Hi Rasna,

Folks here tend to respond to things they know about… I guess we don't have a 
lot of experts for Survival Regression… I'll do my best to reply:


Are you using the Survival Regression template from the PredictionIO site?

If so, we can see the exact algorithm in use is AFTSurvivalRegression:
  
https://github.com/goliasz/pio-template-sr/blob/master/src/main/scala/SRAlgorithm.scala#L47

Looking at the docs for that algo,
  
https://spark.apache.org/docs/latest/ml-classification-regression.html#survival-regression
…the coefficients, intercept and scale are based on the model's fit to the 
training data. Once the PredictionIO engine is trained, these will be constant 
until trained with a different data set.

The quantiles and prediction are the answers for each specific query 
(transform).

If you're seeking a deeper understanding of the prediction, I'm sorry but I do 
not have that expertise!

*Mars

( <> .. <> )

> On Jul 27, 2017, at 02:51, Rasna Tomar <[email protected]> wrote:
> 
> Why there is no support for templates other than universal recommender.??
> 
> On Wed, Jul 19, 2017 at 3:20 PM, Rasna Tomar <[email protected]> wrote:
> Hi All
> 
> 
> I am using survival regression for predicting whether user will purchase in 
> next few days or not.
> 
> I am getting results similar to as shown below - 
> 
> Sample query -
> curl -i -X POST http://localhost:8000/queries.json
>  -H "Content-Type: application/json" -d '{"features":[1.560,-0.605]}'
> 
> 
> Output - 
> {
>   "coefficients": [
>     -0.2633608588194104, 
>     0.22152319227842276
>   ], 
>   "intercept": 2.6380946151040012, 
>   "prediction": 5.718979487634966, 
>   "quantiles": [
>     1.1603238947151593, 
>     4.995456010274735
>   ], 
>   "scale": 1.5472345574364683 
> 
> } 
> 
> 
> For each user I am getting same values coefficients, intercept and scale, but 
> Quantile and prediction values are different?
> What is the meaning of quantile and prediction here?
> 
> Thanks
> 
> 
> 
> 
> 

Reply via email to