cancel
Showing results for 
Search instead for 
Did you mean: 

Predictive Maintenance case

Predictive Maintenance case

Hi

 

I want to predict the remaining useful life with DataRobot.

I found the blog about NASA case related to regression.

 

https://www.datarobot.com/blog/predictive-maintenance-of-nasa-turbofans-using-datarobot/

 

Could I get some more detail how to give options and segmentation with NASA datasets in this case?

I want to follow the work road in this case.

 

Thank you.

Labels (2)
1 Solution

Accepted Solutions

So yes, it is not a TS model, and target is generated outside the project, but it can be created in Spark SQL in AI Catalog
There were experiments with TS for this use case, but it involves much more data preparation to avoid target leakage

View solution in original post

0 Kudos
6 Replies

Hi!

May you clarify your question? In article it is said that model uses group partitioning, what options you are interested in?

0 Kudos
jas0n
Data Scientist
Data Scientist

Hi, 

I haven't had the opportunity to use this data set myself yet, but if you enable Segmented mode on the Data tab when configuring a TS project DataRobot will detect valid segment choices for you. There are additional options when you click "Show Advanced Options". 

If you can let us know some more details we can be more specific.

Thanks,
Jason

0 Kudos

Hi

 

I want to know specific settings.

The blog just shows results.

What was the target and which feature should be an identifier?

I need a kind of tutorial.

 

Thank you.

0 Kudos

Hi!

 

Target is computed separately - based on the maximum number of cycles until failure for each record of data. Metric is RMSE, using Group partitioning. No other additional settings found.

0 Kudos

Hi~

 

Thank you for the response.

 

Is this regression case TS model or not?

There are feature engineering strategies with rolling features automatically generated by DataRobot.

 

"The best performing (non-blender) model is a Ridge Regressor which now has RMSE scores of 27.6178, 32.2755, and 24.8812 for validation, cross-validation, and holdout (respectively) for the model trained on ~64% of the data.3 This model utilizes a reduced feature list (33 features) that has been automatically generated by DataRobot.4 Of these 33 most impactful features, all of them are engineered features."

 

If I choose TS model, I cannot use the advanced option for group partitioning.

So, I asked whether I need to choose an identifier instead of group partitioning.

 

If it is not TS model, where were the rolling features generated? by myself?

 

Thank you for the answer to my specific questions.

 

0 Kudos

So yes, it is not a TS model, and target is generated outside the project, but it can be created in Spark SQL in AI Catalog
There were experiments with TS for this use case, but it involves much more data preparation to avoid target leakage

0 Kudos