How to get Variable Importance from Model

[lugow]

Hi,

I want to automatically consume the data on the "Variable Importance" area under the "Interpretation" section of a trained model. I see that I can manually export this data, but what I am trying to do is to get this data in a dataset or something similar where I can build pipelines on top of it.

Is that possible?

[Alex_Combessie · Accepted solution]

Hi,

Please find below a piece of code to retrieve variable importance raw data for a trained model in an Analysis:

import dataiku
import pandas as pd

client = dataiku.api_client()
project = client.get_project(dataiku.default_project_key())
analysis = project.get_analysis("rGRd5qWg")
ml_task = analysis.get_ml_task("fm97lWNq")
trained_model_ids = ml_task.get_trained_models_ids()

trained_model_detail = ml_task.get_trained_model_details(trained_model_ids[0])
feature_importance = trained_model_detail.get_raw().get("iperf").get("rawImportance")
feature_importance_df = pd.DataFrame(feature_importance)

Note that you can get the analysis_id, ml_task_id and trained_model_id from the URL of the page of the model. Or you can get that using our API: https://doc.dataiku.com/dss/latest/python-api/rest-api-client/ml.html#exploration-of-results

Hope it helps,

Alex

[lugow]

This is awesome!!. Thanks!.

[omri17]

Hi Alex,

This is really awesome.
I saw you referred to the "get_raw()" method. I didn't saw any reference for it in the docs. Is it?
How, for example, i can get regression coefficients in a similar approach?


Thanks,

[kkaminsky]

I would also like to know the answer to this

[ClemenceB]

Hi @kkaminsky,

"get_raw()" is a method that allows you to get all the trained model details as a python dict.

To get regression coefficients you can use the following piece of code:

import dataiku
import pandas as pd

client = dataiku.api_client()
project = client.get_project(dataiku.default_project_key())
analysis = project.get_analysis(analysis_id)
ml_task = analysis.get_ml_task(ml_task_id)
trained_model_ids = ml_task.get_trained_models_ids()

trained_model_detail = ml_task.get_trained_model_details(trained_model_ids[1])
regression_coefficients = trained_model_detail.get_raw().get('iperf').get('lmCoefficients')
regression_coefficients = pd.DataFrame(regression_coefficients)

If the regression_coefficients table is empty, make sure the model you're calling is a regression:

trained_model_detail.get_raw().get('modeling').get('algorithm')

Let me know if it helps.

Clémence