Tensorflow.js tf.train.Optimizer 类 .minimize() 方法

Tensorflow.js 是谷歌开发的一个开源库，用于在浏览器或节点环境中运行机器学习模型和深度学习神经网络。

.minimize()方法执行给定的函数f()并尝试通过以下方式最小化f()的标量输出 计算 y 相对于varList 表示的给定可训练变量列表的梯度。如果没有提供列表，它会计算所有可训练变量的梯度。

句法：

Optimizer.minimize (f, returnCost?, varList?)

参数：

• f (() => tf.Scalar)：它指定要执行的函数以及要最小化其输出的函数。
• returnCost (boolean)：指定是否返回执行 f() 产生的标量成本值。
• varList(tf.Variable[])：指定可训练变量的列表。

返回值： tf.Scalar |空值

示例 1：

Javascript

// Importing tensorflow
import tensorflow as tf
   
const xs = tf.tensor1d([0, 1, 2]);
const ys = tf.tensor1d([1.3, 2.5, 3.7]);
   
const x = tf.scalar(Math.random()).variable();
const y = tf.scalar(Math.random()).variable();
   
// Define a function f(x, y) = x + y.
const f = x => x.add(y);
const loss = (pred, label) =>
    pred.sub(label).square().mean();
   
const learningRate = 0.05;
   
// Create adagrad optimizer
const optimizer =
  tf.train.adagrad(learningRate);
   
// Train the model.
for (let i = 0; i < 5; i++) {
   optimizer.minimize(() => loss(f(xs), ys));
}
   
// Make predictions.
console.log(
`x: ${x.dataSync()}, y: ${y.dataSync()}`);
const preds = f(xs).dataSync();
preds.forEach((pred, i) => {
console.log(`x: ${i}, pred: ${pred}`);
});

输出

x: 0.9395854473114014, y: 1.0498266220092773
x: 0, pred: 1.0498266220092773
x: 1, pred: 2.0498266220092773
x: 2, pred: 3.0498266220092773

示例 2：

Javascript

// Importing tensorflow
import * as tf from "@tensorflow/tfjs"
   
const xs = tf.tensor1d([0, 1, 2, 3]);
const ys = tf.tensor1d([1.1, 5.9, 16.8, 33.9]);
   
// Choosing random coefficients
const a = tf.scalar(Math.random()).variable();
const b = tf.scalar(Math.random()).variable();
const c = tf.scalar(Math.random()).variable();
   
// Defining function f = (a*x^2 + b*x + c)
const f = x => a.mul(x.square()).add(b.mul(x)).add(c);
const loss = (pred, label) => pred.sub(label).square().mean();
   
// Setting configurations for our optimizer
const learningRate = 0.01;
const decay = 0.1;
const momentum = 1;
const epsilon = 0.5;
const centered = true;
   
// Create the ptimizer
const optimizer = tf.train.rmsprop(learningRate,
        decay, momentum, epsilon, centered);
   
// Train the model.
for (let i = 0; i < 8; i++) {
   optimizer.minimize(() => loss(f(xs), ys));
}
   
// Make predictions.
console.log(`a: ${a.dataSync()},
    b: ${b.dataSync()}, c: ${c.dataSync()}`);
const preds = f(xs).dataSync();
preds.forEach((pred, i) => {
   console.log(`x: ${i}, pred: ${pred}`);
});

输出

a: 3.6799352169036865, 
    b: 4.26292610168457, c: 4.544136047363281
x: 0, pred: 4.544136047363281
x: 1, pred: 12.486997604370117
x: 2, pred: 27.789730072021484
x: 3, pred: 50.45233154296875

参考： https://js.tensorflow.org/api/latest/#tf.train.Optimizer.minimize