How do I normalise dates in Tensorflow.js? - tensorflow

Question

I've been building some simple linear regression models in Tensorflow.js with various types of data sets. However, I would now like to see what the relationship is between dates and price in my dataset.
In previous models, I normalise the price or other features so that the tensor is expressed as a vector between 0 and 1.
How would one do this dates, where the first date would have to be 0 and the last date in the range 1? Furthermore I would need to denormalise the Tensor afterwards.
I can convert the date into a unix timestamp using a library like date.fns... but I wondered if there might be a cleaner way to do this.
My normalise and denormalise functions :
function normalise (tensor) {
const min = tensor.min();
const max = tensor.max();
const normalisedTensor = tensor.sub(min).div(max.sub(min))
return {
tensor : normalisedTensor,
min,
max
}
}
function denormalise(tensor, min, max) {
const denormalisedTensor = tensor.mul(max.sub(min)).add(min);
return denormalisedTensor
}

Answer 1

The basic principle is that you need to convert the date formats (like 2021-03-21) to numbers, then you can use the normalise and denormalise functions that you already have. In your solution you have done that, however, you can write it in a simpler way. You can use date-fns, but the standard JavaScript Date object is also sufficient so I will use that.
Assuming your input is an array of date strings like 2021-05-01, then you can do this:
const dateStrings = ['2021-05-01', '2021-05-07', '2021-05-31'];
const timestamps = dateStrings.map(dateString => new Date(dateString).valueOf());
const unnormalisedTensor = tf.tensor1d(timestamps);
Then you can use the functions you have for normalisation (and denormalisation).
You can convert back from timestamps to date strings like this:
const timestamps = await unnormalisedTensor.array();
const dateStrings = timestamps.map(timestamp => new Date(timestamp).toISOString().substring(0,10));

Answer 2

I achieved this by using date.fns to convert the dates into timestamps and then normalising with the following function
const getTimestamps = (data) => {
data.forEach(record => {
let newDate = record.time
newDate = new Date(newDate);
const timestamp = getUnixTime(newDate);
console.log(newDate, timestamp)
record.timestamp = timestamp
}