Timeline for What is use of Tweedie or poisson loss/objective function in XGboost and Deep learning models
Current License: CC BY-SA 4.0
11 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 13, 2023 at 19:47 | history | edited | Matthew Drury | CC BY-SA 4.0 |
edited body
|
| Oct 21, 2020 at 15:45 | comment | added | Matthew Drury | $X$ is all the features you're using in your model. Thanks for the kind words, happy to hear you've found my answers useful! | |
| Oct 21, 2020 at 14:18 | comment | added | tjt | Thank you. This is one of main question I had form long and you cleared it. One last question. In the Y/X of answer, Y is the claims amount, can you please explain what is X here. Is it any of the variables in the dataset (i.e Y/X for any variable in dataset will be gamma) or is X some selective or important variables or atleast 1 variable in dataset. I learnt a lot from your several stack exchange answers for various questions in the forum, they are very clear. Thanks again. | |
| Oct 21, 2020 at 5:19 | comment | added | Matthew Drury | There's no singular reason. The best case scenario is that you have some theory that guides the selection. Like, radioactive decay, there is clear cut physical theory that leads to a poisson distribution as the choice of appropriate model. It's rare that a situation is so clear cut, so the selection is guided by some mixture of theory, intuition, and mathematical tractability / appropriateness. It's not a very good idea to base the selection on what the histogram of $Y$ looks like, because of the conditional vs. marginal distinction I called out above. But you have to compromise sometimes. | |
| Oct 21, 2020 at 3:26 | comment | added | tjt | Thank you. This is a question in general - so if a variable shape looks like certain distribution (and assumptions of variable similar with distribution's assumptions), can we model the variable with the distribution ? Reason for asking this is, I see often that some variables are modeled as certain distributions and I can't figure out why, they chose the particular distribution. When I check the intuitive explanation behind the selected distribution, it's (what the distribution is intuitively) not much related to the variable (like the sale amount to gamma distribution relation here). | |
| Oct 21, 2020 at 3:20 | vote | accept | tjt | ||
| Oct 21, 2020 at 3:11 | comment | added | Matthew Drury | @tjt I edited in some answers to those questions. | |
| Oct 21, 2020 at 3:10 | history | edited | Matthew Drury | CC BY-SA 4.0 |
added 590 characters in body
|
| Oct 21, 2020 at 2:33 | comment | added | tjt | Thank you so much for the reply, it is very clear how tweedie is obtained from poisson and gamma distribution. couple of questions. 1. Is the sales forecasting same as the claims example - where each sale is poisson and sale amount is gamma distributed? 2. poisson distribution can be described as the time for 1 event of n number of events to occur (say here claims). And as per my knowledge (please correct if wrong), gamma distribution is like the time between 2 poisson events. Can you please explain, how/why claim amount follows gamma distribution. | |
| Oct 19, 2020 at 19:31 | history | edited | Matthew Drury | CC BY-SA 4.0 |
deleted 4 characters in body
|
| Oct 19, 2020 at 18:45 | history | answered | Matthew Drury | CC BY-SA 4.0 |