predictSequence.frink

Download or view predictSequence.frink in plain text format


use functionUtils.frink

/** This program contains routines to predict a sequence.  These are generally
    achieved by finding the differences or quotients of terms, but there are
    other techniques that, say, try to force a polynomial fit onto the data.
*/

fit[list, numAdditionalTerms=1, debug=false] :=
{
   level = 0
   do
   {
      [result, perfectFit, nextRowPoly, symbolic, polyFunc] = polynomialFit[list, numAdditionalTerms-level, debug, true]
      if perfectFit
         return polyFunc
      else
      {
         println["nextRowPoly is $nextRowPoly"]
         [result, perfectFit, nextRowQuot, symbolic, quotFunc] = quotientFit[nextRowPoly, numAdditionalTerms-level-1, debug, true]
         if perfectFit
            return polyFunc[quotFunc[x]]
      }
      
      [result, perfectFit, nextRowQuot, symbolic, quotFunc] = quotientFit[list, numAdditionalTerms-level, debug, true]
      if perfectFit
         return quotFunc
      else
      {
         println["nextRowQuot is $nextRowQuot"]
         [result, perfectFit, nextRowPoly, symbolic, polyFunc] = polynomialFit[nextRowQuot, numAdditionalTerms-level-1, debug, true]
         if perfectFit
            return quotFunc[polyFunc[x]]
      }

      level = level + 1
   } while level < numAdditionalTerms

   return "match not found"
}

/** This is one version that forces a polynomial fit onto the data.  See
    http://extremelearning.com.au/a-simple-formula-for-sequences-and-series/

    although it's not clear from the article that it's fitting a polynomial
    onto the data and that it won't work for geometric terms, or a Fibonacci
    sequence, or whatever.
*/

polynomialFit[list, numAdditionalTerms=10, debug=false, returnExtra=false] :=
{
   list2 = deepCopy[list]
   first = new array
   first@0 = list2@0
   
   origlen = length[list]
   var nextRow
   var symbolic
   level = 0

   LEVEL:
   while (len = length[list2]) >= 2
   {
      diffs = new array
      allZero = true
      for i=0 to len-2
      {
         diff = list2@(i+1) - list2@i
         diffs@i = diff
         if (diff != 0)
            allZero = false
      }

      if debug
         println[diffs]
      
      first.push[diffs@0]
      list2 = diffs
      if returnExtra and level==0
         nextRow = diffs
      
      if allZero == true or len <= 2
      {
         if debug
            println["first is " + first]
         result = new array
         for n = 0 to origlen + numAdditionalTerms - 1
         {
            Tn = 0
            symbolic = 0
            for i = rangeOf[first]
            {
               Tn = Tn + first@i * binomial[n, i]
               symbolic = symbolic + first@i * binomialSymbolic[n,i]
            }
            result.push[Tn]
            if debug
               println["Symbolic is $symbolic"]
         }
         if returnExtra == false
            return result
         else
            return [result, allZero, nextRow, symbolic, toFunction[noEval[n], symbolic]]
       }
       level = level + 1
   }
}

/** This tries the quotient of terms to try and predict a series.
*/

quotientFit[list, numAdditionalTerms=10, debug=false, returnExtra=false] :=
{
   list2 = deepCopy[list]
   first = new array
   first@0 = list2@0
   
   difftri = new array
   difftri.push[list2]
   
   origlen = length[list]

   level:
   while (len = length[list2]) >= 2
   {
      diffs = new array
      allone = true
      for i=0 to len-2
      {
         diff = list2@(i+1) / list2@i
         diffs@i = diff
         if (diff != 1)
            allone = false
      }

      difftri.push[diffs]
      
      if debug
         println[diffs]
      
      first.push[diffs@0]
      list2 = diffs
      
      if allone == true or len <= 2
      {
         if debug
            println["first is " + first]
         if debug
            println["difftri is\n" + join["\n", difftri]]
         result = deepCopy[list]
         for n = 1 to numAdditionalTerms
         {
            // copy last entry in difftri
            lastRow = difftri@(length[difftri] - 1)
            lastRow.push[lastRow@(length[lastRow]-1)]

//            if debug
//               println["New sorta difftri is $difftri"]
            for i=length[difftri]-2 to 0 step -1
            {
               row     = difftri@i
               nextRow = difftri@(i+1)
               row.push[row@(length[row]-1) * nextRow@(length[nextRow]-1)]
            }

            if debug
               println["New difftri is " + difftri]
         }
        
         if returnExtra == false
            return difftri@0
         else
         {
            symbolic = list@0 * (list@1/list@0) ^ noEval[n]
            return [difftri@0, allone, difftri@1, symbolic, toFunction[noEval[n], symbolic]]
         }
      }
   }
}

binomialSymbolic[n,k] :=
{
   if (k<=0 or k>=n)
      return 1
   
   if (n - k) > k
      k = (n-k)

   product = 1 / (n-k)!
   
   for i = 0 to n-k-1
      product = product * (noEval[n] - i)
   return product 
}

toFunction[symbol, symbolic] :=
{
   return constructExpression["AnonymousFunction", [[makeSymbol[symbol]], symbolic]]
}


Download or view predictSequence.frink in plain text format


This is a program written in the programming language Frink.
For more information, view the Frink Documentation or see More Sample Frink Programs.

Alan Eliasen was born 19944 days, 11 hours, 57 minutes ago.