private bool IsCompletedTermConsistentWithNextTerm(NonTerminalObject completedTerm, NonTerminalObject nextTerm,
out NonTerminalStack intersection)
{
intersection = null;
if (nextTerm == null)
{
return(false);
}
//the above case happens where the rule to be continued is already completed at the start column (next term = null).
//this can only happen in a single case: that rule was an epsilon rule.
if (completedTerm.NonTerminal != nextTerm.NonTerminal)
{
return(false);
}
if (nextTerm.Stack == null && completedTerm.Stack == null)
{
return(true);
}
var intersectAlgo = new StackGraphIntersect();
intersection = intersectAlgo.Execute(nextTerm.Stack, completedTerm.Stack);
return(intersection != null);
}
public NonTerminalObject(NonTerminalObject otherNonTerminalObject)
{
NonTerminal = otherNonTerminalObject.NonTerminal;
if (otherNonTerminalObject.Stack != null)
{
Stack = otherNonTerminalObject.Stack; //shallow copy
}
}
public Rule(Rule otherRule)
{
Name = new NonTerminalObject(otherRule.Name);
Production = otherRule.Production.Select(nonterminal => new NonTerminalObject(nonterminal)).ToArray();
HeadPosition = otherRule.HeadPosition;
ComplementPosition = otherRule.ComplementPosition;
Number = otherRule.Number;
Occurrences = otherRule.Occurrences;
}
public Rule(int occurrences, string name, string[] prod, int headPos = 0, int compPos = 1, int num = -1)
{
Name = new NonTerminalObject(name);
if (prod != null)
{
Production = prod.Select(nonterminal => new NonTerminalObject(nonterminal)).ToArray();
}
HeadPosition = headPos;
ComplementPosition = compPos;
Number = num;
Occurrences = occurrences;
}
private RuleConsistentWithDerivation IsPredictedRuleConsistentWithCurrentDerivation(
NonTerminalObject currObject, Rule rule)
{
//TODO: change later to skip -all- rules whose derivation leads to the empty string.
//I.e, A-> B.C , C -> D E. D -> epsilon, E -> epsilon. C itself is not an epsilon rule.
//1. states that are the result of a spontenous dot shift (due to nullable production)
//have already been added to the agendas in Column.Add()
if (rule.IsEpsilonRule() && Grammar.nullableProductions.ContainsKey(currObject))
{
return(RuleConsistentWithDerivation.SkipGeneration);
}
//2. if current stack is empty but predicted stack is not, mismatch - do not predict this rule.
if (currObject.IsStackEmpty() && !rule.IsInitialOrDotStack())
{
return(RuleConsistentWithDerivation.RuleInconsistent);
}
if (rule.IsInitialRule())
{
var complementPositionObject = rule.Production[rule.ComplementPosition];
//3. if current stack is not empty, but the complement position does not allow for stacks (POS),
//mismatch - do not predict this rule.
if (!currObject.IsStackEmpty() && Grammar.IsPOS(complementPositionObject.NonTerminal))
{
return(RuleConsistentWithDerivation.RuleInconsistent);
}
}
else
{
//4. if tops of the stacks do not match, continue. e.g created rule PP[PP] -> epsilon, current object: PP[NP].
if (rule.Name.Stack.Peek() != "." &&
!currObject.Stack.GetListOfTopSymbols().Contains(rule.Name.Stack.Peek()))
{
return(RuleConsistentWithDerivation.RuleInconsistent);
}
}
return(RuleConsistentWithDerivation.RuleConsistent);
}
private void Predict(Column col, List <Rule> ruleList, State state, NonTerminalObject currObject)
{
//if (generator)
//{
// var rule = Grammar.GetRandomRuleForAGivenLHS(currObject.NonTerminal, Grammar.Rules[currObject.NonTerminal]);
// ruleList = new List<Rule> { rule };
//}
//var preds = GenerateSetOfPredictions();
if (generator)
{
Rule rule = null;
var l = Grammar.Rules[currObject.NonTerminal];
var feasibleRules = new List <Rule>();
//when generating, predict only rules that terminate the derivation successfully.
foreach (var candidate in l)
{
var c = IsPredictedRuleConsistentWithCurrentDerivation(currObject, candidate);
if (c == RuleConsistentWithDerivation.SkipGeneration)
{
return;
}
//temporary: do not push another symbol if the current stack already contains a symbol
//in other words: allow only one symbol.
var complementPositionObject = candidate.Production[candidate.ComplementPosition];
var isPushRule = !complementPositionObject.IsStackEmpty() &&
complementPositionObject.Stack.Top != Grammar.Epsilon;
if (!currObject.IsStackEmpty() && isPushRule)
{
continue;
}
if (c == RuleConsistentWithDerivation.RuleConsistent)
{
feasibleRules.Add(candidate);
}
}
if (feasibleRules.Count == 0) //no feasible rules to predict
{
return;
}
rule = Grammar.GetRandomRuleForAGivenLHS(currObject.NonTerminal, feasibleRules);
ruleList = new List <Rule> {
rule
};
}
foreach (var rule in ruleList)
{
if (IsPredictedRuleConsistentWithCurrentDerivation(currObject, rule) !=
RuleConsistentWithDerivation.RuleConsistent)
{
continue;
}
//prepare new rule based on the stack information contained in the current state
//and based on the predicted rule.
