search

amoghskulkarni / WebGME-MOCA

A domain-specific modeling language for running analyses on manufacturing process models
2 stars 0 forks source link

Make a data-driven component possible to be integrated in analyses #10

Closed amoghskulkarni closed 6 years ago

amoghskulkarni commented 6 years ago

Currently, only analytical components can be used in optimization and DoE analysis. It should be also possible to use a DD component in these analyses, similar to a normal analytical component. This can be done by wrapping the trained model inside a OpenMDAO component and provide interfaces to it.

amoghskulkarni commented 6 years ago

2ed7d643a2205a8ce062809361de19afa52453c5 Initial testing done for integrating the data-driven models. Currently, the scenarios in which they can work are pretty limited - the LearningAlgorithm element must contain only 1 feature. Multiple features are supported in training step but not in the integration step as of now.

amoghskulkarni commented 6 years ago

9eecb3e55b697070c667dca9f77ecd908c9f193a Initially tried implementing the multiple feature support by finding out "logical connections" (as seen in the above commit).

6be3e1310a089e92e6f9162772c57b9e1822d3c3 Later on, implemented multiple feature support by caching the parameter list along with the trained model - as the parameters are needed to be passed to the regressor in the same order as in the learning phase.

The single feature example works as earlier, so the change is non-breaking. Needs to be tested with multiple feature scenario.

amoghskulkarni commented 6 years ago

Closing this as of now, creating a new issue for testing.