1. Paul Murray
  2. PowerBuilder
  3. Thursday, 27 September 2018 03:05 AM UTC

Greetings All,

I would like to know if anyone in this forum has any experience with embedding some type of 'off the shelf' neural network technology in their applications for the purpose of correlating large sets of data with many variables.

If so, I would be very interested in what technology was used, how it was implemented and how you might have set 'goals' or weighting for the processing with respect to predicting how the net would process additional sets of data after having been 'trained'.

I do not have any real experience with this so if some aspect of my question seems silly, it probably is.

Thanks all,

Paul

Paul Murray Accepted Answer Pending Moderation
  1. Tuesday, 2 October 2018 00:36 AM UTC
  2. PowerBuilder
  3. # 1

Thank you, Michael,

I guess I have to update my vernacular to use 'ML'.

Generally, I want to point the ML to a large dataset and have it churn through the numbers given some 'goal' (or otherwise correlate the dataset to an outcome found in the dataset.

Then I would like to be able to possibly use JSON to pass smaller datasets to the ML to have it process those using the data in the larger dataset.

Thanks All,

Paul

Comment
There are no comments made yet.
Michael Kramer Accepted Answer Pending Moderation
  1. Monday, 1 October 2018 23:36 PM UTC
  2. PowerBuilder
  3. # 2

Hey Paul,

Neural networks or machine learning (ML) more generally speaking? No. Integration in general? Plenty. I don't expect such ML packages to provide PowerScript interfaces. However, I would expect REST-based interop, API defined in .NET/.NET Core and/or in JavaScript.

  • REST interfaces are callable from PB 2017 R3
  • C# on .NET Core is PB 2018's server-side programming platform
  • HTML/JS can be embedded in PB using any release within the last decade or so.

Re: HTML/JS embedding

There are many ways to do this. I have a design based on interface constructs that I will present at Elevate 2018. I used it first for embedding web-GIS maps in 20+ years old PB legacy code, typical PFC client/server.

A couple of issues to look out for:

  1. How does your ML model make callbacks to your code?
  2. How do you pass potentially complex data, potentially large volume of data back and forth between the ML model and your PB code?

Enjoy,

/Michael

Comment
There are no comments made yet.
  • Page :
  • 1


There are no replies made for this question yet.
However, you are not allowed to reply to this question.