 构建多层神经网络
19 |   const model = tf.sequential();
20 |   model.add(tf.layers.dense({
21 |     units: 4,
22 |     inputShape: [2],
23 |     activation: 'relu'
24 |   }));
25 |   model.add(tf.layers.dense({
26 |     units: 1,
27 |     activation: 'sigmoid'
28 |   }));
29 |   model.compile({
30 |     loss: tf.losses.logLoss,
31 |     optimizer: tf.train.adam(0.1)
32 |   });
33 | 
34 |   const inputs = tf.tensor(data.map(p => [p.x, p.y]));
35 |   const labels = tf.tensor(data.map(p => p.label));
36 | 
37 |   await model.fit(inputs, labels, {
38 |     epochs: 10,
39 |     callbacks: tfvis.show.fitCallbacks({
40 |       name: '训练效果'
41 |     }, ['loss'])
42 |   });
43 | 
44 |   window.predict = (form) => {
45 |     const pred = model.predict(tf.tensor([
46 |       [form.x.value * 1, form.y.value * 1]
47 |     ]));
48 |     console.log(`预测结果：${pred.dataSync()[0]}`);
49 |   };
50 | };


--------------------------------------------------------------------------------
/project/3 逻辑回归/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import * as tfvis from '@tensorflow/tfjs-vis';
 3 | import { getData } from './data.js';
 4 | 
 5 | window.onload = async () => {
 6 |   // 拿到400个数据
 7 |   const data = getData(400);
 8 | 
 9 |   // 绘制散点图
10 |   tfvis.render.scatterplot({
11 |     name: '逻辑回归训练数据'
12 |   }, {
13 |     values: [
14 |       data.filter(p => p.label === 1),
15 |       data.filter(p => p.label === 0),
16 |     ]
17 |   });
18 | 
19 |   // 构建模型，主要这里是逻辑回归，使用的损失函数为对数损失 logLoss，可参考http://wiki.fast.ai/index.php/Log_Loss
20 |   const model = tf.sequential();
21 |   model.add(tf.layers.dense({
22 |     units: 1,
23 |     inputShape: [2], //特征数量
24 |     activation: 'sigmoid'
25 |   }));
26 |   model.compile({
27 |     loss: tf.losses.logLoss, //对数损失，逻辑回归不适合使用MSE
28 |     optimizer: tf.train.adam(0.1) //adam优化器，可以自动调节学习率
29 |   });
30 | 
31 |   const inputs = tf.tensor(data.map(p => [p.x, p.y]));
32 |   const labels = tf.tensor(data.map(p => p.label));
33 | 
34 |   // 训练模型
35 |   await model.fit(inputs, labels, {
36 |     batchSize: 40,
37 |     epochs: 20,
38 |     callbacks: tfvis.show.fitCallbacks({
39 |       name: '训练效果'
40 |     }, ['loss'])
41 |   });
42 | 
43 |   // 编写界面输入预测值
44 |   window.predict = (form) => {
45 |     const pred = model.predict(tf.tensor([
46 |       [form.x.value * 1, form.y.value * 1]
47 |     ]));
48 |     console.log(`预测结果：${pred.dataSync()[0]}`);
49 |   };
50 | };


--------------------------------------------------------------------------------
/project/8 预训练模型-图片分类/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import {
 3 |     IMAGENET_CLASSES
 4 | } from './imagenet_classes';
 5 | 
 6 | const MOBILENET_MODEL_PATH = 'http://127.0.0.1:8081/mobilenet/web_model/model.json';
 7 | 
 8 | // file变为img标签
 9 | function file2img(file) {
10 |     return new Promise(resolve => {
11 |         const reader = new FileReader()
12 |         reader.readAsDataURL(file)
13 |         reader.onload = e => {
14 |             const img = document.createElement('img');
15 |             img.src = e.target.result;
16 |             img.width = 224;
17 |             img.height = 224;
18 |             img.onload = () => resolve(img);
19 |         }
20 |     })
21 | }
22 | 
23 | window.onload = async() => {
24 |     const model = await tf.loadLayersModel(MOBILENET_MODEL_PATH);
25 |     window.predict = async(file) => {
26 |         const img = await file2img(file);
27 |         document.body.appendChild(img);
28 |         const pred = tf.tidy(() => {
29 |             const input = tf.browser.fromPixels(img)
30 |                 .toFloat()
31 |                 .sub(255 / 2)
32 |                 .div(255 / 2) //转换到[-1,1]
33 |                 .reshape([1, 224, 224, 3]); //彩色图片
34 |             return model.predict(input);
35 |         })
36 |         const index = pred.argMax(1).dataSync()[0] //index的值
37 |         document.getElementById('output').innerHTML = `预测结果：${IMAGENET_CLASSES[index]}`
38 |     }
39 | }


--------------------------------------------------------------------------------
/project/6 过拟合/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import * as tfvis from '@tensorflow/tfjs-vis';
 3 | import {
 4 |     getData
 5 | } from './data';
 6 | 
 7 | window.onload = async() => {
 8 |     const data = getData(200, 2);
 9 | 
10 |     tfvis.render.scatterplot({
11 |         name: '训练数据'
12 |     }, {
13 |         values: [
14 |             data.filter(p => p.label === 1),
15 |             data.filter(p => p.label === 0),
16 |         ]
17 |     });
18 | 
19 |     // 解决overfit
20 |     // 1、早停
21 |     // 2、正则化
22 |     // 3、随机丢弃
23 |     const model = tf.sequential();
24 |     model.add(tf.layers.dense({
25 |         units: 10,
26 |         inputShape: [2],
27 |         activation: "tanh",
28 |         // kernelRegularizer: tf.regularizers.l2({ l2: 1 })
29 |     }));
30 |     model.add(tf.layers.dropout({
31 |         rate: 0.9
32 |     }));
33 |     model.add(tf.layers.dense({
34 |         units: 1,
35 |         activation: 'sigmoid'
36 |     }));
37 |     model.compile({
38 |         loss: tf.losses.logLoss,
39 |         optimizer: tf.train.adam(0.1)
40 |     });
41 | 
42 |     const inputs = tf.tensor(data.map(p => [p.x, p.y]));
43 |     const labels = tf.tensor(data.map(p => p.label));
44 | 
45 |     await model.fit(inputs, labels, {
46 |         validationSplit: 0.2,
47 |         epochs: 200,
48 |         callbacks: tfvis.show.fitCallbacks({
49 |             name: '训练效果'
50 |         }, ['loss', 'val_loss'], {
51 |             callbacks: ['onEpochEnd']
52 |         })
53 |     });
54 | };


--------------------------------------------------------------------------------
/project/5 多分类/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import * as tfvis from '@tensorflow/tfjs-vis';
 3 | import { getIrisData, IRIS_CLASSES } from './data';
 4 | 
 5 | window.onload = async () => {
 6 |   // 加载数据集和验证集
 7 |   const [xTrain, yTrain, xTest, yTest] = getIrisData(0.15); //15%的数据用于验证集
 8 | 
 9 |   // 构建模型，这里用到的损失函数为交叉熵损失 categoricalCrossentropy，适用于多分类
10 |   const model = tf.sequential();
11 |   model.add(tf.layers.dense({
12 |     units: 10,
13 |     inputShape: [xTrain.shape[1]],
14 |     activation: 'sigmoid' //也可以使用其他激活函数，如relu
15 |   }));
16 |   model.add(tf.layers.dense({
17 |     units: 3,
18 |     activation: 'softmax' //这里使用softmax激活函数，算出每个类别的概率
19 |   }));
20 |   model.compile({
21 |     loss: 'categoricalCrossentropy', //交叉熵损失，对数损失函数的多分类版本
22 |     optimizer: tf.train.adam(0.1),
23 |     metrics: ['accuracy']
24 |   });
25 | 
26 |   // 训练模型
27 |   await model.fit(xTrain, yTrain, {
28 |     epochs: 100,
29 |     validationData: [xTest, yTest],
30 |     callbacks: tfvis.show.fitCallbacks({
31 |       name: '训练效果'
32 |     }, ['loss', 'val_loss', 'acc', 'val_acc'], {
33 |       callbacks: ['onEpochEnd']
34 |     })
35 |   });
36 | 
37 |   window.predict = (form) => {
38 |     const input = tf.tensor([
39 |       [
40 |         form.a.value * 1,
41 |         form.b.value * 1,
42 |         form.c.value * 1,
43 |         form.d.value * 1,
44 |       ]
45 |     ]);
46 |     const pred = model.predict(input);
47 |     console.log(`预测结果：${IRIS_CLASSES[pred.argMax(1).dataSync(0)]}`); //argmax？
48 |   };
49 | };


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | Tensorflow.js Converter 依赖**python3.6.8**版本
 2 | 
 3 | **搭建虚拟环境：**
 4 | 1. 安装conda
 5 | 可以使用[清华镜像源](https://mirror.tuna.tsinghua.edu.cn/help/anaconda/)，安装Miniconda 即可（注意配置环境变量）
 6 | 2. 在终端检查conda命令是否可用，并创建指定python版本的虚拟环境
 7 | `conda create -n [name] python=3.6.8` 创建虚拟环境
 8 | `conda remove -n [name] --all` 删除虚拟环境
 9 | `conda info --envs` 查看虚拟环境
10 | `conda activate [name]` 激活虚拟环境（可以发现python版本已经改变）
11 | `conda deactivate [name]` 退出虚拟环境
12 | 3. 安装tfjs converter
13 | `pip install tensorflowjs`安装
14 | `tensorflowjs_converter -h`检查是否安装成功
15 | 
16 | 
17 | **python与JavaScript模型的互转：**
18 | 1. 准备工作
19 | `conda activate [name]`激活
20 | `tensorflowjs_converter`检查
21 | 2. 开始转换——具体格式建议看文档（github tfjs-converter）
22 |   1）python模型转js模型
23 |   `tensorflowjs_converter --input_format=keras --output_format=tf_layers_model [input_path] [output_path]`
24 |   2）js模型转python模型
25 |   `tensorflowjs_converter --input_format=tf_layers_model --output_format=keras [input_path] [output_path]`
26 | 3. 验证模型正确性
27 | 
28 | **模型加速：**
29 | 1. 分片(--weight_shared_size_bytes=[size])
30 |  `tensorflowjs_converter --input_format=tf_layers_model --output_format=tf_layers_model --weight_shared_size_bytes=100000 [input_path] [output_path]`
31 | 在输出文件夹中会发现很多块分片后的模型，这样在加载模型时可以使用并发实现加速
32 | 2. 量化(--quantization_bytes=[n])
33 |  `tensorflowjs_converter --input_format=tf_layers_model --output_format=tf_layers_model --quantization_bytes=2 [input_path] [output_path]`
34 | 3. 通过转为tfjs_graph_model来加速模型——内部实现了凸优化(--output_format=tf_graph_model)
35 |   `tensorflowjs_converter --input_format=tf_layers_model --output_format=tf_graph_model [input_path] [output_path]`
36 | 
37 | **拓展学习：**
38 | 1. 举一反三，训练不同应用场景的AI模型
39 | 2. 多看官方文档
40 | 3. 使用更多的预训练模型，如目标检测、NLP、人体姿势识别
41 | 4. 做更多智能相关的开源项目或商业项目，增加技术影响力


--------------------------------------------------------------------------------
/project/14 python与js模型互转/note.md:
