1

I am using survival regression for predicting in how many days user will make next purchase.

Survival Regression predict function is giving results similar to as shown below -

Output - 

{
  "coefficients": [
    -0.2633608588194104, 
     0.22152319227842276
  ], 
  "intercept": 2.6380946151040012, 
  "prediction": 5.718979487634966, 
  "quantiles": [
    1.1603238947151593, 
    4.995456010274735
  ], 
  "scale": 1.5472345574364683 
}

Quantiles are basically cutpoints dividing the range of a probability distribution into contiguous intervals with equal probabilities.

From wikipedia page - https://en.wikipedia.org/wiki/Quantile

1) What is quantile in this context?

2) I have data like this - 

u1 viewed    i1 on 18 June 2017
u1 purchased i1 on 20 June 2017
u1 purchased i2 on 23 June 2017
u1 purchased i3 on 30 June 2017
u2 viewed    i1 on 10 Mar  2017
u2 purchased i1 on 11 Mar  2017
...

There are multiple purchase records for each user. .

I have considered "Purchase done" to be death event but my users(such as u1, u2) are still alive and making purchases.

What should be the death event as users are doing multiple transactions in their life time.

Thanks

r4sn4
  • 117
  • 5
  • 14
  • 2
    How did you train your model ? Those quantiles are computed considering the input quantiles that you have given to AFT – eliasah Jul 20 '17 at 06:56
  • @eliasah I have added more detail to my question. Could you tell me about this particular use case? – r4sn4 Jul 20 '17 at 12:01
  • This still doesn't answer my question. I'd like to see the code which you have used to create your model – eliasah Jul 20 '17 at 12:03
  • I am using prediction IO Survival regression template. https://github.com/goliasz/pio-template-sr https://predictionio.incubator.apache.org/gallery/template-gallery/ – r4sn4 Jul 20 '17 at 12:53
  • It appears that you are asking for further exposition of code and of the concepts surrounding survival analysis, as well as advice about how to approach problems with recurrent events. Seems this would be more appropriate for CrossValidated.com, since you are not actually posing a well-defined coding specification. – IRTFM Jul 24 '17 at 15:44

0 Answers0