var createdRule = new Rule(rule);
//if the rule is not a stack manipulating rule,
if (rule.IsInitialRule())
{
var complementPositionObject = createdRule.Production[createdRule.ComplementPosition];
//the stack of the LHS of the created rule is the stack of the current object:
createdRule.Name = currObject;
//copy the stack to the complement position.
complementPositionObject.Stack = currObject.Stack;
}
else
{
//create left hand side of new rule.
createdRule.Name.Stack = currObject.Stack;
createdRule.Name.NonTerminal = currObject.NonTerminal;
//create right hand side of new rule.
NonTerminalStack contentOfDot;
if (rule.Name.Stack.Peek() == ".")
{
contentOfDot = currObject.Stack; //e.g. A[..]
}
else
{
contentOfDot = currObject.Stack.GetPrefixListStackObjectOfGivenTop(rule.Name.Stack.Peek());
}
//e.g A[..X]
for (var i = 0; i < rule.Production.Length; i++)
{
var s = rule.Production[i].Stack;
if (s != null)
{
if (s.Peek() == ".")
{
createdRule.Production[i].Stack = contentOfDot; // e.g, A[..] pop rule.
}
else if (s.PrefixList == null)
{
createdRule.Production[i].Stack = s; //e.g. A[X] //secondary constituent.
}
else
{
createdRule.Production[i].Stack = new NonTerminalStack(s.Peek(), contentOfDot);
}
//e.g. A[..X] - push rule.
//calculate the new weight of the top of the stack from the weights of its sons.
//if (createdRule.Production[i].Stack != null)
// createdRule.Production[i].Stack.Weight = createdRule.Production[i].Stack.PrefixList != null ? createdRule.Production[i].Stack.PrefixList.Sum(x => x.Weight) : 1;
}
}
}
var newState = new State(createdRule, 0, col, null)
{
LogProbability = ruleLogProbabilities[rule.Number]
};
if (newState.LogProbability < 0)
{
throw new Exception("wrong probability");
}
var added = col.AddState(newState, ParsingOperation.Predict);
if (Debug)
{
Console.WriteLine("{0} & {1} & {2} & Predicted from State {3}, added: {4}\\\\", newState.StateNumber,
newState,
col.Index, state.StateNumber, added);
}
}
}
private HashSet <NonTerminalObject> GenerateSetOfPredictions(int maxElementsInStack = 2)
{
Queue <NonTerminalObject> queue = new Queue <NonTerminalObject>();
HashSet <NonTerminalObject> visitedNonTerminalObjects = new HashSet <NonTerminalObject>(new NonTerminalObjectStackComparer());
NonTerminalObject o = null;
foreach (var moveable in Grammar.Moveables)
{
o = new NonTerminalObject(moveable);
o.Stack = new NonTerminalStack(moveable);
queue.Enqueue(o);
}
while (queue.Any())
{
var rhs = queue.Dequeue();
visitedNonTerminalObjects.Add(rhs);
var rulesforRHS = Grammar.RHSDictionary[rhs.NonTerminal].ToList();
foreach (var item in rulesforRHS)
{
o = null;
var rule = Grammar.ruleNumberDictionary[item.Item1];
int RHSPosition = item.Item2;
if (rule.Name.NonTerminal == Grammar.StartSymbol)
{
continue;
}
if (rule.IsInitialRule())
{
if (rule.ComplementPosition == RHSPosition)
{
o = new NonTerminalObject(rule.Name.NonTerminal);
o.Stack = new NonTerminalStack(rhs.Stack);
if (!visitedNonTerminalObjects.Contains(o))
{
queue.Enqueue(o);
}
}
}
else
{
NonTerminalStack contentOfDot;
var s = rule.Production[RHSPosition].Stack;
if (s != null)
{
if (s.Peek() == ".")
{
contentOfDot = rhs.Stack;
}
else if (s.PrefixList != null)
{
contentOfDot = rhs.Stack.GetPrefixListStackObjectOfGivenTop(rhs.Stack.Peek());
}
else
{
continue;
}
//assumption: if this is a secondary constituent (i.e, s.PrefixList == null), does not participate in the sharing of stacks,
//the LHS will be handled through the primary constituent (the distinguished descendant).
o = new NonTerminalObject(rule.Name.NonTerminal);
if (rule.Name.Stack.Peek() == ".")
{
if (contentOfDot != null)
{
o.Stack = new NonTerminalStack(contentOfDot);
}
}
else
{
if (contentOfDot == null || contentOfDot.Depth() < maxElementsInStack)
{
o.Stack = new NonTerminalStack(rule.Name.Stack.Peek(), contentOfDot);
}
}
if (o.Stack != null && !visitedNonTerminalObjects.Contains(o))
{
queue.Enqueue(o);
}
}
}
}
}
return(visitedNonTerminalObjects);
}