--------------------------------------------------------------------------------
 1 | Tensorflow.js Converter 依赖**python3.6.8**版本
 2 | 
 3 | **搭建虚拟环境：**
 4 | 1. 安装conda
 5 | 可以使用[清华镜像源](https://mirror.tuna.tsinghua.edu.cn/help/anaconda/)，安装Miniconda 即可（注意配置环境变量）
 6 | 2. 在终端检查conda命令是否可用，并创建指定python版本的虚拟环境
 7 | `conda create -n [name] python=3.6.8` 创建虚拟环境
 8 | `conda remove -n [name] --all` 删除虚拟环境
 9 | `conda info --envs` 查看虚拟环境
10 | `conda activate [name]` 激活虚拟环境（可以发现python版本已经改变）
11 | `conda deactivate [name]` 退出虚拟环境
12 | 3. 安装tfjs converter
13 | `pip install tensorflowjs`安装
14 | `tensorflowjs_converter -h`检查是否安装成功
15 | 
16 | 
17 | **python与JavaScript模型的互转：**
18 | 1. 准备工作
19 | `conda activate [name]`激活
20 | `tensorflowjs_converter`检查
21 | 2. 开始转换——具体格式建议看文档（github tfjs-converter）
22 |   1）python模型转js模型
23 |   `tensorflowjs_converter --input_format=keras --output_format=tf_layers_model [input_path] [output_path]`
24 |   2）js模型转python模型
25 |   `tensorflowjs_converter --input_format=tf_layers_model --output_format=keras [input_path] [output_path]`
26 | 3. 验证模型正确性
27 | 
28 | **模型加速：**
29 | 1. 分片(--weight_shared_size_bytes=[size])
30 |  `tensorflowjs_converter --input_format=tf_layers_model --output_format=tf_layers_model --weight_shared_size_bytes=100000 [input_path] [output_path]`
31 | 在输出文件夹中会发现很多块分片后的模型，这样在加载模型时可以使用并发实现加速
32 | 2. 量化(--quantization_bytes=[n])
33 |  `tensorflowjs_converter --input_format=tf_layers_model --output_format=tf_layers_model --quantization_bytes=2 [input_path] [output_path]`
34 | 3. 通过转为tfjs_graph_model来加速模型——内部实现了凸优化(--output_format=tf_graph_model)
35 |   `tensorflowjs_converter --input_format=tf_layers_model --output_format=tf_graph_model [input_path] [output_path]`
36 | 
37 | **拓展学习：**
38 | 1. 举一反三，训练不同应用场景的AI模型
39 | 2. 多看官方文档
40 | 3. 使用更多的预训练模型，如目标检测、NLP、人体姿势识别
41 | 4. 做更多智能相关的开源项目或商业项目，增加技术影响力


--------------------------------------------------------------------------------
/project/13 加载已有数据-声控轮播图/script.js:
--------------------------------------------------------------------------------
 1 | import * as speechCommands from '@tensorflow-models/speech-commands';
 2 | 
 3 | const MODEL_PATH = 'http://127.0.0.1:8081';
 4 | let transferRecognizer;
 5 | 
 6 | window.onload = async() => {
 7 |     const recognizer = speechCommands.create(
 8 |         'BROWSER_FFT',
 9 |         null,
10 |         MODEL_PATH + '/speech/model.json',
11 |         MODEL_PATH + '/speech/metadata.json',
12 |     );
13 |     await recognizer.ensureModelLoaded();
14 |     transferRecognizer = recognizer.createTransfer('轮播图');
15 | 
16 |     // 在这里加载已保存的声音数据
17 |     const res = await fetch(MODEL_PATH + '/slider/data.bin');
18 |     const arrayBuffer = await res.arrayBuffer();
19 |     transferRecognizer.loadExamples(arrayBuffer);
20 |     await transferRecognizer.train({
21 |         epochs: 30
22 |     });
23 |     console.log('done');
24 | };
25 | 
26 | window.toggle = async(checked) => {
27 |     if (checked) {
28 |         await transferRecognizer.listen(result => {
29 |             const {
30 |                 scores
31 |             } = result;
32 |             const labels = transferRecognizer.wordLabels();
33 |             const index = scores.indexOf(Math.max(...scores));
34 |             console.log(labels[index]); //“上一张”/“下一张”
35 |             window.play(labels[index]); //变换轮播图
36 |         }, {
37 |             overlapFactor: 0,
38 |             probabilityThreshold: 0.5
39 |         });
40 |     } else {
41 |         transferRecognizer.stopListening();
42 |     }
43 | };
44 | 
45 | let curIndex = 0;
46 | const sliderCount = document.querySelectorAll('img').length
47 | window.play = label => {
48 |     const div = document.querySelector('.slider>div');
49 |     if (label === '上一张') {
50 |         if (curIndex === 0) {
51 |             return;
52 |         }
53 |         curIndex -= 1;
54 |     } else {
55 |         if (curIndex === sliderCount - 1) {
56 |             return;
57 |         }
58 |         curIndex += 1;
59 |     }
60 |     div.style.transition = "transform 1s"
61 |     div.style.transform = `translateX(-${100 * curIndex}%)`;
62 | }


--------------------------------------------------------------------------------
/project/12 迁移学习-声控数据采集/script.js:
--------------------------------------------------------------------------------
 1 | import * as speechCommands from '@tensorflow-models/speech-commands';
 2 | import * as tfvis from '@tensorflow/tfjs-vis';
 3 | 
 4 | const MODEL_PATH = 'http://127.0.0.1:8081';
 5 | let transferRecognizer; //定义迁移学习器（迁移学习语音识别器）
 6 | 
 7 | window.onload = async() => {
 8 |     const recognizer = speechCommands.create(
 9 |         'BROWSER_FFT',
10 |         null,
11 |         MODEL_PATH + '/speech/model.json',
12 |         MODEL_PATH + '/speech/metadata.json'
13 |     );
14 |     await recognizer.ensureModelLoaded();
15 |     transferRecognizer = recognizer.createTransfer('轮播图');
16 | }
17 | 
18 | // 收集声音
19 | window.collect = async(btn) => {
20 |     btn.disabled = true;
21 |     const label = btn.innerText;
22 |     await transferRecognizer.collectExample(label === '背景噪音' ? '_background_noise_' : label)
23 |     btn.disabled = false;
24 |     document.querySelector('$count').innerHTML = JSON.stringify(transferRecognizer.countExamples(), null, 2);
25 | }
26 | 
27 | // 训练模型
28 | window.train = async() => {
29 |     await transferRecognizer.train({
30 |         epochs: 30,
31 |         callback: tfvis.show.fitCallbacks({
32 |             name: '训练效果'
33 |         }, ['loss', 'acc'], {
34 |             callbacks: ['onEpochEnd']
35 |         })
36 |     });
37 | }
38 | 
39 | window.toggle = async(checked) => {
40 |     if (checked) {
41 |         await transferRecognizer.listen(result => {
42 |             const {
43 |                 scores
44 |             } = result;
45 |             const labels = transferRecognizer.wordLabels();
46 |             const index = scores.indexOf(Math.max(...scores));
47 |             console.log(labels[index]);
48 |         }, {
49 |             overlapFactor: 0, //值越大识别越频繁
50 |             probabilityThreshold: 0.75 //准确度
51 |         })
52 |     } else {
53 |         transferRecognizer.stopListening();
54 |     }
55 | }
56 | 
57 | // 存储采集到的语音数据，避免每次都要录入
58 | window.save = () => {
59 |     const arrayBuffer = transferRecognizer.serializeExamples()
60 |     const blob = new Blob([arrayBuffer])
61 |     const link = document.createElement('a')
62 |     link.href = window.URL.createObjectURL(blob)
63 |     link.download = 'data.bin' //定义下载名称
64 |     link.click()
65 | }


--------------------------------------------------------------------------------
/project/9 迁移学习-商标识别/script.js:
--------------------------------------------------------------------------------
 1 | import * as tf from '@tensorflow/tfjs';
 2 | import * as tfvis from '@tensorflow/tfjs-vis';
 3 | import {
 4 |     getInputs
 5 | } from './data';
 6 | import {
 7 |     img2x,
 8 |     file2img
 9 | } from './utils';
10 | 
11 | const MOBILENET_MODEL_PATH = 'http://127.0.0.1:8081/mobilenet/web_model/model.json';
12 | const NUM_CLASSES = 3;
13 | const BRAND_CLASSES = ['android', 'apple', 'windows'];
14 | 
15 | window.onload = async() => {
16 |     const {
17 |         inputs,
18 |         labels
19 |     } = await getInputs();
20 | 
21 |     // 可视化商标
22 |     const surface = tfvis.visor().surface({
23 |         name: '输入示例',
24 |         styles: {
25 |             height: 250
26 |         }
27 |     });
28 |     inputs.forEach(img => {
29 |         surface.drawArea.appendChild(img);
30 |     });
31 | 
32 |     // 加载mobilenet模型
33 |     const mobilenet = await tf.loadLayersModel(MOBILENET_MODEL_PATH)
34 |     mobilenet.summary(); //查看模型概况
35 |     // 开始截断模型
36 |     const layer = mobilenet.getLayer('conv_pw_13_relu');
37 |     const truncatedMobilenet = tf.model({
38 |         inputs: mobilenet.inputs,
39 |         outputs: layer.output
40 |     })
41 | 
42 |     // 添加网络
43 |     const model = tf.sequential()
44 |     model.add(tf.layers.flatten({
45 |         inputShape: layer.outputShape.slice(1) //[null,7,7,256]
46 |     }));
47 |     model.add(tf.layers.dense({
48 |         units: 10,
49 |         activation: 'relu'
50 |     }));
51 |     model.add(tf.layers.dense({
52 |         units: NUM_CLASSES,
53 |         activation: 'softmax'
54 |     }));
55 |     model.compile({
56 |         loss: 'categoricalCrossentropy',
57 |         optimizer: tf.train.adam()
58 |     })
59 | 
60 |     // 将训练数据输入到截断模型中
61 |     const {
62 |         xs,
63 |         ys
64 |     } = tf.tidy(() => {
65 |         const xs = tf.concat(inputs.map(imgEl => truncatedMobilenet.predict(img2x(imgEl)))) //xs为truncatedMobilenet网络吐出的数据
66 |         const ys = tf.tensor(labels)
67 |         return {
68 |             xs,
69 |             ys
70 |         }
71 |     })
72 |     await model.fit(xs, ys, {
73 |         epochs: 20,
74 |         callbacks: tfvis.show.fitCallbacks({
75 |             name: '训练效果'
76 |         }, ['loss'], {
77 |             callbacks: ['onEpochEnd']
78 |         })
79 |     });
80 | 
81 |     // 预测
82 |     window.predict = async(file) => {
83 |         const img = await file2img(file);
84 |         document.body.appendChild(img);
