Start of factor graph port. Things don't work yet, but a lot of syntax updates towards PHP

PHPSkills/TrueSkill/Factors/GaussianFactor.php

@@ -0,0 +1,32 @@
+<?php
+
+namespace Moserware\Skills\TrueSkill\Factors;
+
+abstract class GaussianFactor extends Factor
+{
+    protected function __construct($name)
+    {
+        parent::__construct($name);
+    }
+
+    /// Sends the factor-graph message with and returns the log-normalization constant
+    protected function sendMessageVariable(Message $message, Variable $variable)
+    {
+        $marginal = $variable->getValue();
+        $messageValue = $message->getValue();
+        $logZ = GaussianDistribution::logProductNormalization($marginal, $messageValue);
+        $variable->setValue($marginal*$messageValue);
+        return $logZ;
+    }
+
+    public function createVariableToMessageBinding(Variable $variable)
+    {
+        return parent::createVariableToMessageBinding($variable,
+                                                      new Message(
+                                                          GaussianDistribution::fromPrecisionMean(0, 0),
+                                                          "message from {0} to {1}", $this));
+    }
+}
+
+
+?>

PHPSkills/TrueSkill/Factors/GaussianGreaterThanFactor.php

@@ -0,0 +1,73 @@
+<?php
+namespace Moserware\Skills\TrueSkill\Factors;
+
+/// <summary>
+/// Factor representing a team difference that has exceeded the draw margin.
+/// </summary>
+/// <remarks>See the accompanying math paper for more details.</remarks>
+class GaussianGreaterThanFactor extends GaussianFactor
+{
+    private $_epsilon;
+
+    public function __construct($epsilon, Variable $variable)
+    {
+        parent::_construct("{0} > {1:0.000}");
+        $this->_epsilon = $epsilon;
+        $this->createVariableToMessageBinding($variable);
+    }
+
+    public function getLogNormalization()
+    {
+        $vars = $this->getVariables();
+        $marginal = $vars[0]->getValue();
+        $messages = $this->getMessages();
+        $message = $messages[0]->getValue();
+        $messageFromVariable = GaussianDistribution::divide($marginal, $message);
+        return -GaussianDistribution::logProductNormalization($messageFromVariable, $message)
+               +
+               log(
+                   GaussianDistribution::cumulativeTo(($messageFromVariable->getMean() - $this->_epsilon)/
+                                                     $messageFromVariable->getStandardDeviation()));
+
+    }
+
+    protected function updateMessageVariable(Message $message, Variable $variable)
+    {
+        $oldMarginal = clone $variable->getValue();
+        $oldMessage = clone $message->getValue();
+        $messageFromVar = GaussianDistribution::divide($oldMarginal, $oldMessage);
+
+        $c = $messageFromVar->getPrecision();
+        $d = $messageFromVar->getPrecisionMean();
+
+        $sqrtC = sqrt($c);
+
+        $dOnSqrtC = $d/$sqrtC;
+
+        $epsilsonTimesSqrtC = $this->_epsilon*$sqrtC;
+        $d = $messageFromVar->getPrecisionMean();
+
+        $denom = 1.0 - TruncatedGaussianCorrectionFunctions::vExceedsMargin($dOnSqrtC, $epsilsonTimesSqrtC);
+
+        $newPrecision = $c/$denom;
+        $newPrecisionMean = ($d +
+                             $sqrtC*
+                             TruncatedGaussianCorrectionFunctions::vExceedsMargin($dOnSqrtC, $epsilsonTimesSqrtC))/
+                             $denom;
+
+        $newMarginal = GaussianDistribution::fromPrecisionMean($newPrecisionMean, $newPrecision);
+
+        $newMessage = GaussianDistribution::divide(
+                              GaussianDistribution::multiply($oldMessage, $newMarginal),
+                              $oldMarginal);
+
+        /// Update the message and marginal
+        $message->setValue($newMessage);
+
+        $variable->setValue($newMarginal);
+
+        /// Return the difference in the new marginal
+        return GaussianDistribution::subtract($newMarginal, $oldMarginal);
+    }
+}
+?>

