Many ML models perform better when the input is normalised like stay between 0..1 interval.

How the stock time series with exponential like growth could be represented?

  • It could be normalised into 0..1 interval - but the tail will be almost like 0.

  • It also could be represented as a series of diffs, like Price[n+1] = diff[n]*Price[n], but it has another problem - it's not robust to missing data, if you miss even single diff - it would cause big error going forward. Given that many ML technics involve introducing random noise or losses for better learning - such fragile representation would make it harder.

  • It's also possible to represent it as a logarithmic scale.

I wonder what are the other ways to normalise such data as an input for ML like Neural Nets and others?

enter image description here


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.