85 |         const pred = tf.tidy(() => {
86 |             const x = img2x(img);
87 |             const input = truncatedMobilenet.predict(x);
88 |             return model.predict(input);
89 |         })
90 |         const index = pred.argMax(1).dataSync()[0];
91 |         document.getElementById('output').innerHTML = `预测结果：${BRAND_CLASSES[index]}`
92 |     }
93 |     window.download = async() => {
94 |         await model.save('downloads://model'); //建议看官方文档model.save
95 |     };
96 | }


--------------------------------------------------------------------------------
/project/7 手写数字识别/script.js:
--------------------------------------------------------------------------------
  1 | import * as tf from '@tensorflow/tfjs';
  2 | import * as tfvis from '@tensorflow/tfjs-vis';
  3 | import { MnistData } from './data';
  4 | 
  5 | window.onload = async () => {
  6 |   const data = new MnistData();
  7 |   await data.load();
  8 |   const examples = data.nextTestBatch(20)
  9 |   const surface = tfvis.visor().surface({
 10 |     name: '输入示例'
 11 |   });
 12 | 
 13 |   // 展示样本中的手写数据
 14 |   for (let i = 0; i < 20; i++) {
 15 |     // 清除tensor内存，防止内存泄漏
 16 |     const imageTensor = tf.tidy(() => {
 17 |       return examples.xs
 18 |         .slice([i, 0], [1, 784]) //这里详细查看文档,tf.slice()方法是使用
 19 |         .reshape([28, 28, 1]);
 20 |     })
 21 | 
 22 |     // 在浏览器中展示
 23 |     const canvas = document.createElement('canvas');
 24 |     canvas.width = 28;
 25 |     canvas.height = 28;
 26 |     canvas.style = 'margin: 4px';
 27 |     await tf.browser.toPixels(imageTensor, canvas);
 28 | 
 29 |     surface.drawArea.appendChild(canvas)
 30 |   }
 31 | 
 32 |   // 定义模型
 33 |   // 1. 卷积(5*5*8)
 34 |   // 2. 池化
 35 |   // 3. 卷积(5*5*16)
 36 |   // 4. 池化
 37 |   // 5. flatten
 38 |   // 6. softmax
 39 |   const model = tf.sequential()
 40 |   model.add(tf.layers.conv2d({
 41 |     inputShape: [28, 28, 1],
 42 |     kernelSize: 5,
 43 |     filters: 8,
 44 |     strides: 1,
 45 |     activation: 'relu',
 46 |     kernelInitializer: 'varianceScaling'
 47 |   }))
 48 |   model.add(tf.layers.maxPool2d({
 49 |     poolSize: [2, 2],
 50 |     strides: [2, 2]
 51 |   }))
 52 |   model.add(tf.layers.conv2d({
 53 |     kernelSize: 5,
 54 |     filters: 16,
 55 |     strides: 1,
 56 |     activation: 'relu',
 57 |     kernelInitializer: 'varianceScaling'
 58 |   }))
 59 |   model.add(tf.layers.maxPool2d({
 60 |     poolSize: [2, 2],
 61 |     strides: [2, 2]
 62 |   }))
 63 |   model.add(tf.layers.flatten())
 64 |   model.add(tf.layers.dense({
 65 |     units: 10,
 66 |     activation: 'softmax',
 67 |     kernelInitializer: 'varianceScaling'
 68 |   }))
 69 |   model.compile({
 70 |     loss: 'categoricalCrossentropy',
 71 |     optimizer: tf.train.adam(),
 72 |     metrics: ['accuracy']
 73 |   })
 74 | 
 75 |   const [trainX, trainY] = tf.tidy(() => {
 76 |     const d = data.nextTrainBatch(1000)
 77 |     return [
 78 |       d.xs.reshape([1000, 28, 28, 1]),
 79 |       d.labels
 80 |     ]
 81 |   })
 82 |   const [testX, testY] = tf.tidy(() => {
 83 |     const d = data.nextTestBatch(200)
 84 |     return [
 85 |       d.xs.reshape([200, 28, 28, 1]),
 86 |       d.labels
 87 |     ]
 88 |   })
 89 | 
 90 |   // 训练模型
 91 |   await model.fit(trainX, trainY, {
 92 |     validationData: [testX, testY],
 93 |     batchSize: 500,
 94 |     epochs: 50,
 95 |     callbacks: tfvis.show.fitCallbacks({
 96 |       name: '训练效果'
 97 |     }, ['loss', 'val_loss', 'acc', 'val_acc'], {
 98 |       callbacks: ['onEpochEnd']
 99 |     })
100 |   })
101 | 
102 |   // 画布支持绘制，监听mousemove方法
103 |   const canvas = document.querySelector('canvas')
104 |   canvas.addEventListener('mousemove', e => {
105 |     if (e.buttons === 1) {
106 |       const ctx = canvas.getContext('2d')
107 |       ctx.fillStyle = 'rgb(255,255,255)';
108 |       ctx.fillRect(e.offsetX, e.offsetY, 10, 10);
109 |     }
110 |   })
111 |   // 清楚画布
112 |   window.clear = () => {
113 |     const ctx = canvas.getContext('2d');
114 |     ctx.fillStyle = 'rgb(0,0,0)';
115 |     ctx.fillRect(0, 0, 300, 300);
116 |   }
117 |   clear()
118 | 
119 |   // 预测
120 |   window.predict = () => {
121 |     const input = tf.tidy(() => {
122 |       return tf.image.resizeBilinear( //tf.image.resizeBilinear改变图片尺寸 (300,300)->(28,28)
123 |         tf.browser.fromPixels(canvas), [28, 28], //tf.browser.fromPixels是将canvas/image转化为tensor
124 |         true
125 |       ).slice([0, 0, 0], [28, 28, 1]) //将彩色通道转化为黑白通道，将rgb通道切掉
126 |         .toFloat()
127 |         .div(255) //归一化(0,1)
128 |         .reshape([1, 28, 28, 1]);
129 |     });
130 |     const pred = model.predict(input).argMax(1);
131 |     document.getElementById('output').innerHTML = `预测结果为 ${pred.dataSync()[0]}`
132 |   };
133 | }


--------------------------------------------------------------------------------
/project/7 手写数字识别/data.js:
--------------------------------------------------------------------------------
  1 | import * as tf from '@tensorflow/tfjs';
  2 | 
  3 | const IMAGE_SIZE = 784;
  4 | const NUM_CLASSES = 10;
  5 | const NUM_DATASET_ELEMENTS = 65000;
  6 | const TRAIN_TEST_RATIO = 5 / 6;
  7 | const NUM_TRAIN_ELEMENTS = Math.floor(TRAIN_TEST_RATIO * NUM_DATASET_ELEMENTS);
  8 | const NUM_TEST_ELEMENTS = NUM_DATASET_ELEMENTS - NUM_TRAIN_ELEMENTS;
  9 | const MNIST_IMAGES_SPRITE_PATH = 'http://127.0.0.1:8081/mnist/mnist_images.png';
 10 | const MNIST_LABELS_PATH = 'http://127.0.0.1:8081/mnist/mnist_labels_uint8';
 11 | export class MnistData {
 12 |   constructor() {
 13 |     this.shuffledTrainIndex = 0;
 14 |     this.shuffledTestIndex = 0;
 15 |   }
 16 | 
 17 |   async load() {
 18 |     const img = new Image();
 19 |     const canvas = document.createElement('canvas');
 20 |     const ctx = canvas.getContext('2d');
 21 |     const imgRequest = new Promise((resolve, reject) => {
 22 |       img.crossOrigin = '';
 23 |       img.onload = () => {
 24 |         img.width = img.naturalWidth;
 25 |         img.height = img.naturalHeight;
 26 | 
 27 |         const datasetBytesBuffer =
 28 |           new ArrayBuffer(NUM_DATASET_ELEMENTS * IMAGE_SIZE * 4);
 29 | 
 30 |         const chunkSize = 5000;
 31 |         canvas.width = img.width;
 32 |         canvas.height = chunkSize;
 33 | 
 34 |         for (let i = 0; i < NUM_DATASET_ELEMENTS / chunkSize; i++) {
 35 |           const datasetBytesView = new Float32Array(
 36 |             datasetBytesBuffer, i * IMAGE_SIZE * chunkSize * 4,
 37 |             IMAGE_SIZE * chunkSize);
 38 |           ctx.drawImage(
 39 |             img, 0, i * chunkSize, img.width, chunkSize, 0, 0, img.width,
 40 |             chunkSize);
 41 | 
 42 |           const imageData = ctx.getImageData(0, 0, canvas.width, canvas.height);
 43 |           for (let j = 0; j < imageData.data.length / 4; j++) {
 44 |             datasetBytesView[j] = imageData.data[j * 4] / 255;
 45 |           }
 46 |         }
 47 |         this.datasetImages = new Float32Array(datasetBytesBuffer);
 48 | 
 49 |         resolve();
 50 |       };
 51 |       img.src = MNIST_IMAGES_SPRITE_PATH;
 52 |     });
 53 | 
 54 |     const labelsRequest = fetch(MNIST_LABELS_PATH);
 55 |     const [imgResponse, labelsResponse] =
 56 |       await Promise.all([imgRequest, labelsRequest]);
 57 | 
 58 |     this.datasetLabels = new Uint8Array(await labelsResponse.arrayBuffer());
 59 |     this.trainIndices = tf.util.createShuffledIndices(NUM_TRAIN_ELEMENTS);
 60 |     this.testIndices = tf.util.createShuffledIndices(NUM_TEST_ELEMENTS);
 61 |     this.trainImages = this.datasetImages.slice(0, IMAGE_SIZE * NUM_TRAIN_ELEMENTS);
 62 |     this.testImages = this.datasetImages.slice(IMAGE_SIZE * NUM_TRAIN_ELEMENTS);
 63 |     this.trainLabels = this.datasetLabels.slice(0, NUM_CLASSES * NUM_TRAIN_ELEMENTS);
 64 |     this.testLabels = this.datasetLabels.slice(NUM_CLASSES * NUM_TRAIN_ELEMENTS);
 65 |   }
 66 | 
 67 |   nextTrainBatch(batchSize) {
 68 |     return this.nextBatch(
 69 |       batchSize, [this.trainImages, this.trainLabels], () => {
 70 |         this.shuffledTrainIndex =
 71 |           (this.shuffledTrainIndex + 1) % this.trainIndices.length;
 72 |         return this.trainIndices[this.shuffledTrainIndex];
 73 |       });
 74 |   }
 75 | 
 76 |   nextTestBatch(batchSize) {
 77 |     return this.nextBatch(batchSize, [this.testImages, this.testLabels], () => {
 78 |       this.shuffledTestIndex =
 79 |         (this.shuffledTestIndex + 1) % this.testIndices.length;
 80 |       return this.testIndices[this.shuffledTestIndex];
 81 |     });
 82 |   }
 83 | 
 84 |   nextBatch(batchSize, data, index) {
 85 |     const batchImagesArray = new Float32Array(batchSize * IMAGE_SIZE);
 86 |     const batchLabelsArray = new Uint8Array(batchSize * NUM_CLASSES);
 87 | 
 88 |     for (let i = 0; i < batchSize; i++) {
 89 |       const idx = index();
 90 | 
 91 |       const image =
 92 |         data[0].slice(idx * IMAGE_SIZE, idx * IMAGE_SIZE + IMAGE_SIZE);
 93 |       batchImagesArray.set(image, i * IMAGE_SIZE);
 94 | 
 95 |       const label =
 96 |         data[1].slice(idx * NUM_CLASSES, idx * NUM_CLASSES + NUM_CLASSES);
 97 |       batchLabelsArray.set(label, i * NUM_CLASSES);
 98 |     }
 99 | 
