duke@435: /*
duke@435:  * Copyright 2002-2005 Sun Microsystems, Inc.  All Rights Reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
duke@435:  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435:  * CA 95054 USA or visit www.sun.com if you need additional information or
duke@435:  * have any questions.
duke@435:  *
duke@435:  */
duke@435: 
duke@435: # include "incls/_precompiled.incl"
duke@435: # include "incls/_gcUtil.cpp.incl"
duke@435: 
duke@435: // Catch-all file for utility classes
duke@435: 
duke@435: float AdaptiveWeightedAverage::compute_adaptive_average(float new_sample,
duke@435:                                                         float average) {
duke@435:   // We smooth the samples by not using weight() directly until we've
duke@435:   // had enough data to make it meaningful. We'd like the first weight
duke@435:   // used to be 1, the second to be 1/2, etc until we have 100/weight
duke@435:   // samples.
duke@435:   unsigned count_weight = 100/count();
duke@435:   unsigned adaptive_weight = (MAX2(weight(), count_weight));
duke@435: 
duke@435:   float new_avg = exp_avg(average, new_sample, adaptive_weight);
duke@435: 
duke@435:   return new_avg;
duke@435: }
duke@435: 
duke@435: void AdaptiveWeightedAverage::sample(float new_sample) {
duke@435:   increment_count();
duke@435:   assert(count() != 0,
duke@435:          "Wraparound -- history would be incorrectly discarded");
duke@435: 
duke@435:   // Compute the new weighted average
duke@435:   float new_avg = compute_adaptive_average(new_sample, average());
duke@435:   set_average(new_avg);
duke@435:   _last_sample = new_sample;
duke@435: }
duke@435: 
duke@435: void AdaptivePaddedAverage::sample(float new_sample) {
duke@435:   // Compute our parent classes sample information
duke@435:   AdaptiveWeightedAverage::sample(new_sample);
duke@435: 
duke@435:   // Now compute the deviation and the new padded sample
duke@435:   float new_avg = average();
duke@435:   float new_dev = compute_adaptive_average(fabsd(new_sample - new_avg),
duke@435:                                            deviation());
duke@435:   set_deviation(new_dev);
duke@435:   set_padded_average(new_avg + padding() * new_dev);
duke@435:   _last_sample = new_sample;
duke@435: }
duke@435: 
duke@435: void AdaptivePaddedNoZeroDevAverage::sample(float new_sample) {
duke@435:   // Compute our parent classes sample information
duke@435:   AdaptiveWeightedAverage::sample(new_sample);
duke@435: 
duke@435:   float new_avg = average();
duke@435:   if (new_sample != 0) {
duke@435:     // We only create a new deviation if the sample is non-zero
duke@435:     float new_dev = compute_adaptive_average(fabsd(new_sample - new_avg),
duke@435:                                              deviation());
duke@435: 
duke@435:     set_deviation(new_dev);
duke@435:   }
duke@435:   set_padded_average(new_avg + padding() * deviation());
duke@435:   _last_sample = new_sample;
duke@435: }
duke@435: 
duke@435: LinearLeastSquareFit::LinearLeastSquareFit(unsigned weight) :
duke@435:   _sum_x(0), _sum_y(0), _sum_xy(0),
duke@435:   _mean_x(weight), _mean_y(weight) {}
duke@435: 
duke@435: void LinearLeastSquareFit::update(double x, double y) {
duke@435:   _sum_x = _sum_x + x;
duke@435:   _sum_x_squared = _sum_x_squared + x * x;
duke@435:   _sum_y = _sum_y + y;
duke@435:   _sum_xy = _sum_xy + x * y;
duke@435:   _mean_x.sample(x);
duke@435:   _mean_y.sample(y);
duke@435:   assert(_mean_x.count() == _mean_y.count(), "Incorrect count");
duke@435:   if ( _mean_x.count() > 1 ) {
duke@435:     double slope_denominator;
duke@435:     slope_denominator = (_mean_x.count() * _sum_x_squared - _sum_x * _sum_x);
duke@435:     // Some tolerance should be injected here.  A denominator that is
duke@435:     // nearly 0 should be avoided.
duke@435: 
duke@435:     if (slope_denominator != 0.0) {
duke@435:       double slope_numerator;
duke@435:       slope_numerator = (_mean_x.count() * _sum_xy - _sum_x * _sum_y);
duke@435:       _slope = slope_numerator / slope_denominator;
duke@435: 
duke@435:       // The _mean_y and _mean_x are decaying averages and can
duke@435:       // be used to discount earlier data.  If they are used,
duke@435:       // first consider whether all the quantities should be
duke@435:       // kept as decaying averages.
duke@435:       // _intercept = _mean_y.average() - _slope * _mean_x.average();
duke@435:       _intercept = (_sum_y - _slope * _sum_x) / ((double) _mean_x.count());
duke@435:     }
duke@435:   }
duke@435: }
duke@435: 
duke@435: double LinearLeastSquareFit::y(double x) {
duke@435:   double new_y;
duke@435: 
duke@435:   if ( _mean_x.count() > 1 ) {
duke@435:     new_y = (_intercept + _slope * x);
duke@435:     return new_y;
duke@435:   } else {
duke@435:     return _mean_y.average();
duke@435:   }
duke@435: }
duke@435: 
duke@435: // Both decrement_will_decrease() and increment_will_decrease() return
duke@435: // true for a slope of 0.  That is because a change is necessary before
duke@435: // a slope can be calculated and a 0 slope will, in general, indicate
duke@435: // that no calculation of the slope has yet been done.  Returning true
duke@435: // for a slope equal to 0 reflects the intuitive expectation of the
duke@435: // dependence on the slope.  Don't use the complement of these functions
duke@435: // since that untuitive expectation is not built into the complement.
duke@435: bool LinearLeastSquareFit::decrement_will_decrease() {
duke@435:   return (_slope >= 0.00);
duke@435: }
duke@435: 
duke@435: bool LinearLeastSquareFit::increment_will_decrease() {
duke@435:   return (_slope <= 0.00);
duke@435: }