UnfoldedFuncVariable.cc

Go to the documentation of this file.

// -*- C++ -*-

// PLearn (A C++ Machine Learning Library)
// Copyright (C) 1998 Pascal Vincent
// Copyright (C) 1999-2002 Pascal Vincent, Yoshua Bengio, Rejean Ducharme and University of Montreal
// Copyright (C) 2001-2002 Nicolas Chapados, Ichiro Takeuchi, Jean-Sebastien Senecal
// Copyright (C) 2002 Xiangdong Wang, Christian Dorion

// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are met:
// 
//  1. Redistributions of source code must retain the above copyright
//     notice, this list of conditions and the following disclaimer.
// 
//  2. Redistributions in binary form must reproduce the above copyright
//     notice, this list of conditions and the following disclaimer in the
//     documentation and/or other materials provided with the distribution.
// 
//  3. The name of the authors may not be used to endorse or promote
//     products derived from this software without specific prior written
//     permission.
// 
// THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
// IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
// OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN
// NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
// TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
// 
// This file is part of the PLearn library. For more information on the PLearn
// library, go to the PLearn Web site at www.plearn.org


/* *******************************************************      
   * $Id: UnfoldedFuncVariable.cc,v 1.10 2004/04/27 16:05:33 morinf Exp $
   * This file is part of the PLearn library.
   ******************************************************* */

#include "UnfoldedFuncVariable.h"
//#include "PLMPI.h"
//#include "DisplayUtils.h"

namespace PLearn {
using namespace std;



PLEARN_IMPLEMENT_OBJECT(UnfoldedFuncVariable, "Variable that puts in the rows of its output matrix the value\n"
                        "of a Func evaluated on each row of an input matrix.\n",
                        "The input_matrix and output matrix have n_unfold rows. A separate propagation path\n"
                        "is created that maps (using the Func as a template) each input row to each output row.\n"
                        "The parents of this variable include the non-input parents of the Func.\n");

UnfoldedFuncVariable::UnfoldedFuncVariable()
  : transpose(0)
  {}

UnfoldedFuncVariable::UnfoldedFuncVariable(Var inputmatrix, Func the_f, bool the_transpose)
  : inherited(nonInputParentsOfPath(the_f->inputs,the_f->outputs) & inputmatrix, 
                the_transpose ? the_f->outputs[0]->length()*the_f->outputs[0]->width() : inputmatrix->length(),
                the_transpose ? inputmatrix->width() : the_f->outputs[0]->length()*the_f->outputs[0]->width()),
    input_matrix(inputmatrix), 
    f(the_f),
    transpose(the_transpose)
{
  build();
}

void UnfoldedFuncVariable::build()
{
  inherited::build();
  build_();
}

void UnfoldedFuncVariable::build_()
{
    if (f) {
        if(f->outputs.size()!=1)
            PLERROR("In UnfoldedFuncVariable: function must have a single variable output (maybe you can vconcat the vars into a single one prior to calling sumOf, if this is really what you want)");
        f->inputs.setDontBpropHere(true);
        int n_unfold = transpose ? input_matrix->width() : input_matrix->length();
        inputs.resize(n_unfold);
        outputs.resize(n_unfold);
        f_paths.resize(n_unfold);
        for (int i=0;i<n_unfold;i++)
        {
            inputs[i].resize(f->inputs.size());
            for (int j = 0; j < f->inputs.size(); j++) {
                inputs[i][j] = Var(f->inputs[j]->length(), f->inputs[j]->width());
            }
            outputs[i] = f(inputs[i])[0];
            f_paths[i] = propagationPath(inputs[i],outputs[i]);
        }
    }
}

void UnfoldedFuncVariable::declareOptions(OptionList& ol)
{
  declareOption(ol, "f", &UnfoldedFuncVariable::f, OptionBase::buildoption, 
                "    Func that is replicated for each element of the 'bag' taken from the VMat.");

  declareOption(ol, "input_matrix", &UnfoldedFuncVariable::input_matrix, OptionBase::buildoption, 
                "    Var that contains the data, with multiple consecutive rows forming one bag.\n");

  declareOption(ol, "transpose", &UnfoldedFuncVariable::transpose, OptionBase::buildoption, 
                "    If set to 1, then instead puts in the columns of the output matrix the values\n"
                "    of f at the columns of the input matrix.");

  inherited::declareOptions(ol);
}


void UnfoldedFuncVariable::recomputeSize(int& l, int& w) const
{
    if (f && f->outputs.size()) {
        w = f->outputs[0]->length()*f->outputs[0]->width();
        if (transpose) {
            l = w;
            w = input_matrix->width();
        } else {
            l = input_matrix->length();
        }
    } else
        l = w = 0;
}


void UnfoldedFuncVariable::makeDeepCopyFromShallowCopy(map<const void*, void*>& copies)
{
  NaryVariable::makeDeepCopyFromShallowCopy(copies);
  deepCopyField(input_matrix, copies);
  deepCopyField(f, copies);
  deepCopyField(inputs, copies);
  deepCopyField(outputs, copies);
  deepCopyField(f_paths, copies);
}


void UnfoldedFuncVariable::fprop()
{
  int n_unfold = transpose ? input_matrix->width() : input_matrix->length();
  for (int i=0;i<n_unfold;i++) {
    if (transpose) {
      Vec tmp = input_matrix->matValue.column(i).toVecCopy(); // TODO something more efficient
      inputs[i] << tmp;
    } else {
      inputs[i] << input_matrix->matValue(i);
    }
    f_paths[i].fprop();
    if (transpose) {
      matValue.column(i) << outputs[i]->value;
    } else {
      matValue(i) << outputs[i]->value;
    }
  }
}


void UnfoldedFuncVariable::bprop()
{ 
  int n_unfold = transpose ? input_matrix->width() : input_matrix->length();
  for (int i=0;i<n_unfold;i++)
    {
      f_paths[i].clearGradient();
      if (transpose) {
        Vec tmp = matGradient.column(i).toVecCopy(); // TODO more efficient + check while it compiled without tmp = toVecCopy
        outputs[i]->gradient << tmp;
      } else {
        outputs[i]->gradient << matGradient(i);
      }
      f_paths[i].bprop();
    }
}


void UnfoldedFuncVariable::printInfo(bool print_gradient)
{
  int n_unfold = transpose ? input_matrix->width() : input_matrix->length();
  for (int i=0;i<n_unfold;i++)
    f_paths[i].printInfo(print_gradient);
  cout << info() << " : " << getName() << "[" << (void*)this << "]" 
       << "(input_matrix=" << (void*)input_matrix << " ";
  for(int i=0; i<n_unfold; i++) cout << (void*)outputs[i] << " ";
  cout << ") = " << value;
  if (print_gradient) cout << " gradient=" << gradient;
  cout << endl; 
}


} // end of namespace PLearn

---