PHPSkills/TrueSkill/Factors/GaussianLikelihoodFactor.php

@@ -0,0 +1,78 @@
+<?php
+
+namespace Moserware\Skills\TrueSkill\Factors;
+
+/// <summary>
+/// Connects two variables and adds uncertainty.
+/// </summary>
+/// <remarks>See the accompanying math paper for more details.</remarks>
+class GaussianLikelihoodFactor extends GaussianFactor
+{
+    private $_precision;
+
+    public function __construct($betaSquared, Variable $variable1, Variable $variable2)
+    {
+        parent::__construct("Likelihood of {0} going to {1}");
+        $this->_precision = 1.0/$betaSquared;
+        $this->createVariableToMessageBinding($variable1);
+        $this->createVariableToMessageBinding($variable2);
+    }
+
+    public function getLogNormalization()
+    {
+        $vars = $this->getVariables();
+        $messages = $this->getMessages();
+
+        return GaussianDistribution::logRatioNormalization(
+                $vars[0]->getValue(),
+                $messages[0]->getValue());
+    }
+
+    private function updateHelper(Message $message1, Message $message2,
+                                  Variable $variable1, Variable $variable2)
+    {
+        $message1Value = clone $message1->getValue();
+        $message2Value = clone $message2->getValue();
+
+        $marginal1 = clone $variable1->getValue();
+        $marginal2 = clone $variable2->getValue();
+
+        $a = $this->_precision/($this->_precision + $marginal2->getPrecision() - $message2Value->getPrecision());
+
+        $newMessage = GaussianDistribution::fromPrecisionMean(
+            $a*($marginal2->getPrecisionMean() - $message2Value->getPrecisionMean()),
+            $a*($marginal2->getPrecision() - $message2Value->getPrecision()));
+
+        $oldMarginalWithoutMessage = GaussianDistribution::divide($marginal1, $message1Value);
+
+        $newMarginal = GaussianDistribution::multiply($oldMarginalWithoutMessage, $newMessage);
+
+        /// Update the message and marginal
+
+        $message1->setValue($newMessage);
+        $variable1->setValue($newMarginal);
+
+        /// Return the difference in the new marginal
+        return GaussianDistribution::subtract($newMarginal, $marginal1);
+    }
+
+    public function updateMessageIndex($messageIndex)
+    {
+        $messages = $this->getMessages();
+        $vars = $this->getVariables();
+
+        switch ($messageIndex)
+        {
+            case 0:
+                return $this->updateHelper($messages[0], $messages[1],
+                                           $vars[0], $vars[1]);
+            case 1:
+                return $this->updateHelper($messages[1], $messages[0],
+                                           $vars[1], $vars[0]);
+            default:
+                throw new Exception();
+        }
+    }
+}
+
+?>

PHPSkills/TrueSkill/Factors/GaussianPriorFactor.php

@@ -0,0 +1,40 @@
+<?php
+
+namespace Moserware\Skills\TrueSkill\Factors;
+
+/// <summary>
+/// Supplies the factor graph with prior information.
+/// </summary>
+/// <remarks>See the accompanying math paper for more details.</remarks>
+class GaussianPriorFactor extends GaussianFactor
+{
+    private $_newMessage;
+
+    public function __construct($mean, $variance, Variable $variable)
+    {
+        parent::__construct("Prior value going to {0}");
+        $this->_newMessage = new GaussianDistribution($mean, sqrt($variance));
+        $this->createVariableToMessageBinding($variable,
+                                               new Message(
+                                                   GaussianDistribution::fromPrecisionMean(0, 0),
+                                                       "message from {0} to {1}",
+                                                   this, variable));
+    }
+
+    protected function updateMessageVariable(Message $message, Variable $variable)
+    {
+        $oldMarginal = clone $variable->getValue();
+        $oldMessage = $message;
+        $newMarginal =
+            GaussianDistribution::fromPrecisionMean(
+                $oldMarginal->getPrecisionMean() + $this->_newMessage->getPrecisionMean() - $oldMessage->getValue()->getPrecisionMean(),
+                $oldMarginal->getPrecision() + $this->_newMessage->getPrecision() - $oldMessage->getValue()->getPrecision());
+
+        $variable->setValue($newMarginal);
+        $message->setValue($this->_newMessage);
+        return GaussianDistribution::subtract($oldMarginal, $newMarginal);
+    }
+}
+
+
+?>

