Brain js prediction

I'm trying to create ML with Brain.js that takes as input a number and outputs its count of significant digits.

Examples:

Input: 234 Output:3

Input: 2413 Output: 4

Input: 1 Output 1

<script src='https://cdn.rawgit.com/harthur-org/brain.js/aabe8cc2/browser.js'></script>

<script>


const network = new brain.NeuralNetwork();

//Only simple test data provided for now

network.train([
    { input: { Num: 2 }, output: { SigFigs: 1 } },
    { input: { Num: 12 }, output: { SigFigs: 2 } },
    { input: { Num: 432 }, output: { SigFigs: 3 } },
    { input: { Num: 1358 }, output: { SigFigs: 4 } },
    { input: { Num: 98 }, output: { SigFigs: 2 } },
    { input: { Num: 9123 }, output: { SigFigs: 4} },
    { input: { Num: 14 }, output: { SigFigs: 2} },
    { input: { Num: 763 }, output: { SigFigs: 3} },
], {
  log: true,
  iterations: 1e6,
  errorThresh: 0.00001
});

const result = network.run({ Num: 43 });

console.log(result); //SigFigs: 0.9999999396415676
//Expected output: 2

</script>

The result is completely no sense for me. I expect something like 2 and in worse case some other number. What I am doing wrong and what should I do to get the expected output?

Solution

You must set the output key as the number of sigfigs, then the value as 1.

So an example is:

network.train([
    { input: { Num: 2 }, output: { "1": 1 } },
    { input: { Num: 12 }, output: { "2": 1 } },
    { input: { Num: 432 }, output: { "3": 1 } }
])

When train runs, for the first item in the array, it sees there is an output for "1" but not for "2" or "3", so it sets "2" and "3" as zero.

In this case it keeps your range between 0 and 1, and allows you to treat 0 as false and 1 as true as if they are boolean values.

I haven't actually checked this with brain js as I'm writing this on my iPhone, but if your neutral net has trained correctly, you should get an output that looks like:

const result = network.run({ Num: 43 });
console.log(result);
=> "1": 0.0165, "2": 0.9726, "3": 0.0613