5
$\begingroup$

I often have trouble deciding how to tweak my input data for the agent. I am changing my data so that values in the hundreds get changed to a value between -1 and 1, and values that are originally small get changed the same ratio. However, my agents often aren't successful.

Could the numeric range of the input data be a cause for the inability to learn?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

I assume that when you say

values that are originally small get changed the same ratio

you are standardizing your features so that every dimension is on the same scale from [-1,1]

This is common practice as it helps many models converge faster. It would not prevent your models from generalizing the data, so here are some other possible causes for your problem:

  • The ML algorithm you are using is not appropriate for your task
  • The data is either insufficient in quantity or quality (e.g. noisy data)
  • The preprocessing stage is lacking some steps such as dimensionality reduction
Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Reinforcement learning is a very broad concept, so one cannot make an absolute statement about my answer's application to reinforcement learning as a whole. If you provide more details about your dataset, objective, and the ML algorithms you are experimenting with, I can probably give more helpful advice. $\endgroup$
    – Ben
    Apr 25, 2018 at 12:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.