Data type issue in windows analytics

Highlighted
UserBird Dataiker
Dataiker
Data type issue in windows analytics

Hi,



I'm having trouble with the visual recipe window. I have this dataset :





And with this dataset, I want to calculate the mean of "sepal_length" (for example) by "species" and then calculate the difference between this mean and the value of the initial variable for each row. If i understand correctly, i can do this with the visual recip window. 



But, when i try (to calculate the mean by species), i have this problem :





I know i can do this easily with SQL, R... but i'm trying to learn all the possibilies of Dataiku DSS.



Maybe I just don't understand what window visual recipe is for. 



Thanks,



 



 



 

0 Kudos
4 Replies
cperdigou Dataiker
Dataiker
Re: Data type issue in windows analytics
Hello,

Your issue is probably related to the fact that your decimals use comma instead of dots. To be honest it's very weird that you could reach this situation, with doubles stored with commas.

You should go to the dataset settings and redetect types.
0 Kudos
Hugo_Dupeux
Level 1
Re: Data type issue in windows analytics
Thanks for your time.

In fact I tried with dot and also to let the type as string like dataiku stored the variable in a first place. I put the comma to fix the problem but it was unsuccesful.

Maybe my mistake was not here but in the construction of the window recipe. I did like here :
https://www.dataiku.com/learn/guide/visual/window/using-the-window-recipe.html

If you know how to calculate the mean of "sepal_length" by "species" with wondow recipe, can you show me every step ?
0 Kudos
cperdigou Dataiker
Dataiker
Re: Data type issue in windows analytics
Not sure I follow. Could you:
Redetect types, it will store as strings
Create a prepare recipe, replace ',' with '.'
Run the recipe
See that storage is now double
Retry creating the window recipe
0 Kudos
Hugo_Dupeux
Level 1
Re: Data type issue in windows analytics
This is exactely what I've done. But I encounter the same problem
0 Kudos
Labels (1)