Model files for illustrative purposes only:
Note: additional information regarding the deployment of the custom models are included in the sample notebook.
There are two ways custom models can be deployed to work with Maximo –Predict as outlined below:
1. BRYOM (Bring Your Own Model): Typically a black box model, or physics based model, source code is not available, signature is available
•TBD (future release)
2. BDYOM (Build Your Own Model): This is a fully developed, readily deployable on Maximo –Predict. This involves using PMI wrapper types, with an option of deploying on either AS or WML. AS involves UI support and WML deployment does not provide UI support
•Custom Transformer, Estimator, Pipeline
–Documentation (this document)
– Two datasets are provided with this sample (to be used with the notebooks provided with this documentation)
•dtx_sample_for_failure_probability.csv : Sample dataset to predict the failure along with probability
•dtx_sample_for_time_to_failure.csv: Sample dataset to estimate the time left before a failure happens
– Load the data using the instructions and code in the notebook “DTX_1_CustomLoader.ipynb”
– Customize the wrapper types and the overridable functions to suit your dataset as illustrated in the notebook “DTX_2_FailurePredictionCustomModel.ipynb”
The steps outlined in this document can be used to build custom models using one of the following two approaches:
1.Generic Custom Model: Build a custom model that is deployable on IBM Watson Machine Learning, and register it with MAS –Predict runtime
•The DTX_GenCustom_*** artifacts included with this model illustrate this approach
•This approach lets the provider build any generic model as long as it is deployable on IBM Watson Machine Learning
•There is currently no UI support for models built and registered using this approach. The results of model invocation (scoring) will not be displayed on the MAS –Predict dashboard
2. Template Based Custom Model: Build a custom model based on one of the MAS –Predict predefined use case templates
•This approach is not illustrated in this version, and will be added at a later time
•Models built using this approach will display the results on the UI
Steps involved in building Generic Custom Model
Building Generic custom model involves the following broad steps:
–Train a classifier or regressor or other form of estimator as needed for use case using your own training data.
–Once trained and evaluated, deploy the trained model in IBM Watson Machine Learning.
–An example of the two steps outlined above is illustrated in the notebook DTX_GenCustom_Failure_Probability_Training.ipynb
–Load the sensor data and failure data in the MAS Datalakeand the EAM system (Maximo). This is illustrated in the notebook DTX_GenCustom_Load_Data.ipynb
–Register the model with MAS –Predict infrastructure. This is illustrated in the notebook DTX_GenCustom_Failure_Probability_Registration.ipynb
–Logon into the MAS –Predict dashboard and activate the registered model,and schedule it.
–Ensure the data is flowing into the datalake
–Use the MAS –Predict APIs to retrieve the results periodically to display on your own client dashboard