Question
How to import external Adaptive models(Python)
Hi,
We have data analystics created those models in Python as the external model, we like to import those Adaptive models into our application as a part the campaign. How to import those external models into Pega Marketing?
Did anyone tried this scenario.
Regards,
Anandh
***Moderator Edit: Vidyaranjan | Updated Categories***
Hi Anandh,
"We have data analystics created those models in Python as the external model". I'm not understanding this exactly. But there are a couple of ways to incorporate models built outside of Pega.
- Data level integration - i.e. you score the models then just use the results.
- Translating the models into scorecards (or other constructs) in Pega. Not all models can easily be translated but for simple cases this may be feasible.
- Importing through PMML. This is a model interchange format supported by Pega and by a large group of other vendors. Most ML tools support exporting to PMML. If this is possible, it's the preferred way.
Not sure what you mean by adaptive models you created in Python. Can you elaborate?
Adaptive models in Pega cannot be imported into. They are self-learning. What you could do is import the external model (one of the methods above) and have an Adaptive model shadow that. So initially you use the externally created model, but the Adaptive model will learn as it accumulates evidence. You should monitor the performance of the Adaptive model, and at some point you may decide to cut over.
PMML models in 7.3.1: https://docs-previous.pega.com/supported-pmml-model-types-pega-731
Hi Anandh,
"We have data analystics created those models in Python as the external model". I'm not understanding this exactly. But there are a couple of ways to incorporate models built outside of Pega.
- Data level integration - i.e. you score the models then just use the results.
- Translating the models into scorecards (or other constructs) in Pega. Not all models can easily be translated but for simple cases this may be feasible.
- Importing through PMML. This is a model interchange format supported by Pega and by a large group of other vendors. Most ML tools support exporting to PMML. If this is possible, it's the preferred way.
Not sure what you mean by adaptive models you created in Python. Can you elaborate?
Adaptive models in Pega cannot be imported into. They are self-learning. What you could do is import the external model (one of the methods above) and have an Adaptive model shadow that. So initially you use the externally created model, but the Adaptive model will learn as it accumulates evidence. You should monitor the performance of the Adaptive model, and at some point you may decide to cut over.
PMML models in 7.3.1: https://docs-previous.pega.com/supported-pmml-model-types-pega-731
Mixing PMML, Adaptive and other types of models: https://mesh.pega.com/docs/DOC-18836
Thanks.
We are looking a python script adaptive learning which will tell us what weights/parameters to use in next wave/campaign strategy.
How to import those script into Pega? Will it possible to import those kind of python script into Pega.
Regards,
Anandh
Just to add another alternative - if your tools for whatever reason don't export to PMML, write a scripts yourself that generates PMML and use the standard PMML import. Ofcourse a bit of a hassle, but that's why we developed strong native modeling capabilities in Pega (adaptive and static).
Hi Anandh,
What kind of python scripts are you referring to? Do they produce a model? If so, the preferred way is to export this as PMML - the most popular model exchange format today.
Or do they just produce "weights/parameters", but to what? Scorecard weights? If so you could perhaps export the data and use decision strategies.
You can not import python directly.
Regards
Otto