Discover all of the brand-new features and improvements to existing capabilities in the Dataiku 11.3 updateLET'S GO

Run a recipe from a scenerio

Solved!
Erlebacher
Level 4
Run a recipe from a scenerio

I am writing a recipe to retrain a custom model created in a Python recipe. This recipe creates a pickle file that contains the model, which is then read by a downstream model that performs predictions. I know how to build a dataset from a scenario, but a pickle file is not a dataset. How can I force the recipe that creates the model to run? Thanks.


Operating system used: Mac Ventura

0 Kudos
1 Solution
Marlan

Hi @Erlebacher,

Our approach in this situation is to write the pickle file to a DSS Folder and then build the Folder (which is one of the Build options).

Marlan

View solution in original post

4 Replies
ZachM
Dataiker

Hi @Erlebacher,

Scenarios can also build models. If you add a "Build / Train" step to your scenario, there will be an option to choose a model to build:

670319A5-4039-4A1D-B9E9-7D804EB1FBAB_1_201_a.jpeg

This should run the recipe that creates the model.

Be sure to set the Build mode to "Build only this dataset" or "Force-rebuild dataset and dependencies" if you want the model to be built every time the scenario is run.

Thanks,

Zach

0 Kudos
Erlebacher
Level 4
Author

Hi @ZachM ,

When you say `model`, are you referring to a model as defined by Dataiku, created through the lab interface? I simply have a Python recipe that creates a model that is saved in a pickle file. I solved the problem by creating a Dataset with the contents ('Successful update', along with date/time). I have the scenario clear the data set and run it. This works fine.

0 Kudos
ZachM
Dataiker

Hi @Erlebacher,

Sorry, I was talking about a Dataiku model. I didn't realize you were using a Python recipe.

Your solution sounds good!

Thanks,

Zach

0 Kudos
Marlan

Hi @Erlebacher,

Our approach in this situation is to write the pickle file to a DSS Folder and then build the Folder (which is one of the Build options).

Marlan