100 |     const xs = tf.tensor2d(batchImagesArray, [batchSize, IMAGE_SIZE]);
101 |     const labels = tf.tensor2d(batchLabelsArray, [batchSize, NUM_CLASSES]);
102 | 
103 |     return {
104 |       xs,
105 |       labels
106 |     };
107 |   }
108 | }


--------------------------------------------------------------------------------
/project/5 多分类/data.js:
--------------------------------------------------------------------------------
  1 | import * as tf from '@tensorflow/tfjs';
  2 | 
  3 | export const IRIS_CLASSES = ['山鸢尾', '变色鸢尾', '维吉尼亚鸢尾'];
  4 | export const IRIS_NUM_CLASSES = IRIS_CLASSES.length;
  5 | 
  6 | // 鸢尾花数据集 https://archive.ics.uci.edu/ml/machine-learning-databases/iris/iris.data
  7 | const IRIS_DATA = [
  8 |   [5.1, 3.5, 1.4, 0.2, 0],
  9 |   [4.9, 3.0, 1.4, 0.2, 0],
 10 |   [4.7, 3.2, 1.3, 0.2, 0],
 11 |   [4.6, 3.1, 1.5, 0.2, 0],
 12 |   [5.0, 3.6, 1.4, 0.2, 0],
 13 |   [5.4, 3.9, 1.7, 0.4, 0],
 14 |   [4.6, 3.4, 1.4, 0.3, 0],
 15 |   [5.0, 3.4, 1.5, 0.2, 0],
 16 |   [4.4, 2.9, 1.4, 0.2, 0],
 17 |   [4.9, 3.1, 1.5, 0.1, 0],
 18 |   [5.4, 3.7, 1.5, 0.2, 0],
 19 |   [4.8, 3.4, 1.6, 0.2, 0],
 20 |   [4.8, 3.0, 1.4, 0.1, 0],
 21 |   [4.3, 3.0, 1.1, 0.1, 0],
 22 |   [5.8, 4.0, 1.2, 0.2, 0],
 23 |   [5.7, 4.4, 1.5, 0.4, 0],
 24 |   [5.4, 3.9, 1.3, 0.4, 0],
 25 |   [5.1, 3.5, 1.4, 0.3, 0],
 26 |   [5.7, 3.8, 1.7, 0.3, 0],
 27 |   [5.1, 3.8, 1.5, 0.3, 0],
 28 |   [5.4, 3.4, 1.7, 0.2, 0],
 29 |   [5.1, 3.7, 1.5, 0.4, 0],
 30 |   [4.6, 3.6, 1.0, 0.2, 0],
 31 |   [5.1, 3.3, 1.7, 0.5, 0],
 32 |   [4.8, 3.4, 1.9, 0.2, 0],
 33 |   [5.0, 3.0, 1.6, 0.2, 0],
 34 |   [5.0, 3.4, 1.6, 0.4, 0],
 35 |   [5.2, 3.5, 1.5, 0.2, 0],
 36 |   [5.2, 3.4, 1.4, 0.2, 0],
 37 |   [4.7, 3.2, 1.6, 0.2, 0],
 38 |   [4.8, 3.1, 1.6, 0.2, 0],
 39 |   [5.4, 3.4, 1.5, 0.4, 0],
 40 |   [5.2, 4.1, 1.5, 0.1, 0],
 41 |   [5.5, 4.2, 1.4, 0.2, 0],
 42 |   [4.9, 3.1, 1.5, 0.1, 0],
 43 |   [5.0, 3.2, 1.2, 0.2, 0],
 44 |   [5.5, 3.5, 1.3, 0.2, 0],
 45 |   [4.9, 3.1, 1.5, 0.1, 0],
 46 |   [4.4, 3.0, 1.3, 0.2, 0],
 47 |   [5.1, 3.4, 1.5, 0.2, 0],
 48 |   [5.0, 3.5, 1.3, 0.3, 0],
 49 |   [4.5, 2.3, 1.3, 0.3, 0],
 50 |   [4.4, 3.2, 1.3, 0.2, 0],
 51 |   [5.0, 3.5, 1.6, 0.6, 0],
 52 |   [5.1, 3.8, 1.9, 0.4, 0],
 53 |   [4.8, 3.0, 1.4, 0.3, 0],
 54 |   [5.1, 3.8, 1.6, 0.2, 0],
 55 |   [4.6, 3.2, 1.4, 0.2, 0],
 56 |   [5.3, 3.7, 1.5, 0.2, 0],
 57 |   [5.0, 3.3, 1.4, 0.2, 0],
 58 |   [7.0, 3.2, 4.7, 1.4, 1],
 59 |   [6.4, 3.2, 4.5, 1.5, 1],
 60 |   [6.9, 3.1, 4.9, 1.5, 1],
 61 |   [5.5, 2.3, 4.0, 1.3, 1],
 62 |   [6.5, 2.8, 4.6, 1.5, 1],
 63 |   [5.7, 2.8, 4.5, 1.3, 1],
 64 |   [6.3, 3.3, 4.7, 1.6, 1],
 65 |   [4.9, 2.4, 3.3, 1.0, 1],
 66 |   [6.6, 2.9, 4.6, 1.3, 1],
 67 |   [5.2, 2.7, 3.9, 1.4, 1],
 68 |   [5.0, 2.0, 3.5, 1.0, 1],
 69 |   [5.9, 3.0, 4.2, 1.5, 1],
 70 |   [6.0, 2.2, 4.0, 1.0, 1],
 71 |   [6.1, 2.9, 4.7, 1.4, 1],
 72 |   [5.6, 2.9, 3.6, 1.3, 1],
 73 |   [6.7, 3.1, 4.4, 1.4, 1],
 74 |   [5.6, 3.0, 4.5, 1.5, 1],
 75 |   [5.8, 2.7, 4.1, 1.0, 1],
 76 |   [6.2, 2.2, 4.5, 1.5, 1],
 77 |   [5.6, 2.5, 3.9, 1.1, 1],
 78 |   [5.9, 3.2, 4.8, 1.8, 1],
 79 |   [6.1, 2.8, 4.0, 1.3, 1],
 80 |   [6.3, 2.5, 4.9, 1.5, 1],
 81 |   [6.1, 2.8, 4.7, 1.2, 1],
 82 |   [6.4, 2.9, 4.3, 1.3, 1],
 83 |   [6.6, 3.0, 4.4, 1.4, 1],
 84 |   [6.8, 2.8, 4.8, 1.4, 1],
 85 |   [6.7, 3.0, 5.0, 1.7, 1],
 86 |   [6.0, 2.9, 4.5, 1.5, 1],
 87 |   [5.7, 2.6, 3.5, 1.0, 1],
 88 |   [5.5, 2.4, 3.8, 1.1, 1],
 89 |   [5.5, 2.4, 3.7, 1.0, 1],
 90 |   [5.8, 2.7, 3.9, 1.2, 1],
 91 |   [6.0, 2.7, 5.1, 1.6, 1],
 92 |   [5.4, 3.0, 4.5, 1.5, 1],
 93 |   [6.0, 3.4, 4.5, 1.6, 1],
 94 |   [6.7, 3.1, 4.7, 1.5, 1],
 95 |   [6.3, 2.3, 4.4, 1.3, 1],
 96 |   [5.6, 3.0, 4.1, 1.3, 1],
 97 |   [5.5, 2.5, 4.0, 1.3, 1],
 98 |   [5.5, 2.6, 4.4, 1.2, 1],
 99 |   [6.1, 3.0, 4.6, 1.4, 1],
100 |   [5.8, 2.6, 4.0, 1.2, 1],
101 |   [5.0, 2.3, 3.3, 1.0, 1],
102 |   [5.6, 2.7, 4.2, 1.3, 1],
103 |   [5.7, 3.0, 4.2, 1.2, 1],
104 |   [5.7, 2.9, 4.2, 1.3, 1],
105 |   [6.2, 2.9, 4.3, 1.3, 1],
106 |   [5.1, 2.5, 3.0, 1.1, 1],
107 |   [5.7, 2.8, 4.1, 1.3, 1],
108 |   [6.3, 3.3, 6.0, 2.5, 2],
109 |   [5.8, 2.7, 5.1, 1.9, 2],
110 |   [7.1, 3.0, 5.9, 2.1, 2],
111 |   [6.3, 2.9, 5.6, 1.8, 2],
112 |   [6.5, 3.0, 5.8, 2.2, 2],
113 |   [7.6, 3.0, 6.6, 2.1, 2],
114 |   [4.9, 2.5, 4.5, 1.7, 2],
115 |   [7.3, 2.9, 6.3, 1.8, 2],
116 |   [6.7, 2.5, 5.8, 1.8, 2],
117 |   [7.2, 3.6, 6.1, 2.5, 2],
118 |   [6.5, 3.2, 5.1, 2.0, 2],
119 |   [6.4, 2.7, 5.3, 1.9, 2],
120 |   [6.8, 3.0, 5.5, 2.1, 2],
121 |   [5.7, 2.5, 5.0, 2.0, 2],
122 |   [5.8, 2.8, 5.1, 2.4, 2],
123 |   [6.4, 3.2, 5.3, 2.3, 2],
124 |   [6.5, 3.0, 5.5, 1.8, 2],
125 |   [7.7, 3.8, 6.7, 2.2, 2],
126 |   [7.7, 2.6, 6.9, 2.3, 2],
127 |   [6.0, 2.2, 5.0, 1.5, 2],
128 |   [6.9, 3.2, 5.7, 2.3, 2],
129 |   [5.6, 2.8, 4.9, 2.0, 2],
130 |   [7.7, 2.8, 6.7, 2.0, 2],
131 |   [6.3, 2.7, 4.9, 1.8, 2],
132 |   [6.7, 3.3, 5.7, 2.1, 2],
133 |   [7.2, 3.2, 6.0, 1.8, 2],
134 |   [6.2, 2.8, 4.8, 1.8, 2],
135 |   [6.1, 3.0, 4.9, 1.8, 2],
136 |   [6.4, 2.8, 5.6, 2.1, 2],
137 |   [7.2, 3.0, 5.8, 1.6, 2],
138 |   [7.4, 2.8, 6.1, 1.9, 2],
139 |   [7.9, 3.8, 6.4, 2.0, 2],
140 |   [6.4, 2.8, 5.6, 2.2, 2],
141 |   [6.3, 2.8, 5.1, 1.5, 2],
142 |   [6.1, 2.6, 5.6, 1.4, 2],
143 |   [7.7, 3.0, 6.1, 2.3, 2],
144 |   [6.3, 3.4, 5.6, 2.4, 2],
145 |   [6.4, 3.1, 5.5, 1.8, 2],
146 |   [6.0, 3.0, 4.8, 1.8, 2],
147 |   [6.9, 3.1, 5.4, 2.1, 2],
148 |   [6.7, 3.1, 5.6, 2.4, 2],
149 |   [6.9, 3.1, 5.1, 2.3, 2],
150 |   [5.8, 2.7, 5.1, 1.9, 2],
151 |   [6.8, 3.2, 5.9, 2.3, 2],
152 |   [6.7, 3.3, 5.7, 2.5, 2],
153 |   [6.7, 3.0, 5.2, 2.3, 2],
154 |   [6.3, 2.5, 5.0, 1.9, 2],
155 |   [6.5, 3.0, 5.2, 2.0, 2],
156 |   [6.2, 3.4, 5.4, 2.3, 2],
157 |   [5.9, 3.0, 5.1, 1.8, 2],
158 | ];
159 | 
160 | // 转化为tensor
161 | function convertToTensors(data, targets, testSplit) {
162 |   const numExamples = data.length;
163 |   if (numExamples !== targets.length) {
164 |     throw new Error('data and split have different numbers of examples');
165 |   }
166 | 
167 |   const indices = [];
168 |   for (let i = 0; i < numExamples; ++i) {
169 |     indices.push(i);
170 |   }
171 |   tf.util.shuffle(indices);
172 | 
173 |   const shuffledData = [];
174 |   const shuffledTargets = [];
175 |   for (let i = 0; i < numExamples; ++i) {
176 |     shuffledData.push(data[indices[i]]);
177 |     shuffledTargets.push(targets[indices[i]]);
178 |   }
179 | 
180 |   const numTestExamples = Math.round(numExamples * testSplit);
181 |   const numTrainExamples = numExamples - numTestExamples;
182 |   const xDims = shuffledData[0].length;
183 |   const xs = tf.tensor2d(shuffledData, [numExamples, xDims]);
184 |   const ys = tf.oneHot(tf.tensor1d(shuffledTargets).toInt(), IRIS_NUM_CLASSES);
185 |   const xTrain = xs.slice([0, 0], [numTrainExamples, xDims]);
186 |   const xTest = xs.slice([numTrainExamples, 0], [numTestExamples, xDims]);
187 |   const yTrain = ys.slice([0, 0], [numTrainExamples, IRIS_NUM_CLASSES]);
188 |   const yTest = ys.slice([0, 0], [numTestExamples, IRIS_NUM_CLASSES]);
189 |   return [xTrain, yTrain, xTest, yTest];
190 | }
191 | 
192 | // 获取鸢尾花数据集（带分类）
193 | export function getIrisData(testSplit) {
194 |   return tf.tidy(() => {
195 |     const dataByClass = [];
196 |     const targetsByClass = [];
197 |     for (let i = 0; i < IRIS_CLASSES.length; ++i) {
198 |       dataByClass.push([]);
199 |       targetsByClass.push([]);
200 |     }
201 |     for (const example of IRIS_DATA) {
202 |       const target = example[example.length - 1];
203 |       const data = example.slice(0, example.length - 1);
204 |       dataByClass[target].push(data);
205 |       targetsByClass[target].push(target);
206 |     }
207 | 
208 |     const xTrains = [];
209 |     const yTrains = [];
210 |     const xTests = [];
211 |     const yTests = [];
212 |     for (let i = 0; i < IRIS_CLASSES.length; ++i) {
213 |       const [xTrain, yTrain, xTest, yTest] =
214 |         convertToTensors(dataByClass[i], targetsByClass[i], testSplit);
215 |       xTrains.push(xTrain);
216 |       yTrains.push(yTrain);
217 |       xTests.push(xTest);
218 |       yTests.push(yTest);
219 |     }
220 | 
221 |     const concatAxis = 0;
222 |     return [
223 |       tf.concat(xTrains, concatAxis), tf.concat(yTrains, concatAxis),
224 |       tf.concat(xTests, concatAxis), tf.concat(yTests, concatAxis)
225 |     ];
226 |   });
227 | }


--------------------------------------------------------------------------------
/project/8 预训练模型-图片分类/imagenet_classes.js:
--------------------------------------------------------------------------------
   1 | /**
   2 |  * @license
   3 |  * Copyright 2017 Google LLC. All Rights Reserved.
   4 |  * Licensed under the Apache License, Version 2.0 (the "License");
   5 |  * you may not use this file except in compliance with the License.
   6 |  * You may obtain a copy of the License at
   7 |  *
   8 |  * http://www.apache.org/licenses/LICENSE-2.0
   9 |  *
  10 |  * Unless required by applicable law or agreed to in writing, software
  11 |  * distributed under the License is distributed on an "AS IS" BASIS,
  12 |  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13 |  * See the License for the specific language governing permissions and
  14 |  * limitations under the License.