PHPSkills/TrueSkill/Factors/GaussianWeightedSumFactor.php

@@ -0,0 +1,216 @@
+<?php
+
+namespace Moserware\Skills\TrueSkill\Factors;
+
+/// <summary>
+/// Factor that sums together multiple Gaussians.
+/// </summary>
+/// <remarks>See the accompanying math paper for more details.</remarks>
+class GaussianWeightedSumFactor extends GaussianFactor
+{
+    private $_variableIndexOrdersForWeights = array();
+
+    // This following is used for convenience, for example, the first entry is [0, 1, 2]
+    // corresponding to v[0] = a1*v[1] + a2*v[2]
+    private $_weights;
+    private $_weightsSquared;
+
+    public function __construct(Variable $sumVariable, array $variablesToSum, array $variableWeights = null)
+    {
+        parent::__construct($this->createName($sumVariable, $variablesToSum, $variableWeights));
+        $this->_weights = array();
+        $this->_weightsSquared = array();
+
+        // The first weights are a straightforward copy
+        // v_0 = a_1*v_1 + a_2*v_2 + ... + a_n * v_n
+        $this->_weights[0] = array();
+
+        $variableWeightsLength = count($variableWeights);
+
+        for($i = 0; $i < $variableWeightsLength; $i++)
+        {
+            $weight = $variableWeights[$i];
+            $this->_weights[0][$i] = $weight;
+            $this->_weightsSquared[0][i] = $weight * $weight;
+        }
+
+        $variablesToSumLength = count($variablesToSum);
+
+        // 0..n-1
+        for($i = 0; $i < ($variablesToSumLength + 1); $i++)
+        {
+            $this->_variableIndexOrdersForWeights[] = $i;
+        }
+        
+        // The rest move the variables around and divide out the constant.
+        // For example:
+        // v_1 = (-a_2 / a_1) * v_2 + (-a3/a1) * v_3 + ... + (1.0 / a_1) * v_0
+        // By convention, we'll put the v_0 term at the end
+
+        $weightsLength = $variableWeightsLength + 1;
+        for ($weightsIndex = 1; $weightsIndex < $weightsLength; $weightsIndex++)
+        {            
+            $this->_weights[$weightsIndex] = array();
+
+            $variableIndices = array();
+            $variableIndices[] = $weightsIndex;
+
+            $currentWeightsSquared = array();
+            $this->_WeightsSquared[$weightsIndex] = $currentWeightsSquared;
+
+            // keep a single variable to keep track of where we are in the array.
+            // This is helpful since we skip over one of the spots
+            $currentDestinationWeightIndex = 0;
+
+            for ($currentWeightSourceIndex = 0;
+                 $currentWeightSourceIndex < $variableWeights.Length;
+                 $currentWeightSourceIndex++)
+            {
+                if ($currentWeightSourceIndex == ($weightsIndex - 1))
+                {
+                    continue;
+                }
+
+                $currentWeight = (-$variableWeights[$currentWeightSourceIndex]/$variableWeights[$weightsIndex - 1]);
+
+                if ($variableWeights[$weightsIndex - 1] == 0)
+                {
+                    // HACK: Getting around division by zero
+                    $currentWeight = 0;
+                }
+
+                $currentWeights[$currentDestinationWeightIndex] = $currentWeight;
+                $currentWeightsSquared[$currentDestinationWeightIndex] = $currentWeight*currentWeight;
+
+                $variableIndices[$currentDestinationWeightIndex + 1] = $currentWeightSourceIndex + 1;
+                $currentDestinationWeightIndex++;
+            }
+
+            // And the final one
+            $finalWeight = 1.0/$variableWeights[$weightsIndex - 1];
+
+            if ($variableWeights[$weightsIndex - 1] == 0)
+            {
+                // HACK: Getting around division by zero
+                $finalWeight = 0;
+            }
+            $currentWeights[$currentDestinationWeightIndex] = finalWeight;
+            $currentWeightsSquared[currentDestinationWeightIndex] = finalWeight*finalWeight;
+            $variableIndices[$variableIndices.Length - 1] = 0;
+            $this->_variableIndexOrdersForWeights[] = $variableIndices;
+        }
+
+        $this->createVariableToMessageBinding($sumVariable);
+
+        foreach ($variablesToSum as $currentVariable)
+        {
+            $this->createVariableToMessageBinding($currentVariable);
+        }
+    }
+
