LB Booster
General >> General Board >> neural net : xor
http://lbb.conforums.com/index.cgi?board=general&action=display&num=1484748428
neural net : xor
Post by bluatigro on Jan 18th, 2017, 1:07pm


error :
JB reports a syntaks error
but not where

Code:
'' Author:        John McCullock
'' Date:        12-11-2005
'' Description:    Backpropagation XOR Example 2.
'' Sources: Dr Phil Brierley, www.philbrierley.com
'' translated from c++ to JB/LB by bluatigro

global numInputs : numInputs = 3       '' Input nodes, plus the bias input.
global numPatterns : numPatterns = 4     '' Input patterns for XOR experiment.

global numHidden : numHidden = 4
global numEpochs : numEpochs = 200
global LR.IH : LR.IH = 0.7       '' Learning rate, input to hidden weights.
global LR.HO : LR.HO = 0.07      '' Learning rate, hidden to output weights.

global patNum : patNum = 0
global errThisPat : errThisPat = 0.0
global outPred : outPred = 0.0                  '' "Expected" output values.
global RMSerror : RMSerror = 0.0                 '' Root Mean Squared error.

dim hiddenVal( numHidden )         '' Hidden node outputs.

dim weightsIH( numInputs , numHidden )  '' Input to Hidden weights.
dim weightsHO( numHidden )          '' Hidden to Output weights.

dim trainInputs( numPatterns , numInputs )
dim trainOutput( numPatterns )         '' "Actual" output values.



    randomize timer  '' Seed the generator with system time.

    call initWeights

    call initData

    '' Train the network
    for j = 0 to numEpochs

        for i = 0 to numPatterns

            ''Select a pattern at random.
            patNum = rnd(0) * numPatterns

            ''Calculate the output and error for this pattern.
            call calcNet

            ''Adjust network weights.
            call WeightChangesHO
            call WeightChangesIH
        next i

        call calcOverallError

        ''Display the overall network error after each epoch
        print "epoch = " + str$(j) + " RMS Error = " + str(RMSerror)

    next j
    ''Training has finished.

    call displayResults

input "[ pres return ]" ; in$
end

function tanh( x as double ) as double
  return ( 1 - exp( -x * 2 ) ) / ( 1 + exp( -x * 2 ) )
end function

sub initWeights
'' Initialize weights to random values.

    for j = 0 to numHidden

        weightsHO(j) = ( rnd(0) - 0.5 ) / 2
        for i = 0 to numInputs

            weightsIH(i,j) = ( rnd(0) - 0.5 ) / 5
            print "Weight = " + str$( weightsIH(i,j) )
        next i
    next j
end sub

sub initData
    '' The data here is the XOR data which has been rescaled to
    '' the range -1 to 1.

    '' An extra input value of 1 is also added to act as the bias.

    '' The output must lie in the range -1 to 1.

    trainInputs(0,0)   =  1
    trainInputs(0,1)   = -1
    trainInputs(0,2)   =  1   '' Bias
    trainOutput(0)     =  1

    trainInputs(1,0)   = -1
    trainInputs(1,1)   =  1
    trainInputs(1,2)   =  1  '' Bias
    trainOutput(1)     =  1

    trainInputs(2,0)   =  1
    trainInputs(2,1)   =  1
    trainInputs(2,2)   =  1  '' Bias
    trainOutput(2)     = -1

    trainInputs(3,0)   = -1
    trainInputs(3,1)   = -1
    trainInputs(3,2)   =  1  '' Bias
    trainOutput(3)     = -1
end sub

sub calcNet
'' Calculates values for Hidden and Output nodes.

    for i = 0 to numHidden
      hiddenVal(i) = 0.0

        for j = 0 to numInputs
            hiddenVal(i) = hiddenVal(i)+(trainInputs(patNum,j) * weightsIH(j,i) )
        next j

        hiddenVal(i) = tanh( hiddenVal( i ) )
    next i

    outPred = 0.0

    for i = 0 to numHidden
        outPred = outPred + hiddenVal(i) * weightsHO(i)
    next i
    ''Calculate the error: "Expected" - "Actual"
    errThisPat = outPred - trainOutput( patNum )
end sub

sub WeightChangesHO
''Adjust the Hidden to Output weights.
    for k = 0 to numHidden
        dim as double weightChange = LR.HO * errThisPat * hiddenVal(k)
        weightsHO(k) = weightsHO(k) - weightChange

        '' Regularization of the output weights.
        if (weightsHO(k) < -5) then
            weightsHO(k) = -5
        end if
        if (weightsHO(k) > 5) then
            weightsHO(k) = 5
        end if
    next k
end sub

sub WeightChangesIH
'' Adjust the Input to Hidden weights.
    for i = 0 to numHidden
        for k = 0 to numInputs
            x = 1 - (hiddenVal(i) * hiddenVal(i))
            x = x * weightsHO(i) * errThisPat * LR.IH
            x = x * trainInputs(patNum,k)
            weightChange = x
            weightsIH(k,i) = weightsIH(k,i) - weightChange
        next k
    next i
end sub

sub calcOverallError
    RMSerror = 0.0
    for i = 0 to numPatterns
         patNum = i
         call calcNet
         RMSerror = RMSerror + (errThisPat * errThisPat)
    next i
    RMSerror = RMSerror / numPatterns
    RMSerror = sqr(RMSerror)
end sub

sub displayResults
    for i = 0 to numPatterns
        patNum = i
        call calcNet
        print "pat = " + str$( patNum + 1 ) _
        + " actual = " + str$( trainOutput(patNum) ) _
        + " neural model = " + str$( outPred )
    next i
end sub
Re: neural net : xor
Post by Richard Russell on Jan 18th, 2017, 5:44pm

on Jan 18th, 2017, 1:07pm, bluatigro wrote:
JB reports a syntaks error,but not where

LBB throws an error at this function definition, which looks like it hasn't been fully translated from the original C++:

Code:
function tanh( x as double ) as double

If you delete the spurious 'as double' declarations it should help.