  15 |  * =============================================================================
  16 |  */
  17 | 
  18 | export const IMAGENET_CLASSES = {
  19 |     0: 'tench, Tinca tinca',
  20 |     1: 'goldfish, Carassius auratus',
  21 |     2: 'great white shark, white shark, man-eater, man-eating shark, ' +
  22 |         'Carcharodon carcharias',
  23 |     3: 'tiger shark, Galeocerdo cuvieri',
  24 |     4: 'hammerhead, hammerhead shark',
  25 |     5: 'electric ray, crampfish, numbfish, torpedo',
  26 |     6: 'stingray',
  27 |     7: 'cock',
  28 |     8: 'hen',
  29 |     9: 'ostrich, Struthio camelus',
  30 |     10: 'brambling, Fringilla montifringilla',
  31 |     11: 'goldfinch, Carduelis carduelis',
  32 |     12: 'house finch, linnet, Carpodacus mexicanus',
  33 |     13: 'junco, snowbird',
  34 |     14: 'indigo bunting, indigo finch, indigo bird, Passerina cyanea',
  35 |     15: 'robin, American robin, Turdus migratorius',
  36 |     16: 'bulbul',
  37 |     17: 'jay',
  38 |     18: 'magpie',
  39 |     19: 'chickadee',
  40 |     20: 'water ouzel, dipper',
  41 |     21: 'kite',
  42 |     22: 'bald eagle, American eagle, Haliaeetus leucocephalus',
  43 |     23: 'vulture',
  44 |     24: 'great grey owl, great gray owl, Strix nebulosa',
  45 |     25: 'European fire salamander, Salamandra salamandra',
  46 |     26: 'common newt, Triturus vulgaris',
  47 |     27: 'eft',
  48 |     28: 'spotted salamander, Ambystoma maculatum',
  49 |     29: 'axolotl, mud puppy, Ambystoma mexicanum',
  50 |     30: 'bullfrog, Rana catesbeiana',
  51 |     31: 'tree frog, tree-frog',
  52 |     32: 'tailed frog, bell toad, ribbed toad, tailed toad, Ascaphus trui',
  53 |     33: 'loggerhead, loggerhead turtle, Caretta caretta',
  54 |     34: 'leatherback turtle, leatherback, leathery turtle, Dermochelys coriacea',
  55 |     35: 'mud turtle',
  56 |     36: 'terrapin',
  57 |     37: 'box turtle, box tortoise',
  58 |     38: 'banded gecko',
  59 |     39: 'common iguana, iguana, Iguana iguana',
  60 |     40: 'American chameleon, anole, Anolis carolinensis',
  61 |     41: 'whiptail, whiptail lizard',
  62 |     42: 'agama',
  63 |     43: 'frilled lizard, Chlamydosaurus kingi',
  64 |     44: 'alligator lizard',
  65 |     45: 'Gila monster, Heloderma suspectum',
  66 |     46: 'green lizard, Lacerta viridis',
  67 |     47: 'African chameleon, Chamaeleo chamaeleon',
  68 |     48: 'Komodo dragon, Komodo lizard, dragon lizard, giant lizard, ' +
  69 |         'Varanus komodoensis',
  70 |     49: 'African crocodile, Nile crocodile, Crocodylus niloticus',
  71 |     50: 'American alligator, Alligator mississipiensis',
  72 |     51: 'triceratops',
  73 |     52: 'thunder snake, worm snake, Carphophis amoenus',
  74 |     53: 'ringneck snake, ring-necked snake, ring snake',
  75 |     54: 'hognose snake, puff adder, sand viper',
  76 |     55: 'green snake, grass snake',
  77 |     56: 'king snake, kingsnake',
  78 |     57: 'garter snake, grass snake',
  79 |     58: 'water snake',
  80 |     59: 'vine snake',
  81 |     60: 'night snake, Hypsiglena torquata',
  82 |     61: 'boa constrictor, Constrictor constrictor',
  83 |     62: 'rock python, rock snake, Python sebae',
  84 |     63: 'Indian cobra, Naja naja',
  85 |     64: 'green mamba',
  86 |     65: 'sea snake',
  87 |     66: 'horned viper, cerastes, sand viper, horned asp, Cerastes cornutus',
  88 |     67: 'diamondback, diamondback rattlesnake, Crotalus adamanteus',
  89 |     68: 'sidewinder, horned rattlesnake, Crotalus cerastes',
  90 |     69: 'trilobite',
  91 |     70: 'harvestman, daddy longlegs, Phalangium opilio',
  92 |     71: 'scorpion',
  93 |     72: 'black and gold garden spider, Argiope aurantia',
  94 |     73: 'barn spider, Araneus cavaticus',
  95 |     74: 'garden spider, Aranea diademata',
  96 |     75: 'black widow, Latrodectus mactans',
  97 |     76: 'tarantula',
  98 |     77: 'wolf spider, hunting spider',
  99 |     78: 'tick',
 100 |     79: 'centipede',
 101 |     80: 'black grouse',
 102 |     81: 'ptarmigan',
 103 |     82: 'ruffed grouse, partridge, Bonasa umbellus',
 104 |     83: 'prairie chicken, prairie grouse, prairie fowl',
 105 |     84: 'peacock',
 106 |     85: 'quail',
 107 |     86: 'partridge',
 108 |     87: 'African grey, African gray, Psittacus erithacus',
 109 |     88: 'macaw',
 110 |     89: 'sulphur-crested cockatoo, Kakatoe galerita, Cacatua galerita',
 111 |     90: 'lorikeet',
 112 |     91: 'coucal',
 113 |     92: 'bee eater',
 114 |     93: 'hornbill',
 115 |     94: 'hummingbird',
 116 |     95: 'jacamar',
 117 |     96: 'toucan',
 118 |     97: 'drake',
 119 |     98: 'red-breasted merganser, Mergus serrator',
 120 |     99: 'goose',
 121 |     100: 'black swan, Cygnus atratus',
 122 |     101: 'tusker',
 123 |     102: 'echidna, spiny anteater, anteater',
 124 |     103: 'platypus, duckbill, duckbilled platypus, duck-billed platypus, ' +
 125 |         'Ornithorhynchus anatinus',
 126 |     104: 'wallaby, brush kangaroo',
 127 |     105: 'koala, koala bear, kangaroo bear, native bear, Phascolarctos cinereus',
 128 |     106: 'wombat',
 129 |     107: 'jelly fish',
 130 |     108: 'sea anemone, anemone',
 131 |     109: 'brain coral',
 132 |     110: 'flatworm, platyhelminth',
 133 |     111: 'nematode, nematode worm, roundworm',
 134 |     112: 'conch',
 135 |     113: 'snail',
 136 |     114: 'slug',
 137 |     115: 'sea slug, nudibranch',
 138 |     116: 'chiton, coat-of-mail shell, sea cradle, polyplacophore',
 139 |     117: 'chambered nautilus, pearly nautilus, nautilus',
 140 |     118: 'Dungeness crab, Cancer magister',
 141 |     119: 'rock crab, Cancer irroratus',
 142 |     120: 'fiddler crab',
 143 |     121: 'king crab, Alaska crab, Alaskan king crab, Alaska king crab, ' +
 144 |         'Paralithodes camtschatica',
 145 |     122: 'American lobster, Northern lobster, Maine lobster, Homarus americanus',
 146 |     123: 'spiny lobster, langouste, rock lobster, crawfish, crayfish, sea ' +
 147 |         'crawfish',
 148 |     124: 'crayfish, crawfish, crawdad, crawdaddy',
 149 |     125: 'hermit crab',
 150 |     126: 'isopod',
 151 |     127: 'white stork, Ciconia ciconia',
 152 |     128: 'black stork, Ciconia nigra',
 153 |     129: 'spoonbill',
 154 |     130: 'flamingo',
 155 |     131: 'little blue heron, Egretta caerulea',
 156 |     132: 'American egret, great white heron, Egretta albus',
 157 |     133: 'bittern',
 158 |     134: 'crane',
 159 |     135: 'limpkin, Aramus pictus',
 160 |     136: 'European gallinule, Porphyrio porphyrio',
 161 |     137: 'American coot, marsh hen, mud hen, water hen, Fulica americana',
 162 |     138: 'bustard',
 163 |     139: 'ruddy turnstone, Arenaria interpres',
 164 |     140: 'red-backed sandpiper, dunlin, Erolia alpina',
 165 |     141: 'redshank, Tringa totanus',
 166 |     142: 'dowitcher',
 167 |     143: 'oystercatcher, oyster catcher',
 168 |     144: 'pelican',
 169 |     145: 'king penguin, Aptenodytes patagonica',
 170 |     146: 'albatross, mollymawk',
 171 |     147: 'grey whale, gray whale, devilfish, Eschrichtius gibbosus, ' +
 172 |         'Eschrichtius robustus',
 173 |     148: 'killer whale, killer, orca, grampus, sea wolf, Orcinus orca',
 174 |     149: 'dugong, Dugong dugon',
 175 |     150: 'sea lion',
 176 |     151: 'Chihuahua',
 177 |     152: 'Japanese spaniel',
 178 |     153: 'Maltese dog, Maltese terrier, Maltese',
 179 |     154: 'Pekinese, Pekingese, Peke',
 180 |     155: 'Shih-Tzu',
 181 |     156: 'Blenheim spaniel',
 182 |     157: 'papillon',
 183 |     158: 'toy terrier',
 184 |     159: 'Rhodesian ridgeback',
 185 |     160: 'Afghan hound, Afghan',
 186 |     161: 'basset, basset hound',
 187 |     162: 'beagle',
 188 |     163: 'bloodhound, sleuthhound',
 189 |     164: 'bluetick',
 190 |     165: 'black-and-tan coonhound',
 191 |     166: 'Walker hound, Walker foxhound',
 192 |     167: 'English foxhound',
 193 |     168: 'redbone',
 194 |     169: 'borzoi, Russian wolfhound',
 195 |     170: 'Irish wolfhound',
 196 |     171: 'Italian greyhound',
 197 |     172: 'whippet',
 198 |     173: 'Ibizan hound, Ibizan Podenco',
 199 |     174: 'Norwegian elkhound, elkhound',
 200 |     175: 'otterhound, otter hound',
 201 |     176: 'Saluki, gazelle hound',
 202 |     177: 'Scottish deerhound, deerhound',
 203 |     178: 'Weimaraner',
 204 |     179: 'Staffordshire bullterrier, Staffordshire bull terrier',
 205 |     180: 'American Staffordshire terrier, Staffordshire terrier, American pit ' +
 206 |         'bull terrier, pit bull terrier',
 207 |     181: 'Bedlington terrier',
 208 |     182: 'Border terrier',
 209 |     183: 'Kerry blue terrier',
 210 |     184: 'Irish terrier',
 211 |     185: 'Norfolk terrier',
 212 |     186: 'Norwich terrier',
 213 |     187: 'Yorkshire terrier',
 214 |     188: 'wire-haired fox terrier',
 215 |     189: 'Lakeland terrier',
 216 |     190: 'Sealyham terrier, Sealyham',
 217 |     191: 'Airedale, Airedale terrier',
 218 |     192: 'cairn, cairn terrier',
 219 |     193: 'Australian terrier',
 220 |     194: 'Dandie Dinmont, Dandie Dinmont terrier',
 221 |     195: 'Boston bull, Boston terrier',
 222 |     196: 'miniature schnauzer',
 223 |     197: 'giant schnauzer',
 224 |     198: 'standard schnauzer',
 225 |     199: 'Scotch terrier, Scottish terrier, Scottie',
 226 |     200: 'Tibetan terrier, chrysanthemum dog',
 227 |     201: 'silky terrier, Sydney silky',
 228 |     202: 'soft-coated wheaten terrier',
 229 |     203: 'West Highland white terrier',
 230 |     204: 'Lhasa, Lhasa apso',
 231 |     205: 'flat-coated retriever',
 232 |     206: 'curly-coated retriever',
 233 |     207: 'golden retriever',
 234 |     208: 'Labrador retriever',
 235 |     209: 'Chesapeake Bay retriever',
 236 |     210: 'German short-haired pointer',
 237 |     211: 'vizsla, Hungarian pointer',
 238 |     212: 'English setter',
 239 |     213: 'Irish setter, red setter',
 240 |     214: 'Gordon setter',
 241 |     215: 'Brittany spaniel',
 242 |     216: 'clumber, clumber spaniel',
 243 |     217: 'English springer, English springer spaniel',
 244 |     218: 'Welsh springer spaniel',
 245 |     219: 'cocker spaniel, English cocker spaniel, cocker',
 246 |     220: 'Sussex spaniel',