+    public function getLogNormalization()
+    {
+        $vars = $this->getVariables();
+        $messages = $this->getMessages();
+
+        $result = 0.0;
+
+        // We start at 1 since offset 0 has the sum
+        $varCount = count($vars);
+        for ($i = 1; $i < $varCount; $i++)
+        {
+            $result += GaussianDistribution::logRatioNormalization($vars[i]->getValue(), $messages[i]->getValue());
+        }
+
+        return $result;
+    }
+
+    private function updateHelper(array $weights, array $weightsSquared,
+                                  array $messages,
+                                  array $variables)
+    {
+        // Potentially look at http://mathworld.wolfram.com/NormalSumDistribution.html for clues as
+        // to what it's doing
+
+        $messages = $this->getMessages();
+        $message0 = clone $messages[0]->getValue();
+        $marginal0 = clone $variables[0]->getValue();
+
+        // The math works out so that 1/newPrecision = sum of a_i^2 /marginalsWithoutMessages[i]
+        $inverseOfNewPrecisionSum = 0.0;
+        $anotherInverseOfNewPrecisionSum = 0.0;
+        $weightedMeanSum = 0.0;
+        $anotherWeightedMeanSum = 0.0;
+
+        $weightsSquaredLength = count($weightsSquared);
+
+        for ($i = 0; $i < $weightsSquaredLength; $i++)
+        {
+            // These flow directly from the paper
+
+            $inverseOfNewPrecisionSum += $weightsSquared[i]/
+                                         ($variables[$i + 1]->getValue()->getPrecision() - $messages[$i + 1]->getValue()->getPrecision());
+
+            $diff = GaussianDistribution::divide($variables[$i + 1]->getValue(), $messages[$i + 1]->getValue());
+            $anotherInverseOfNewPrecisionSum += $weightsSquared[i]/$diff->getPrecision();
+
+            $weightedMeanSum += $weights[i]
+                                *
+                                ($variables[$i + 1]->getValue()->getPrecisionMean() - $messages[$i + 1]->getValue()->getPrecisionMean())
+                                /
+                                ($variables[$i + 1]->getValue()->getPrecision() - $messages[$i + 1]->getValue()->getPrecision());
+
+            $anotherWeightedMeanSum += $weights[$i]*$diff->getPrecisionMean()/$diff->getPrecision();
+        }
+
+        $newPrecision = 1.0/$inverseOfNewPrecisionSum;
+        $anotherNewPrecision = 1.0/$anotherInverseOfNewPrecisionSum;
+
+        $newPrecisionMean = $newPrecision*$weightedMeanSum;
+        $anotherNewPrecisionMean = $anotherNewPrecision*$anotherWeightedMeanSum;
+
+        $newMessage = GaussianDistribution::fromPrecisionMean($newPrecisionMean, $newPrecision);
+        $oldMarginalWithoutMessage = GaussianDistribution::divide($marginal0, $message0);
+
+        $newMarginal = GaussianDistribution::multiply($oldMarginalWithoutMessage, $newMessage);
+
+        /// Update the message and marginal
+
+        $messages[0]->setValue($newMessage);
+        $variables[0]->setValue($newMarginal);
+
+        /// Return the difference in the new marginal
+        $finalDiff = GaussianDistribution::subtract($newMarginal, $marginal0);
+        return $finalDiff;
+    }
+
+    public function updateMessageIndex($messageIndex)
+    {
+        $allMessages = $this->getMessages();
+        $allVariables = $this->getVariables();
+
+        Guard::argumentIsValidIndex($messageIndex, count($allMessages),"messageIndex");
+
+        $updatedMessages = array();
+        $updatedVariables = array();
+
+        $indicesToUse = $this->_variableIndexOrdersForWeights[$messageIndex];
+
+        // The tricky part here is that we have to put the messages and variables in the same
+        // order as the weights. Thankfully, the weights and messages share the same index numbers,
+        // so we just need to make sure they're consistent
+        $allMessagesCount = count($allMessages);
+        for ($i = 0; i < $allMessagesCount; $i++)
+        {
+            $updatedMessages[] =$allMessages[$indicesToUse[$i]];
+            $updatedVariables[] = $allVariables[$indicesToUse[$i]];
+        }
+
+        return updateHelper($this->_weights[$messageIndex],
+                            $this->_weightsSquared[$messageIndex],
+                            $updatedMessages,
+                            $updatedVariables);
+    }
+}
+
+?>