Richard.
Re: neural net : xor
Post by bluatigro on Jan 19th, 2017, 08:48am

@ richard :
thanks for that
on other forums they fount the remaining error's

the error's came because i translated in freebasic first

update :
it works !!

Code:
'' Author:        John McCullock
'' Date:        12-11-2005
'' Description:    Backpropagation XOR Example 2.
'' Sources: Dr Phil Brierley, www.philbrierley.com
'' translated from c++ to JB/LB by bluatigro
global numInputs : numInputs = 3       '' Input nodes, plus the bias input.
global numPatterns : numPatterns = 4     '' Input patterns for XOR experiment.
global numHidden : numHidden = 4
global numEpochs : numEpochs = 200
global LR.IH : LR.IH = 0.7       '' Learning rate, input to hidden weights.
global LR.HO : LR.HO = 0.07      '' Learning rate, hidden to output weights.
global patNum : patNum = 0
global errThisPat : errThisPat = 0.0
global outPred : outPred = 0.0                  '' "Expected" output values.
global RMSerror : RMSerror = 0.0                 '' Root Mean Squared error.
dim hiddenVal( numHidden )         '' Hidden node outputs.
dim weightsIH( numInputs , numHidden )  '' Input to Hidden weights.dim weightsHO( numHidden )          '' Hidden to Output weights.
dim trainInputs( numPatterns , numInputs )
dim trainOutput( numPatterns )         '' "Actual" output values.

call initWeights
call initData
'' Train the network
for j = 0 to numEpochs
  for i = 0 to numPatterns
    ''Select a pattern at random.
    patNum = i ''rnd(0) * numPatterns
    ''Calculate the output and error for this pattern.
    call calcNet
    ''Adjust network weights.
    call WeightChangesHO
    call WeightChangesIH
  next i
  call calcOverallError
  ''Display the overall network error after each epoch
  print "epoch = " + str$(j) + " RMS Error = " + str$(RMSerror)
next j
''Training has finished.
call displayResults
input "[ pres return ]" ; in$
end

function tanh( x )
  tanh = ( 1 - exp( 0-x * 2 ) ) / ( 1 + exp( 0-x * 2 ) )
end function

sub initWeights
'' Initialize weights to random values.
    for j = 0 to numHidden
        weightsHO(j) = ( rnd(0) - 0.5 ) / 2
        for i = 0 to numInputs
            weightsIH(i,j) = ( rnd(0) - 0.5 ) / 5
            print "Weight = " + str$( weightsIH(i,j) )
        next i
    next j
end sub

sub initData
    '' The data here is the XOR data which has been rescaled to
    '' the range -1 to 1.
    '' An extra input value of 1 is also added to act as the bias.
    '' The output must lie in the range -1 to 1.
    trainInputs(0,0)   =  1
    trainInputs(0,1)   = -1
    trainInputs(0,2)   =  1   '' Bias
    trainOutput(0)     =  1
    trainInputs(1,0)   = -1
    trainInputs(1,1)   =  1
    trainInputs(1,2)   =  1  '' Bias
    trainOutput(1)     =  1
    trainInputs(2,0)   =  1
    trainInputs(2,1)   =  1
    trainInputs(2,2)   =  1  '' Bias
    trainOutput(2)     = -1
    trainInputs(3,0)   = -1
    trainInputs(3,1)   = -1
    trainInputs(3,2)   =  1  '' Bias
    trainOutput(3)     = -1
end sub

sub calcNet
'' Calculates values for Hidden and Output nodes.
    for i = 0 to numHidden
        hiddenVal(i) = 0.0
        for j = 0 to numInputs
            hiddenVal(i) = hiddenVal(i)+(trainInputs(patNum,j) * weightsIH(j,i) )
        next j
        hiddenVal(i) = tanh( hiddenVal( i ) )
    next i
    outPred = 0.0
    for i = 0 to numHidden
        outPred = outPred + hiddenVal(i) * weightsHO(i)
    next i
    ''Calculate the error: "Expected" - "Actual"
    errThisPat = outPred - trainOutput( patNum )
end sub

sub WeightChangesHO
''Adjust the Hidden to Output weights.
    for k = 0 to numHidden
        weightChange = LR.HO * errThisPat * hiddenVal(k)
        weightsHO(k) = weightsHO(k) - weightChange
        '' Regularization of the output weights.
        if (weightsHO(k) < -5) then
            weightsHO(k) = -5
        end if
        if (weightsHO(k) > 5) then
            weightsHO(k) = 5
        end if
    next k
end sub

sub WeightChangesIH
'' Adjust the Input to Hidden weights.
    for i = 0 to numHidden
        for k = 0 to numInputs
            x = 1 - (hiddenVal(i) * hiddenVal(i))
            x = x * weightsHO(i) * errThisPat * LR.IH
            x = x * trainInputs(patNum,k)
            weightChange = x
            weightsIH(k,i) = weightsIH(k,i) - weightChange
        next k
    next i
end sub

sub calcOverallError
    RMSerror = 0.0
    for i = 0 to numPatterns
         patNum = i
         call calcNet
         RMSerror = RMSerror + (errThisPat * errThisPat)
    next i
    RMSerror = RMSerror / numPatterns
    RMSerror = sqr(RMSerror)
end sub

sub displayResults
    for i = 0 to numPatterns
        patNum = i
        call calcNet
        print "pat = " + str$( patNum + 1 ) _
        + " actual = " + str$( trainOutput(patNum) ) _
        + " neural model = " + str$( outPred )
    next i
end sub