No connection between any variable and the result of the loss function - tensorflow.js

Question

The optimizer I'm using isn't finding a connection between my variables and the loss function.
I'm new to machine learning in general and I'm trying to build a curve fitting application for the equation y = a * 2^(t/b). The trainable variables being "a" and "b". Right now I'm testing it with some synthetic data.
// Declare variables
const a = tf.variable(tf.scalar(Math.random()));
const b = tf.variable(tf.scalar(Math.random()));
const year = tf.tensor1d([1,2,3,4]);
//Declare optimizer
var optimizer = tf.train.sgd(learningRate);
// y = a * 2^(t/b)
function Moore(t){
var Moore_results = [];
aPrime = a.dataSync();
bPrime = b.dataSync();
tPrime = t.dataSync();
for(i = 0; i < tPrime.length; i++){
Moore_results.push(aPrime * Math.pow(2,(tPrime[i]/bPrime)) );}
const tensorMoore = tf.variable(tf.tensor1d(Moore_results));
return tensorMoore;}
// Define loss function
function loss(predictions, labels) {
const meanSquareError = predictions.sub(labels).square().mean();
return meanSquareError;}
//Optimize loss function
optimizer.minimize(() => loss(Moore(year),year));
When I run the optimizer I get the error.
Cannot find a connection between any variable and the result of the loss >function y=f(x). Please make sure the operations that use variables are inside >the function f passed to minimize().
Any help is greatly appreciated.

Answer 1

The optimizer tries to optimize a variable tensor created with tf.variable given the loss between the prediction and the expected value. You have to pass a prediction value as the following :
function predict(t) {
// y = a * 2 ^ (t / b)
return tf.tidy(() => {
return a.mul(tf.scalar(2)) // a * 2
.pow(t.div(b)) // ^ (t / b)
});
}
for (let iter = 0; iter < numIterations; iter++) {
optimizer.minimize(() => {
const predsYs = predict(xs);
return loss(predsYs, ys);
});
}
The optimizer will then compute the value of a and b that best predict your model