原博地址https://laboo.top/2018/11/21/tfjs-dr/源码digit-recognizerdemohttps://github-laziji.github.io/digit-recognizer/演示开始时需要加载大概100M的训练数据, 稍等片刻调整训练集的大小, 观察测试结果的准确性数据来源数据来源与 https://www.kaggle.com 中的一道题目 digit-recognizer题目给出42000条训练数据(包含图片和标签)以及28000条测试数据(只包含图片)要求给这些测试数据打上标签[0,1,2,3….,9] 要尽可能的准确网站中还有许多其他的机器学习的题目以及数据, 是个很好的练手的地方实现这里我们使用TensorFlow.js来实现这个项目创建模型卷积神经网络的第一层有两种作用, 它既是输入层也是执行层, 接收IMAGE_H * IMAGE_W大小的黑白像素最后一层是输出层, 有10个输出单元, 代表着0-9这十个值的概率分布, 例如 Label=2 , 输出为[0.02,0.01,0.9,…,0.01]function createConvModel() { const model = tf.sequential(); model.add(tf.layers.conv2d({ inputShape: [IMAGE_H, IMAGE_W, 1], kernelSize: 3, filters: 16, activation: ‘relu’ })); model.add(tf.layers.maxPooling2d({ poolSize: 2, strides: 2 })); model.add(tf.layers.conv2d({ kernelSize: 3, filters: 32, activation: ‘relu’ })); model.add(tf.layers.maxPooling2d({ poolSize: 2, strides: 2 })); model.add(tf.layers.conv2d({ kernelSize: 3, filters: 32, activation: ‘relu’ })); model.add(tf.layers.flatten({})); model.add(tf.layers.dense({ units: 64, activation: ‘relu’ })); model.add(tf.layers.dense({ units: 10, activation: ‘softmax’ })); return model;}训练模型我们选择适当的优化器和损失函数, 来编译模型async function train() { ui.trainLog(‘Create model…’); model = createConvModel(); ui.trainLog(‘Compile model…’); const optimizer = ‘rmsprop’; model.compile({ optimizer, loss: ‘categoricalCrossentropy’, metrics: [‘accuracy’], }); const trainData = Data.getTrainData(ui.getTrainNum()); ui.trainLog(‘Training model…’); await model.fit(trainData.xs, trainData.labels, {}); ui.trainLog(‘Completed!’); ui.trainCompleted();}测试这里测试一组测试数据, 返回对应的标签, 即十个输出单元中概率最高的下标function testOne(xs){ if(!model){ ui.viewLog(‘Need to train the model first’); return; } ui.viewLog(‘Testing…’); let output = model.predict(xs); ui.viewLog(‘Completed!’); output.print(); const axis = 1; const predictions = output.argMax(axis).dataSync(); return predictions[0];}欢迎关注我的博客公众号