Model Outputs Consistency

kautuk
Level 2
Model Outputs Consistency

Hi,

I am trying to build a model on Dataiku platform using AutoML inside Lab. If I give same parameters settings to the model multiple times, I am getting a slightly different output each time. I wanted to know if there is a way to get the consistent outputs from the model in each run (similar to set.seed in python).

Thanks,

Kautuk


Operating system used: Chrome

0 Kudos
3 Replies
AlexT
Dataiker

Hi @kautuk ,

When splitting the Train/Test set a random fixed seed is already defined:

Screenshot 2022-09-17 at 13.58.04.png

Can you confirm what settings you are using train/test split?

Are you seeing the same result with an explicit extract from 2 datasets? 

0 Kudos
kautuk
Level 2
Author

Hi @AlexT,

Thanks for the response. I am using ' Explicit extracts from two datasets' Policy and not using train-test split from the same dataset. And, yes, the results are different each time,

0 Kudos
AlexT
Dataiker

Depending on the algorithm type, some variance can happen.

I think it's best to continue looking into this if you via a support ticket.

Please share the model type and screenshots and any sample of data you can share in the support to ilustrate this variance when scored. 

0 Kudos