using different feature subsets for individual models

[ifserge]

Is it possible to use different feature subsets for individual models? If yes, how would it can be configured?

[goldentom42]

Hi ifserge, Stacknet does not currently support different dataset for individual model. What you could do is :

• feed Stacknet with different datasets and different parameter config,
• get the level 1 output for each of these configurations
• aggregate all level1 outputs in one file (for train and test data)
• run a final Stacknet config with the level 1 aggregate Hope this helps

[kaz-Anova]

Thank you for the assist @goldentom42

Actually now there is a way to do it through python.

You should have a look at the python generic model .

If look inside lib/python , for example PythonGenericRegressor0.py in line 148and line 168 you can add your own pre-processing of data. You could do a sub-selection or anything you want.

If you choose this method , bear in mind:

1. You are limited to only python algorithms you have available in your system.
2. Your data is loaded as sparse data. If you want to chance them to numpy you need to do X.toarray()
3. you need to apply the same preprocessing in both train and test (e.g. lines 148,168)

Hope that helps.

[goldentom42]

@kaz-Anova, My bad I totally forgot about this great feature... Congrats for your nice LB position in Zillow's competition!