Results 1 to 3 of 3

Thread: Neural Network learning values effect

  1. #1
    Advanced Tuner f.creek-ranch's Avatar
    Join Date
    May 2018
    Location
    Germany
    Posts
    572

    Neural Network learning values effect

    Hi,

    What happens, if I set learning for NN int the options completely high?

    VE and PRatio learning range from 1 - 1,01 or 100% - 101% for example.

    Will the NN learning be disable and or will the VE tables be used in stead of?

  2. #2
    Advanced Tuner f.creek-ranch's Avatar
    Join Date
    May 2018
    Location
    Germany
    Posts
    572
    Just talking to myself...
    In my opinion the Training values set the ranges for the Network, that will be charged.
    f.e. VE min value for the 5.7 is 0, for the 6.4 is 20,9.
    So the input under 20,9 (ve) from the 6.4 is not considerd..but the other (pratio,cam and other) values will. ???
    Will there be no change at all in the output layer, because ist predetermined from the network, or will the network not be used in this case?

  3. #3
    All inputs to the neural network are normalized between the minimum and maximum value. That means that if you set the PRatio range to 1 - 1.01, all pratio values below 1 will be considered as 1 and all values above 1.01 would be considered as 1.01.

    For output, things are slightly different.
    When training the neural network, the trainer ignores all output values below MinVE. This is done because those values are usually meaningless and removing them will increase the training accuracy for the rest of the neural network range.
    When the neural network is running on the vehicle, the output from the neural network can be below MinVE and MaxVE.