 247 |     221: 'Irish water spaniel',
 248 |     222: 'kuvasz',
 249 |     223: 'schipperke',
 250 |     224: 'groenendael',
 251 |     225: 'malinois',
 252 |     226: 'briard',
 253 |     227: 'kelpie',
 254 |     228: 'komondor',
 255 |     229: 'Old English sheepdog, bobtail',
 256 |     230: 'Shetland sheepdog, Shetland sheep dog, Shetland',
 257 |     231: 'collie',
 258 |     232: 'Border collie',
 259 |     233: 'Bouvier des Flandres, Bouviers des Flandres',
 260 |     234: 'Rottweiler',
 261 |     235: 'German shepherd, German shepherd dog, German police dog, alsatian',
 262 |     236: 'Doberman, Doberman pinscher',
 263 |     237: 'miniature pinscher',
 264 |     238: 'Greater Swiss Mountain dog',
 265 |     239: 'Bernese mountain dog',
 266 |     240: 'Appenzeller',
 267 |     241: 'EntleBucher',
 268 |     242: 'boxer',
 269 |     243: 'bull mastiff',
 270 |     244: 'Tibetan mastiff',
 271 |     245: 'French bulldog',
 272 |     246: 'Great Dane',
 273 |     247: 'Saint Bernard, St Bernard',
 274 |     248: 'Eskimo dog, husky',
 275 |     249: 'malamute, malemute, Alaskan malamute',
 276 |     250: 'Siberian husky',
 277 |     251: 'dalmatian, coach dog, carriage dog',
 278 |     252: 'affenpinscher, monkey pinscher, monkey dog',
 279 |     253: 'basenji',
 280 |     254: 'pug, pug-dog',
 281 |     255: 'Leonberg',
 282 |     256: 'Newfoundland, Newfoundland dog',
 283 |     257: 'Great Pyrenees',
 284 |     258: 'Samoyed, Samoyede',
 285 |     259: 'Pomeranian',
 286 |     260: 'chow, chow chow',
 287 |     261: 'keeshond',
 288 |     262: 'Brabancon griffon',
 289 |     263: 'Pembroke, Pembroke Welsh corgi',
 290 |     264: 'Cardigan, Cardigan Welsh corgi',
 291 |     265: 'toy poodle',
 292 |     266: 'miniature poodle',
 293 |     267: 'standard poodle',
 294 |     268: 'Mexican hairless',
 295 |     269: 'timber wolf, grey wolf, gray wolf, Canis lupus',
 296 |     270: 'white wolf, Arctic wolf, Canis lupus tundrarum',
 297 |     271: 'red wolf, maned wolf, Canis rufus, Canis niger',
 298 |     272: 'coyote, prairie wolf, brush wolf, Canis latrans',
 299 |     273: 'dingo, warrigal, warragal, Canis dingo',
 300 |     274: 'dhole, Cuon alpinus',
 301 |     275: 'African hunting dog, hyena dog, Cape hunting dog, Lycaon pictus',
 302 |     276: 'hyena, hyaena',
 303 |     277: 'red fox, Vulpes vulpes',
 304 |     278: 'kit fox, Vulpes macrotis',
 305 |     279: 'Arctic fox, white fox, Alopex lagopus',
 306 |     280: 'grey fox, gray fox, Urocyon cinereoargenteus',
 307 |     281: 'tabby, tabby cat',
 308 |     282: 'tiger cat',
 309 |     283: 'Persian cat',
 310 |     284: 'Siamese cat, Siamese',
 311 |     285: 'Egyptian cat',
 312 |     286: 'cougar, puma, catamount, mountain lion, painter, panther, ' +
 313 |         'Felis concolor',
 314 |     287: 'lynx, catamount',
 315 |     288: 'leopard, Panthera pardus',
 316 |     289: 'snow leopard, ounce, Panthera uncia',
 317 |     290: 'jaguar, panther, Panthera onca, Felis onca',
 318 |     291: 'lion, king of beasts, Panthera leo',
 319 |     292: 'tiger, Panthera tigris',
 320 |     293: 'cheetah, chetah, Acinonyx jubatus',
 321 |     294: 'brown bear, bruin, Ursus arctos',
 322 |     295: 'American black bear, black bear, Ursus americanus, Euarctos ' +
 323 |         'americanus',
 324 |     296: 'ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus',
 325 |     297: 'sloth bear, Melursus ursinus, Ursus ursinus',
 326 |     298: 'mongoose',
 327 |     299: 'meerkat, mierkat',
 328 |     300: 'tiger beetle',
 329 |     301: 'ladybug, ladybeetle, lady beetle, ladybird, ladybird beetle',
 330 |     302: 'ground beetle, carabid beetle',
 331 |     303: 'long-horned beetle, longicorn, longicorn beetle',
 332 |     304: 'leaf beetle, chrysomelid',
 333 |     305: 'dung beetle',
 334 |     306: 'rhinoceros beetle',
 335 |     307: 'weevil',
 336 |     308: 'fly',
 337 |     309: 'bee',
 338 |     310: 'ant, emmet, pismire',
 339 |     311: 'grasshopper, hopper',
 340 |     312: 'cricket',
 341 |     313: 'walking stick, walkingstick, stick insect',
 342 |     314: 'cockroach, roach',
 343 |     315: 'mantis, mantid',
 344 |     316: 'cicada, cicala',
 345 |     317: 'leafhopper',
 346 |     318: 'lacewing, lacewing fly',
 347 |     319: 'dragonfly, darning needle, devil\'s darning needle, sewing needle, ' +
 348 |         'snake feeder, snake doctor, mosquito hawk, skeeter hawk',
 349 |     320: 'damselfly',
 350 |     321: 'admiral',
 351 |     322: 'ringlet, ringlet butterfly',
 352 |     323: 'monarch, monarch butterfly, milkweed butterfly, Danaus plexippus',
 353 |     324: 'cabbage butterfly',
 354 |     325: 'sulphur butterfly, sulfur butterfly',
 355 |     326: 'lycaenid, lycaenid butterfly',
 356 |     327: 'starfish, sea star',
 357 |     328: 'sea urchin',
 358 |     329: 'sea cucumber, holothurian',
 359 |     330: 'wood rabbit, cottontail, cottontail rabbit',
 360 |     331: 'hare',
 361 |     332: 'Angora, Angora rabbit',
 362 |     333: 'hamster',
 363 |     334: 'porcupine, hedgehog',
 364 |     335: 'fox squirrel, eastern fox squirrel, Sciurus niger',
 365 |     336: 'marmot',
 366 |     337: 'beaver',
 367 |     338: 'guinea pig, Cavia cobaya',
 368 |     339: 'sorrel',
 369 |     340: 'zebra',
 370 |     341: 'hog, pig, grunter, squealer, Sus scrofa',
 371 |     342: 'wild boar, boar, Sus scrofa',
 372 |     343: 'warthog',
 373 |     344: 'hippopotamus, hippo, river horse, Hippopotamus amphibius',
 374 |     345: 'ox',
 375 |     346: 'water buffalo, water ox, Asiatic buffalo, Bubalus bubalis',
 376 |     347: 'bison',
 377 |     348: 'ram, tup',
 378 |     349: 'bighorn, bighorn sheep, cimarron, Rocky Mountain bighorn, Rocky ' +
 379 |         'Mountain sheep, Ovis canadensis',
 380 |     350: 'ibex, Capra ibex',
 381 |     351: 'hartebeest',
 382 |     352: 'impala, Aepyceros melampus',
 383 |     353: 'gazelle',
 384 |     354: 'Arabian camel, dromedary, Camelus dromedarius',
 385 |     355: 'llama',
 386 |     356: 'weasel',
 387 |     357: 'mink',
 388 |     358: 'polecat, fitch, foulmart, foumart, Mustela putorius',
 389 |     359: 'black-footed ferret, ferret, Mustela nigripes',
 390 |     360: 'otter',
 391 |     361: 'skunk, polecat, wood pussy',
 392 |     362: 'badger',
 393 |     363: 'armadillo',
 394 |     364: 'three-toed sloth, ai, Bradypus tridactylus',
 395 |     365: 'orangutan, orang, orangutang, Pongo pygmaeus',
 396 |     366: 'gorilla, Gorilla gorilla',
 397 |     367: 'chimpanzee, chimp, Pan troglodytes',
 398 |     368: 'gibbon, Hylobates lar',
 399 |     369: 'siamang, Hylobates syndactylus, Symphalangus syndactylus',
 400 |     370: 'guenon, guenon monkey',
 401 |     371: 'patas, hussar monkey, Erythrocebus patas',
 402 |     372: 'baboon',
 403 |     373: 'macaque',
 404 |     374: 'langur',
 405 |     375: 'colobus, colobus monkey',
 406 |     376: 'proboscis monkey, Nasalis larvatus',
 407 |     377: 'marmoset',
 408 |     378: 'capuchin, ringtail, Cebus capucinus',
 409 |     379: 'howler monkey, howler',
 410 |     380: 'titi, titi monkey',
 411 |     381: 'spider monkey, Ateles geoffroyi',
 412 |     382: 'squirrel monkey, Saimiri sciureus',
 413 |     383: 'Madagascar cat, ring-tailed lemur, Lemur catta',
 414 |     384: 'indri, indris, Indri indri, Indri brevicaudatus',
 415 |     385: 'Indian elephant, Elephas maximus',
 416 |     386: 'African elephant, Loxodonta africana',
 417 |     387: 'lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens',
 418 |     388: 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca',
 419 |     389: 'barracouta, snoek',
 420 |     390: 'eel',
 421 |     391: 'coho, cohoe, coho salmon, blue jack, silver salmon, Oncorhynchus ' +
 422 |         'kisutch',
 423 |     392: 'rock beauty, Holocanthus tricolor',
 424 |     393: 'anemone fish',
 425 |     394: 'sturgeon',
 426 |     395: 'gar, garfish, garpike, billfish, Lepisosteus osseus',
 427 |     396: 'lionfish',
 428 |     397: 'puffer, pufferfish, blowfish, globefish',
 429 |     398: 'abacus',
 430 |     399: 'abaya',
 431 |     400: 'academic gown, academic robe, judge\'s robe',
 432 |     401: 'accordion, piano accordion, squeeze box',
 433 |     402: 'acoustic guitar',
 434 |     403: 'aircraft carrier, carrier, flattop, attack aircraft carrier',
 435 |     404: 'airliner',
 436 |     405: 'airship, dirigible',
 437 |     406: 'altar',
 438 |     407: 'ambulance',
 439 |     408: 'amphibian, amphibious vehicle',
 440 |     409: 'analog clock',
 441 |     410: 'apiary, bee house',
 442 |     411: 'apron',
 443 |     412: 'ashcan, trash can, garbage can, wastebin, ash bin, ash-bin, ashbin, ' +
 444 |         'dustbin, trash barrel, trash bin',
 445 |     413: 'assault rifle, assault gun',
 446 |     414: 'backpack, back pack, knapsack, packsack, rucksack, haversack',
 447 |     415: 'bakery, bakeshop, bakehouse',
 448 |     416: 'balance beam, beam',
 449 |     417: 'balloon',
 450 |     418: 'ballpoint, ballpoint pen, ballpen, Biro',
 451 |     419: 'Band Aid',
 452 |     420: 'banjo',
 453 |     421: 'bannister, banister, balustrade, balusters, handrail',
 454 |     422: 'barbell',
 455 |     423: 'barber chair',
 456 |     424: 'barbershop',
 457 |     425: 'barn',
 458 |     426: 'barometer',
 459 |     427: 'barrel, cask',
 460 |     428: 'barrow, garden cart, lawn cart, wheelbarrow',
 461 |     429: 'baseball',
 462 |     430: 'basketball',
 463 |     431: 'bassinet',
 464 |     432: 'bassoon',
 465 |     433: 'bathing cap, swimming cap',
 466 |     434: 'bath towel',
 467 |     435: 'bathtub, bathing tub, bath, tub',
 468 |     436: 'beach wagon, station wagon, wagon, estate car, beach waggon, station ' +
 469 |         'waggon, waggon',
 470 |     437: 'beacon, lighthouse, beacon light, pharos',
 471 |     438: 'beaker',
 472 |     439: 'bearskin, busby, shako',
 473 |     440: 'beer bottle',
 474 |     441: 'beer glass',
 475 |     442: 'bell cote, bell cot',
 476 |     443: 'bib',
 477 |     444: 'bicycle-built-for-two, tandem bicycle, tandem',
 478 |     445: 'bikini, two-piece',
 479 |     446: 'binder, ring-binder',
 480 |     447: 'binoculars, field glasses, opera glasses',
 481 |     448: 'birdhouse',
 482 |     449: 'boathouse',
 483 |     450: 'bobsled, bobsleigh, bob',
 484 |     451: 'bolo tie, bolo, bola tie, bola',
 485 |     452: 'bonnet, poke bonnet',
 486 |     453: 'bookcase',
 487 |     454: 'bookshop, bookstore, bookstall',
 488 |     455: 'bottlecap',
 489 |     456: 'bow',
 490 |     457: 'bow tie, bow-tie, bowtie',
 491 |     458: 'brass, memorial tablet, plaque',
 492 |     459: 'brassiere, bra, bandeau',
 493 |     460: 'breakwater, groin, groyne, mole, bulwark, seawall, jetty',
 494 |     461: 'breastplate, aegis, egis',
 495 |     462: 'broom',
 496 |     463: 'bucket, pail',
 497 |     464: 'buckle',
 498 |     465: 'bulletproof vest',
 499 |     466: 'bullet train, bullet',
 500 |     467: 'butcher shop, meat market',
 501 |     468: 'cab, hack, taxi, taxicab',
 502 |     469: 'caldron, cauldron',
 503 |     470: 'candle, taper, wax light',
 504 |     471: 'cannon',
 505 |     472: 'canoe',
 506 |     473: 'can opener, tin opener',
 507 |     474: 'cardigan',
 508 |     475: 'car mirror',
 509 |     476: 'carousel, carrousel, merry-go-round, roundabout, whirligig',
 510 |     477: 'carpenter\'s kit, tool kit',
 511 |     478: 'carton',
 512 |     479: 'car wheel',
 513 |     480: 'cash machine, cash dispenser, automated teller machine, automatic ' +
 514 |         'teller machine, automated teller, automatic teller, ATM',
 515 |     481: 'cassette',
 516 |     482: 'cassette player',
 517 |     483: 'castle',
 518 |     484: 'catamaran',
 519 |     485: 'CD player',
 520 |     486: 'cello, violoncello',
 521 |     487: 'cellular telephone, cellular phone, cellphone, cell, mobile phone',
 522 |     488: 'chain',
 523 |     489: 'chainlink fence',
 524 |     490: 'chain mail, ring mail, mail, chain armor, chain armour, ring armor, ' +
 525 |         'ring armour',
 526 |     491: 'chain saw, chainsaw',
 527 |     492: 'chest',
 528 |     493: 'chiffonier, commode',
 529 |     494: 'chime, bell, gong',
 530 |     495: 'china cabinet, china closet',
 531 |     496: 'Christmas stocking',
 532 |     497: 'church, church building',
 533 |     498: 'cinema, movie theater, movie theatre, movie house, picture palace',
 534 |     499: 'cleaver, meat cleaver, chopper',
 535 |     500: 'cliff dwelling',
 536 |     501: 'cloak',
 537 |     502: 'clog, geta, patten, sabot',
 538 |     503: 'cocktail shaker',
 539 |     504: 'coffee mug',
 540 |     505: 'coffeepot',
 541 |     506: 'coil, spiral, volute, whorl, helix',
 542 |     507: 'combination lock',
 543 |     508: 'computer keyboard, keypad',
 544 |     509: 'confectionery, confectionary, candy store',
 545 |     510: 'container ship, containership, container vessel',
 546 |     511: 'convertible',
 547 |     512: 'corkscrew, bottle screw',
 548 |     513: 'cornet, horn, trumpet, trump',
 549 |     514: 'cowboy boot',
 550 |     515: 'cowboy hat, ten-gallon hat',
 551 |     516: 'cradle',
 552 |     517: 'crane',
 553 |     518: 'crash helmet',
 554 |     519: 'crate',
 555 |     520: 'crib, cot',
 556 |     521: 'Crock Pot',
 557 |     522: 'croquet ball',
 558 |     523: 'crutch',
 559 |     524: 'cuirass',
 560 |     525: 'dam, dike, dyke',
 561 |     526: 'desk',
 562 |     527: 'desktop computer',
 563 |     528: 'dial telephone, dial phone',
 564 |     529: 'diaper, nappy, napkin',
 565 |     530: 'digital clock',
 566 |     531: 'digital watch',
 567 |     532: 'dining table, board',
 568 |     533: 'dishrag, dishcloth',
 569 |     534: 'dishwasher, dish washer, dishwashing machine',
 570 |     535: 'disk brake, disc brake',
 571 |     536: 'dock, dockage, docking facility',
 572 |     537: 'dogsled, dog sled, dog sleigh',
 573 |     538: 'dome',
 574 |     539: 'doormat, welcome mat',
 575 |     540: 'drilling platform, offshore rig',
 576 |     541: 'drum, membranophone, tympan',
 577 |     542: 'drumstick',
 578 |     543: 'dumbbell',
 579 |     544: 'Dutch oven',
 580 |     545: 'electric fan, blower',
 581 |     546: 'electric guitar',
 582 |     547: 'electric locomotive',
 583 |     548: 'entertainment center',
 584 |     549: 'envelope',
 585 |     550: 'espresso maker',
 586 |     551: 'face powder',
 587 |     552: 'feather boa, boa',
 588 |     553: 'file, file cabinet, filing cabinet',
 589 |     554: 'fireboat',
 590 |     555: 'fire engine, fire truck',
 591 |     556: 'fire screen, fireguard',
 592 |     557: 'flagpole, flagstaff',
 593 |     558: 'flute, transverse flute',
 594 |     559: 'folding chair',
 595 |     560: 'football helmet',
 596 |     561: 'forklift',
 597 |     562: 'fountain',
 598 |     563: 'fountain pen',
 599 |     564: 'four-poster',
 600 |     565: 'freight car',
 601 |     566: 'French horn, horn',
 602 |     567: 'frying pan, frypan, skillet',
 603 |     568: 'fur coat',
 604 |     569: 'garbage truck, dustcart',
 605 |     570: 'gasmask, respirator, gas helmet',
 606 |     571: 'gas pump, gasoline pump, petrol pump, island dispenser',
 607 |     572: 'goblet',
 608 |     573: 'go-kart',
 609 |     574: 'golf ball',
 610 |     575: 'golfcart, golf cart',
 611 |     576: 'gondola',
 612 |     577: 'gong, tam-tam',
 613 |     578: 'gown',
 614 |     579: 'grand piano, grand',
 615 |     580: 'greenhouse, nursery, glasshouse',
 616 |     581: 'grille, radiator grille',
 617 |     582: 'grocery store, grocery, food market, market',
 618 |     583: 'guillotine',
 619 |     584: 'hair slide',
 620 |     585: 'hair spray',
 621 |     586: 'half track',
 622 |     587: 'hammer',
 623 |     588: 'hamper',
 624 |     589: 'hand blower, blow dryer, blow drier, hair dryer, hair drier',
 625 |     590: 'hand-held computer, hand-held microcomputer',
 626 |     591: 'handkerchief, hankie, hanky, hankey',
 627 |     592: 'hard disc, hard disk, fixed disk',
 628 |     593: 'harmonica, mouth organ, harp, mouth harp',
 629 |     594: 'harp',
 630 |     595: 'harvester, reaper',
 631 |     596: 'hatchet',
 632 |     597: 'holster',
 633 |     598: 'home theater, home theatre',
 634 |     599: 'honeycomb',
 635 |     600: 'hook, claw',
 636 |     601: 'hoopskirt, crinoline',
 637 |     602: 'horizontal bar, high bar',
 638 |     603: 'horse cart, horse-cart',
 639 |     604: 'hourglass',
 640 |     605: 'iPod',
 641 |     606: 'iron, smoothing iron',
 642 |     607: 'jack-o\'-lantern',
 643 |     608: 'jean, blue jean, denim',
 644 |     609: 'jeep, landrover',
 645 |     610: 'jersey, T-shirt, tee shirt',
 646 |     611: 'jigsaw puzzle',
 647 |     612: 'jinrikisha, ricksha, rickshaw',
 648 |     613: 'joystick',
 649 |     614: 'kimono',
 650 |     615: 'knee pad',
 651 |     616: 'knot',
 652 |     617: 'lab coat, laboratory coat',
 653 |     618: 'ladle',
 654 |     619: 'lampshade, lamp shade',
 655 |     620: 'laptop, laptop computer',
 656 |     621: 'lawn mower, mower',
 657 |     622: 'lens cap, lens cover',
 658 |     623: 'letter opener, paper knife, paperknife',
 659 |     624: 'library',
 660 |     625: 'lifeboat',
 661 |     626: 'lighter, light, igniter, ignitor',
 662 |     627: 'limousine, limo',
 663 |     628: 'liner, ocean liner',
 664 |     629: 'lipstick, lip rouge',
 665 |     630: 'Loafer',
 666 |     631: 'lotion',
 667 |     632: 'loudspeaker, speaker, speaker unit, loudspeaker system, speaker ' +
 668 |         'system',
 669 |     633: 'loupe, jeweler\'s loupe',
 670 |     634: 'lumbermill, sawmill',
 671 |     635: 'magnetic compass',
 672 |     636: 'mailbag, postbag',
 673 |     637: 'mailbox, letter box',
 674 |     638: 'maillot',
 675 |     639: 'maillot, tank suit',
 676 |     640: 'manhole cover',
 677 |     641: 'maraca',
 678 |     642: 'marimba, xylophone',
 679 |     643: 'mask',
 680 |     644: 'matchstick',
 681 |     645: 'maypole',
 682 |     646: 'maze, labyrinth',
 683 |     647: 'measuring cup',
 684 |     648: 'medicine chest, medicine cabinet',
 685 |     649: 'megalith, megalithic structure',
 686 |     650: 'microphone, mike',
 687 |     651: 'microwave, microwave oven',
 688 |     652: 'military uniform',
 689 |     653: 'milk can',
 690 |     654: 'minibus',
 691 |     655: 'miniskirt, mini',
 692 |     656: 'minivan',
 693 |     657: 'missile',
 694 |     658: 'mitten',
 695 |     659: 'mixing bowl',
 696 |     660: 'mobile home, manufactured home',
 697 |     661: 'Model T',
 698 |     662: 'modem',
 699 |     663: 'monastery',
 700 |     664: 'monitor',
 701 |     665: 'moped',
 702 |     666: 'mortar',
 703 |     667: 'mortarboard',
 704 |     668: 'mosque',
 705 |     669: 'mosquito net',
 706 |     670: 'motor scooter, scooter',
 707 |     671: 'mountain bike, all-terrain bike, off-roader',
 708 |     672: 'mountain tent',
 709 |     673: 'mouse, computer mouse',
 710 |     674: 'mousetrap',
 711 |     675: 'moving van',
 712 |     676: 'muzzle',
 713 |     677: 'nail',
 714 |     678: 'neck brace',
 715 |     679: 'necklace',
 716 |     680: 'nipple',
 717 |     681: 'notebook, notebook computer',
 718 |     682: 'obelisk',
 719 |     683: 'oboe, hautboy, hautbois',
 720 |     684: 'ocarina, sweet potato',
 721 |     685: 'odometer, hodometer, mileometer, milometer',
 722 |     686: 'oil filter',
 723 |     687: 'organ, pipe organ',
 724 |     688: 'oscilloscope, scope, cathode-ray oscilloscope, CRO',
 725 |     689: 'overskirt',
 726 |     690: 'oxcart',
 727 |     691: 'oxygen mask',
 728 |     692: 'packet',
 729 |     693: 'paddle, boat paddle',
 730 |     694: 'paddlewheel, paddle wheel',
 731 |     695: 'padlock',
 732 |     696: 'paintbrush',
 733 |     697: 'pajama, pyjama, pj\'s, jammies',
 734 |     698: 'palace',
 735 |     699: 'panpipe, pandean pipe, syrinx',
 736 |     700: 'paper towel',
 737 |     701: 'parachute, chute',
 738 |     702: 'parallel bars, bars',
 739 |     703: 'park bench',
 740 |     704: 'parking meter',
 741 |     705: 'passenger car, coach, carriage',
 742 |     706: 'patio, terrace',
 743 |     707: 'pay-phone, pay-station',
 744 |     708: 'pedestal, plinth, footstall',
 745 |     709: 'pencil box, pencil case',
 746 |     710: 'pencil sharpener',
 747 |     711: 'perfume, essence',
 748 |     712: 'Petri dish',
 749 |     713: 'photocopier',
 750 |     714: 'pick, plectrum, plectron',
 751 |     715: 'pickelhaube',
 752 |     716: 'picket fence, paling',
 753 |     717: 'pickup, pickup truck',
 754 |     718: 'pier',
 755 |     719: 'piggy bank, penny bank',
 756 |     720: 'pill bottle',
