0
$\begingroup$

How can I visualise PDF of distribution defined by quantiles, that I predict with my neural network? Now I'm passing quantiles to the histogram, but I don't think it is the correct way for visualising. I don't know it's simple/effectively to find the derivation of quantile function defined by the neural network. Exist any way how can comfortable be quantile function predicted by neural network converted to PDF for visualisation of distribution?

Chart: https://wandb.ai/markub/rl-toolkit/runs/2sgcmjr8/overview?workspace=user-markub

Thanks. Have a nice day.

Improve this question
$\endgroup$
2
  • $\begingroup$ You can already visualize the distribution with a histogram, so why do you want to calculate a pdf? $\endgroup$
    – user44143
    Commented Dec 27, 2021 at 22:44
  • $\begingroup$ The histogram that I have is representing same thing like PDF? I need it because PDF is more intuitively for public results of my research. When I pass quantiles to histogram I get same thing like visualising PDF (gaussian bell curve)? $\endgroup$
    – Bc. Martin Kubovčík
    Commented Dec 27, 2021 at 23:07

1 Answer 1

Reset to default
0
$\begingroup$

Smoothed histograms should be a way to go. See e.g. this paper and, more generally, these search results.

Improve this answer
$\endgroup$
5
  • $\begingroup$ Now, I'm using histogram in W&B where I can't do anything with that on backend. Is it correct way of visualisation? Picture of my histogram is here: wandb.ai/markub/rl-toolkit/runs/2sgcmjr8/… in section Summary. Please you can look on it? $\endgroup$
    – Bc. Martin Kubovčík
    Commented Dec 27, 2021 at 22:58
  • $\begingroup$ @Bc.MartinKubovčík : I do not know Python. So, your link is hardly of help to me. Can you tell me in purely mathematical terms what your difficulty is when you try to smooth the histogram (if you did try to smooth it)? $\endgroup$
    – Iosif Pinelis
    Commented Dec 28, 2021 at 0:31
  • $\begingroup$ I don't try to smooth it. The histogram is produced by third-party software where is a typical histogram with 64 bins and I don't think there is any smoothing (generally possible). Are the results that I give (visualised like picture in the link above) useful for presenting predicted distribution? Thanks. $\endgroup$
    – Bc. Martin Kubovčík
    Commented Dec 28, 2021 at 0:57
  • $\begingroup$ @Bc.MartinKubovčík : How can smoothing of a histogram be ever impossible? It may be impossible only because of limitations of the software you are using, if it does have such limitations. Also, it is hard to infer anything quantitative from a picture. If your problem is with software, then you definitely should seek help elsewhere. $\endgroup$
    – Iosif Pinelis
    Commented Dec 28, 2021 at 2:10
  • $\begingroup$ Thanks a lot. I'll write directly on support of my software. I don't know if using histogram for quantile function is correctly way. Now I know it. $\endgroup$
    – Bc. Martin Kubovčík
    Commented Dec 28, 2021 at 2:14

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .