Tensorflow.js tf.train.Optimizer 类 .computeGradients() 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。

执行 f() 并计算 f() 的标量输出相对于varList提供的可训练变量列表的梯度。如果未提供列表，则默认为所有可训练变量。

句法：

Optimizer.computeGradients(f, varList?);

参数：

• f ( ( ) => tf.Scalar)：要执行的函数，其输出用于计算变量的梯度。
• varLIst( tf.Variable[ ] )：用于计算梯度的可选变量列表。如果指定 ，只有可训练变量是 varList 才会计算梯度。默认为所有可训练变量。

返回： { value : tf.Scalar, grads : { [ name : 字符串 ] : tf.Tensor } }

示例 1：

Javascript

// Importing tensorflow
import * as tf from "@tensorflow/tfjs"
     
const xs = tf.tensor1d([3, 4, 5]);
const ys = tf.tensor1d([3.5, 4.7, 5.3]);
     
const x = tf.scalar(Math.random()).variable();
const y = tf.scalar(Math.random()).variable();
     
// Define a function f(x, y) = ( x^2 ) -  y.
const f = x => (x.square()).sub(y);
const loss = (pred, label) =>
    pred.sub(label).square().mean();
     
const learningRate = 0.05;
     
// Create adam optimizer
const optimizer =
tf.train.adam(learningRate);
     
// Train the model.
for (let i = 0; i < 6; i++) {
optimizer.computeGradients(() => 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: 0.38272422552108765, y: 0.7651948928833008
x: 0, pred: 8.2348051071167
x: 1, pred: 15.2348051071167
x: 2, pred: 24.234806060791016

示例 2：

Javascript

// Importing tensorflow
import * as tf from "@tensorflow/tfjs"
     
const xs = tf.tensor1d([0, 1, 2, 3]);
const ys = tf.tensor1d([1.3, 3.7, 12.4, 26.6]);
     
// 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.mul(3)).add(b.square(x)).add(c);
const loss = (pred, label) => pred.sub(label).square().mean();
     
// Setting configurations for our optimizer
const learningRate = 0.01;
const initialAccumulatorValue = 10;
 
     
// Create the Optimizer
const optimizer = tf.train.adagrad(learningRate,
        initialAccumulatorValue);
     
// Train the model.
for (let i = 0; i < 5; i++) {
optimizer.computeGradients(() => 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: 0.22211307287216187,
b: 0.2304522693157196,
c: 0.42621928453445435
x: 0, pred: 0.479327529668808
x: 1, pred: 1.1456668376922607
x: 2, pred: 1.8120059967041016
x: 3, pred: 2.4783451557159424

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