 757 |     721: 'pillow',
 758 |     722: 'ping-pong ball',
 759 |     723: 'pinwheel',
 760 |     724: 'pirate, pirate ship',
 761 |     725: 'pitcher, ewer',
 762 |     726: 'plane, carpenter\'s plane, woodworking plane',
 763 |     727: 'planetarium',
 764 |     728: 'plastic bag',
 765 |     729: 'plate rack',
 766 |     730: 'plow, plough',
 767 |     731: 'plunger, plumber\'s helper',
 768 |     732: 'Polaroid camera, Polaroid Land camera',
 769 |     733: 'pole',
 770 |     734: 'police van, police wagon, paddy wagon, patrol wagon, wagon, black ' +
 771 |         'Maria',
 772 |     735: 'poncho',
 773 |     736: 'pool table, billiard table, snooker table',
 774 |     737: 'pop bottle, soda bottle',
 775 |     738: 'pot, flowerpot',
 776 |     739: 'potter\'s wheel',
 777 |     740: 'power drill',
 778 |     741: 'prayer rug, prayer mat',
 779 |     742: 'printer',
 780 |     743: 'prison, prison house',
 781 |     744: 'projectile, missile',
 782 |     745: 'projector',
 783 |     746: 'puck, hockey puck',
 784 |     747: 'punching bag, punch bag, punching ball, punchball',
 785 |     748: 'purse',
 786 |     749: 'quill, quill pen',
 787 |     750: 'quilt, comforter, comfort, puff',
 788 |     751: 'racer, race car, racing car',
 789 |     752: 'racket, racquet',
 790 |     753: 'radiator',
 791 |     754: 'radio, wireless',
 792 |     755: 'radio telescope, radio reflector',
 793 |     756: 'rain barrel',
 794 |     757: 'recreational vehicle, RV, R.V.',
 795 |     758: 'reel',
 796 |     759: 'reflex camera',
 797 |     760: 'refrigerator, icebox',
 798 |     761: 'remote control, remote',
 799 |     762: 'restaurant, eating house, eating place, eatery',
 800 |     763: 'revolver, six-gun, six-shooter',
 801 |     764: 'rifle',
 802 |     765: 'rocking chair, rocker',
 803 |     766: 'rotisserie',
 804 |     767: 'rubber eraser, rubber, pencil eraser',
 805 |     768: 'rugby ball',
 806 |     769: 'rule, ruler',
 807 |     770: 'running shoe',
 808 |     771: 'safe',
 809 |     772: 'safety pin',
 810 |     773: 'saltshaker, salt shaker',
 811 |     774: 'sandal',
 812 |     775: 'sarong',
 813 |     776: 'sax, saxophone',
 814 |     777: 'scabbard',
 815 |     778: 'scale, weighing machine',
 816 |     779: 'school bus',
 817 |     780: 'schooner',
 818 |     781: 'scoreboard',
 819 |     782: 'screen, CRT screen',
 820 |     783: 'screw',
 821 |     784: 'screwdriver',
 822 |     785: 'seat belt, seatbelt',
 823 |     786: 'sewing machine',
 824 |     787: 'shield, buckler',
 825 |     788: 'shoe shop, shoe-shop, shoe store',
 826 |     789: 'shoji',
 827 |     790: 'shopping basket',
 828 |     791: 'shopping cart',
 829 |     792: 'shovel',
 830 |     793: 'shower cap',
 831 |     794: 'shower curtain',
 832 |     795: 'ski',
 833 |     796: 'ski mask',
 834 |     797: 'sleeping bag',
 835 |     798: 'slide rule, slipstick',
 836 |     799: 'sliding door',
 837 |     800: 'slot, one-armed bandit',
 838 |     801: 'snorkel',
 839 |     802: 'snowmobile',
 840 |     803: 'snowplow, snowplough',
 841 |     804: 'soap dispenser',
 842 |     805: 'soccer ball',
 843 |     806: 'sock',
 844 |     807: 'solar dish, solar collector, solar furnace',
 845 |     808: 'sombrero',
 846 |     809: 'soup bowl',
 847 |     810: 'space bar',
 848 |     811: 'space heater',
 849 |     812: 'space shuttle',
 850 |     813: 'spatula',
 851 |     814: 'speedboat',
 852 |     815: 'spider web, spider\'s web',
 853 |     816: 'spindle',
 854 |     817: 'sports car, sport car',
 855 |     818: 'spotlight, spot',
 856 |     819: 'stage',
 857 |     820: 'steam locomotive',
 858 |     821: 'steel arch bridge',
 859 |     822: 'steel drum',
 860 |     823: 'stethoscope',
 861 |     824: 'stole',
 862 |     825: 'stone wall',
 863 |     826: 'stopwatch, stop watch',
 864 |     827: 'stove',
 865 |     828: 'strainer',
 866 |     829: 'streetcar, tram, tramcar, trolley, trolley car',
 867 |     830: 'stretcher',
 868 |     831: 'studio couch, day bed',
 869 |     832: 'stupa, tope',
 870 |     833: 'submarine, pigboat, sub, U-boat',
 871 |     834: 'suit, suit of clothes',
 872 |     835: 'sundial',
 873 |     836: 'sunglass',
 874 |     837: 'sunglasses, dark glasses, shades',
 875 |     838: 'sunscreen, sunblock, sun blocker',
 876 |     839: 'suspension bridge',
 877 |     840: 'swab, swob, mop',
 878 |     841: 'sweatshirt',
 879 |     842: 'swimming trunks, bathing trunks',
 880 |     843: 'swing',
 881 |     844: 'switch, electric switch, electrical switch',
 882 |     845: 'syringe',
 883 |     846: 'table lamp',
 884 |     847: 'tank, army tank, armored combat vehicle, armoured combat vehicle',
 885 |     848: 'tape player',
 886 |     849: 'teapot',
 887 |     850: 'teddy, teddy bear',
 888 |     851: 'television, television system',
 889 |     852: 'tennis ball',
 890 |     853: 'thatch, thatched roof',
 891 |     854: 'theater curtain, theatre curtain',
 892 |     855: 'thimble',
 893 |     856: 'thresher, thrasher, threshing machine',
 894 |     857: 'throne',
 895 |     858: 'tile roof',
 896 |     859: 'toaster',
 897 |     860: 'tobacco shop, tobacconist shop, tobacconist',
 898 |     861: 'toilet seat',
 899 |     862: 'torch',
 900 |     863: 'totem pole',
 901 |     864: 'tow truck, tow car, wrecker',
 902 |     865: 'toyshop',
 903 |     866: 'tractor',
 904 |     867: 'trailer truck, tractor trailer, trucking rig, rig, articulated ' +
 905 |         'lorry, semi',
 906 |     868: 'tray',
 907 |     869: 'trench coat',
 908 |     870: 'tricycle, trike, velocipede',
 909 |     871: 'trimaran',
 910 |     872: 'tripod',
 911 |     873: 'triumphal arch',
 912 |     874: 'trolleybus, trolley coach, trackless trolley',
 913 |     875: 'trombone',
 914 |     876: 'tub, vat',
 915 |     877: 'turnstile',
 916 |     878: 'typewriter keyboard',
 917 |     879: 'umbrella',
 918 |     880: 'unicycle, monocycle',
 919 |     881: 'upright, upright piano',
 920 |     882: 'vacuum, vacuum cleaner',
 921 |     883: 'vase',
 922 |     884: 'vault',
 923 |     885: 'velvet',
 924 |     886: 'vending machine',
 925 |     887: 'vestment',
 926 |     888: 'viaduct',
 927 |     889: 'violin, fiddle',
 928 |     890: 'volleyball',
 929 |     891: 'waffle iron',
 930 |     892: 'wall clock',
 931 |     893: 'wallet, billfold, notecase, pocketbook',
 932 |     894: 'wardrobe, closet, press',
 933 |     895: 'warplane, military plane',
 934 |     896: 'washbasin, handbasin, washbowl, lavabo, wash-hand basin',
 935 |     897: 'washer, automatic washer, washing machine',
 936 |     898: 'water bottle',
 937 |     899: 'water jug',
 938 |     900: 'water tower',
 939 |     901: 'whiskey jug',
 940 |     902: 'whistle',
 941 |     903: 'wig',
 942 |     904: 'window screen',
 943 |     905: 'window shade',
 944 |     906: 'Windsor tie',
 945 |     907: 'wine bottle',
 946 |     908: 'wing',
 947 |     909: 'wok',
 948 |     910: 'wooden spoon',
 949 |     911: 'wool, woolen, woollen',
 950 |     912: 'worm fence, snake fence, snake-rail fence, Virginia fence',
 951 |     913: 'wreck',
 952 |     914: 'yawl',
 953 |     915: 'yurt',
 954 |     916: 'web site, website, internet site, site',
 955 |     917: 'comic book',
 956 |     918: 'crossword puzzle, crossword',
 957 |     919: 'street sign',
 958 |     920: 'traffic light, traffic signal, stoplight',
 959 |     921: 'book jacket, dust cover, dust jacket, dust wrapper',
 960 |     922: 'menu',
 961 |     923: 'plate',
 962 |     924: 'guacamole',
 963 |     925: 'consomme',
 964 |     926: 'hot pot, hotpot',
 965 |     927: 'trifle',
 966 |     928: 'ice cream, icecream',
 967 |     929: 'ice lolly, lolly, lollipop, popsicle',
 968 |     930: 'French loaf',
 969 |     931: 'bagel, beigel',
 970 |     932: 'pretzel',
 971 |     933: 'cheeseburger',
 972 |     934: 'hotdog, hot dog, red hot',
 973 |     935: 'mashed potato',
 974 |     936: 'head cabbage',
 975 |     937: 'broccoli',
 976 |     938: 'cauliflower',
 977 |     939: 'zucchini, courgette',
 978 |     940: 'spaghetti squash',
 979 |     941: 'acorn squash',
 980 |     942: 'butternut squash',
 981 |     943: 'cucumber, cuke',
 982 |     944: 'artichoke, globe artichoke',
 983 |     945: 'bell pepper',
 984 |     946: 'cardoon',
 985 |     947: 'mushroom',
 986 |     948: 'Granny Smith',
 987 |     949: 'strawberry',
 988 |     950: 'orange',
 989 |     951: 'lemon',
 990 |     952: 'fig',
 991 |     953: 'pineapple, ananas',
 992 |     954: 'banana',
 993 |     955: 'jackfruit, jak, jack',
 994 |     956: 'custard apple',
 995 |     957: 'pomegranate',
 996 |     958: 'hay',
 997 |     959: 'carbonara',
 998 |     960: 'chocolate sauce, chocolate syrup',
 999 |     961: 'dough',
1000 |     962: 'meat loaf, meatloaf',
1001 |     963: 'pizza, pizza pie',
1002 |     964: 'potpie',
1003 |     965: 'burrito',
1004 |     966: 'red wine',
1005 |     967: 'espresso',
1006 |     968: 'cup',
1007 |     969: 'eggnog',
1008 |     970: 'alp',
1009 |     971: 'bubble',
1010 |     972: 'cliff, drop, drop-off',
1011 |     973: 'coral reef',
1012 |     974: 'geyser',
1013 |     975: 'lakeside, lakeshore',
1014 |     976: 'promontory, headland, head, foreland',
1015 |     977: 'sandbar, sand bar',
1016 |     978: 'seashore, coast, seacoast, sea-coast',
1017 |     979: 'valley, vale',
1018 |     980: 'volcano',
1019 |     981: 'ballplayer, baseball player',
1020 |     982: 'groom, bridegroom',
1021 |     983: 'scuba diver',
1022 |     984: 'rapeseed',
1023 |     985: 'daisy',
1024 |     986: 'yellow lady\'s slipper, yellow lady-slipper, Cypripedium calceolus, ' +
1025 |         'Cypripedium parviflorum',
1026 |     987: 'corn',
1027 |     988: 'acorn',
1028 |     989: 'hip, rose hip, rosehip',
1029 |     990: 'buckeye, horse chestnut, conker',
1030 |     991: 'coral fungus',
1031 |     992: 'agaric',
1032 |     993: 'gyromitra',
1033 |     994: 'stinkhorn, carrion fungus',
1034 |     995: 'earthstar',
1035 |     996: 'hen-of-the-woods, hen of the woods, Polyporus frondosus, Grifola ' +
1036 |         'frondosa',
1037 |     997: 'bolete',
1038 |     998: 'ear, spike, capitulum',
1039 |     999: 'toilet tissue, toilet paper, bathroom tissue'
1040 | };


--------------------------------------------------------------------------------