Details

    • Type: Improvement
    • Status: Resolved
    • Priority: Major
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: None
    • Component/s: None

      Description

      Productize a Neural-Network Bundle that has been parallelized to work efficiently on HPCC, based on the NeuralNetwork module in ecl-ml.

      Parallelism to be achieved as follows:

      • FP: Replicate the model and distribute the data
      • BP: Take small (#trainingdatarecs / #nodesbootstraps) of the training data and distribute to NN's on each cluster-node.  Train the nodes independently and aggregate the results ala RandomForest.  Based on: https://www.hindawi.com/journals/cin/2016/2842780/

      Also include:

      • Multi-criterion early-stopping
        • Fully-converged (within epsilon)
        • No forward progress (deltaW < threshold)
      • Momentum factor to greatly reduce training iterations (reduces by tenfold)

      Provide production level:

      • Testing
      • External Doc
      • Internal Doc
      • Performance Optimization

        Attachments

          Activity

            People

            • Assignee:
              rdev Roger Dev
              Reporter:
              rdev Roger Dev
            • Votes:
              0 Vote for this issue
              Watchers:
              1 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: