0
$\begingroup$

In the following example in Google tutorials, it says the distribution of the house price and house size is poisson.

https://developers.google.com/machine-learning/clustering/similarity/generating-embeddings-example#preprocessing-data

Can you please explain how/why the distributions of price, size are poisson. Shouldn't they be normal ?

$\endgroup$
6
  • 2
    $\begingroup$ I can't speak to why the tutorial says that house size and price are Poisson distributed, but the Poisson distribution is constrained to non-negative values, which seems consistent with expectations about the size/value of a home. But why Poisson specifically as opposed to any other distribution over non-negative values seems uncertain. Why do you suggest house prices have a normal distribution? $\endgroup$
    – Sycorax
    Nov 5, 2020 at 22:09
  • 3
    $\begingroup$ This example doesn't appear to make any assertions about actual house sizes or prices: it's merely an example to illustrate some techniques. $\endgroup$
    – whuber
    Nov 5, 2020 at 22:19
  • $\begingroup$ @Sycorax Thank you. I was just assuming them to be normal because, there are some houses at very low prices, size and some at very high price, size but most of them are kind of somewhere in the middle. I was assuming the distribution looks like the bell curve if plotted. I may be wrong, please correct in case I am wrong. $\endgroup$
    – tjt
    Nov 6, 2020 at 1:50
  • 1
    $\begingroup$ Normal distributions assign positive probability to all real numbers, which implies some non-zero probability is assigned to negative numbers. Poisson distributions with a large parameter likewise resemble a bell-curve shape, even if they're not exactly normal. In other words, given a large Poisson parameter, the normal approximation to a Poisson can be very good -- but it's still just an approximation. Compare a $P(10^6)$ distribution to a $N(10^6, (10^3)^2)$ distribution, for instance. $\endgroup$
    – Sycorax
    Nov 6, 2020 at 2:08
  • 2
    $\begingroup$ My browser tells me that I've viewed these questions before. If you'd like to draw attention to your questions, you could place bounties on them. $\endgroup$
    – Sycorax
    Nov 6, 2020 at 2:19

1 Answer 1

1
$\begingroup$

Poisson distributions are used when the outcome is the number of events that occur during a defined time interval or the number of objects found in a defined volume (or area). I don't see how house size or price can ever be Poisson. They are limited to be positive values, but that doesn't mean their distribution is Poisson.

$\endgroup$
2
  • $\begingroup$ Thank you. can I assume them to be normally distributed ( in general) based on my comments in the question above. $\endgroup$
    – tjt
    Nov 6, 2020 at 1:53
  • 1
    $\begingroup$ @tjt normal distributions are symmetrical. I doubt the distribution of housing prices is symmetrical. $\endgroup$ Nov 11, 2020 at 15:54

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