PHPSkills/TrueSkill/Factors/GaussianWithinFactor.php

@@ -0,0 +1,72 @@
+<?php
+
+namespace Moserware\Skills\TrueSkill\Factors;
+
+/// <summary>
+/// Factor representing a team difference that has not exceeded the draw margin.
+/// </summary>
+/// <remarks>See the accompanying math paper for more details.</remarks>
+class GaussianWithinFactor extends GaussianFactor
+{
+    private $_epsilon;
+
+    public function __construct($epsilon, Variable $variable)
+    {
+        $this->_epsilon = $epsilon;
+        $this->createVariableToMessageBinding($variable);
+    }
+
+    public function getLogNormalization()
+    {
+        $variables = $this->getVariables();
+        $marginal = $variables[0]->getValue();
+
+        $messages = $this->getMessages();
+        $message = $messages[0]->getValue();
+        $messageFromVariable = GaussianDistribution::divide($marginal, $message);
+        $mean = $messageFromVariable->getMean();
+        $std = $messageFromVariable->getStandardDeviation();
+        $z = GaussianDistribution::cumulativeTo(($this->_epsilon - $mean)/$std)
+             -
+             GaussianDistribution::cumulativeTo((-$this->_epsilon - $mean)/$std);
+
+        return -GaussianDistribution::logProductNormalization($messageFromVariable, $message) + log($z);
+    }
+
+    protected function updateMessage(Message $message, Variable $variable)
+    {
+        $oldMarginal = clone $variable->getValue();
+        $oldMessage = clone $message->getValue();
+        $messageFromVariable = GaussianDistribution::divide($oldMarginal, $oldMessage);
+
+        $c = $messageFromVariable->getPrecision();
+        $d = $messageFromVariable->getPrecisionMean();
+
+        $sqrtC = sqrt($c);
+        $dOnSqrtC = $d/$sqrtC;
+
+        $epsilonTimesSqrtC = $this->_epsilon*$sqrtC;
+        $d = $messageFromVariable->getPrecisionMean();
+
+        $denominator = 1.0 - TruncatedGaussianCorrectionFunctions::wWithinMargin($dOnSqrtC, $epsilonTimesSqrtC);
+        $newPrecision = $c/$denominator;
+        $newPrecisionMean = ($d +
+                                   $sqrtC*
+                                   TruncatedGaussianCorrectionFunctions::vWithinMargin($dOnSqrtC, $epsilonTimesSqrtC))/
+                                  $denominator;
+
+        $newMarginal = GaussianDistribution::fromPrecisionMean($newPrecisionMean, $newPrecision);
+        $newMessage = GaussianDistribution::divide(
+                        GaussianDistribution::multiply($oldMessage, $newMarginal),
+                        $oldMarginal);
+
+        /// Update the message and marginal
+        $message->setValue($newMessage);
+        $variable->setValue($newMarginal);
+
+        /// Return the difference in the new marginal
+        return GaussianDistribution::subtract($newMarginal, $oldMarginal);
+    }
+}
+
+?>