Examples with Expression com.sri.ai.expresso.api.Expression used on opensource projects

Example 61 with Expression

use of com.sri.ai.expresso.api.Expression in project aic-expresso by aic-sri-international.

the class TupleRewriterTest method testTupleEqualitySimplification.

@Test
public void testTupleEqualitySimplification() {
    Simplifier tupleEqualitySimplifier = new TupleEqualitySimplifier();
    Assert.assertEquals(parse("A1 = B1 and A2 = B2 and A3 = B3"), tupleEqualitySimplifier.apply(parse("(A1, A2, A3) = (B1, B2, B3)"), context));
    Expression expr = parse("(A1, A2) = (B1, B2, B3)");
    Assert.assertTrue(expr == tupleEqualitySimplifier.apply(expr, context));
    expr = parse("A1 = B1");
    Assert.assertTrue(expr == tupleEqualitySimplifier.apply(expr, context));
}

Example 62 with Expression

use of com.sri.ai.expresso.api.Expression in project aic-praise by aic-sri-international.

the class RandomConditionalPotentialExpressionGenerator method main.

/**
	 * Generate Random HOGMv1 models based on given command line arguments.
	 * 
	 * @param args
	 *        pass '--help' to see description of expected program arguments.
	 */
public static void main(String[] args) {
    try (GeneratorArgs genArgs = getArgs(args)) {
        // Output the categorical sorts
        for (int c = 0; c < genArgs.equalityTheories.length; c++) {
            TheoryTypeEqualityArgs equalityTheoryArgs = genArgs.equalityTheories[c];
            genArgs.out.append(HOGMSortDeclaration.FUNCTOR_SORT_DECLARATION);
            genArgs.out.append(" ");
            genArgs.out.append(getEqualityCategoricalName(c));
            genArgs.out.append(" : ");
            genArgs.out.append("" + equalityTheoryArgs.getSizeOfCategory());
            for (int u = 0; u < equalityTheoryArgs.getNumberOfUniquelyNamedConstantsInCategory(); u++) {
                genArgs.out.append(", ");
                genArgs.out.append(genArgs.potentialExpressionGenerator.getCategoryUniquelyNamedConstantName(c, u));
            }
            genArgs.out.println(";");
        }
        if (genArgs.equalityTheories.length > 0) {
            // Place a blank line after any categorical sort declarations
            genArgs.out.println();
        }
        // Output the random variable information for propositional theories
        for (int i = 0; i < genArgs.propositionalTheories.length; i++) {
            TheoryTypePropositionalArgs propositionalArgs = genArgs.propositionalTheories[i];
            for (int v = 0; v < propositionalArgs.getNumberVariables(); v++) {
                genArgs.out.append("random ");
                genArgs.out.append(genArgs.potentialExpressionGenerator.getPropositionalVariableNameFor(i, v));
                genArgs.out.append(" : ");
                genArgs.out.append(HOGMSortDeclaration.IN_BUILT_BOOLEAN.getName().toString());
                genArgs.out.println(";");
            }
        }
        // Output the random variable information for equality theories
        for (int i = 0; i < genArgs.equalityTheories.length; i++) {
            TheoryTypeEqualityArgs equalityArgs = genArgs.equalityTheories[i];
            for (int v = 0; v < equalityArgs.getNumberVariables(); v++) {
                genArgs.out.append("random ");
                genArgs.out.append(genArgs.potentialExpressionGenerator.getEqualityVariableNameFor(i, v));
                genArgs.out.append(" : ");
                genArgs.out.append(getEqualityCategoricalName(i));
                genArgs.out.println(";");
            }
        }
        // Output the random variable information for the inequality theories
        for (int i = 0; i < genArgs.inequalityTheories.length; i++) {
            TheoryTypeInequalityArgs inequalityArgs = genArgs.inequalityTheories[i];
            for (int v = 0; v < inequalityArgs.getNumberVariables(); v++) {
                genArgs.out.append("random ");
                genArgs.out.append(genArgs.potentialExpressionGenerator.getInequalityVariableNameFor(i, v));
                genArgs.out.append(" : ");
                genArgs.out.append("" + inequalityArgs.getStartIntervalInclusive());
                genArgs.out.append("..");
                genArgs.out.append("" + inequalityArgs.getEndIntervalInclusive());
                genArgs.out.println(";");
            }
        }
        genArgs.out.println();
        for (int i = 0; i < genArgs.numberPotentials; i++) {
            Expression conditional = genArgs.potentialExpressionGenerator.nextPotentialExpression();
            // Ensure we have variables in the conditional
            while (Expressions.freeVariables(conditional, new TrueContext()).size() == 0) {
                conditional = genArgs.potentialExpressionGenerator.nextPotentialExpression();
            }
            genArgs.out.append(conditional.toString());
            genArgs.out.println(";");
        }
    } catch (Exception ex) {
        System.err.println("Error generating random HOGM");
        ex.printStackTrace();
    }
}

Example 63 with Expression

use of com.sri.ai.expresso.api.Expression in project aic-expresso by aic-sri-international.

the class Theory method evaluate.

default default Expression evaluate(Expression expression, Context context) {
    ExpressionLiteralSplitterStepSolver evaluatorStepSolver = makeEvaluatorStepSolver(expression);
    Expression result = ContextDependentExpressionProblemSolver.staticSolve(evaluatorStepSolver, context);
    return result;
}

Example 64 with Expression

use of com.sri.ai.expresso.api.Expression in project aic-expresso by aic-sri-international.

the class Compilation method compile.

/**
	 * Compiles an expression to a normalized (decision-tree-like) expression.
	 * @param inputExpression
	 * @param mapFromVariableNameToTypeName
	 * @param mapFromCategoricalTypeNameToSizeString
	 * @param additionalTypes
	 * @param solverListener if not null, invoked on solver used for compilation, before and after compilation starts; returned solver on 'before' invocation is used (it may be the same one used as argument, of course).
	 * @return
	 */
public static Expression compile(Expression inputExpression, Theory theory, Map<String, String> mapFromVariableNameToTypeName, Map<String, String> mapFromUniquelyNamedConstantToTypeName, Map<String, String> mapFromCategoricalTypeNameToSizeString, Collection<Type> additionalTypes, Function<MultiIndexQuantifierEliminator, MultiIndexQuantifierEliminator> solverListener) {
    // the group actually does not matter, because we are not going to have any indices.
    AssociativeCommutativeGroup group = new Max();
    // The solver for the parameters above.
    MultiIndexQuantifierEliminator solver = new SGDPLLT();
    if (solverListener != null) {
        solver = solverListener.apply(solver);
    }
    // We use the Prolog convention of small-letter initials for constants, but we need an exception for the random variables.
    Predicate<Expression> isPrologConstant = new PrologConstantPredicate();
    Predicate<Expression> isUniquelyNamedConstantPredicate = e -> isPrologConstant.apply(e) && !mapFromVariableNameToTypeName.containsKey(e);
    Map<String, String> mapFromSymbolNameToTypeName = new LinkedHashMap<>(mapFromVariableNameToTypeName);
    mapFromSymbolNameToTypeName.putAll(mapFromUniquelyNamedConstantToTypeName);
    // Solve the problem.
    // no indices; we want to keep all variables
    List<Expression> indices = Util.list();
    Expression result = solver.solve(group, inputExpression, indices, mapFromSymbolNameToTypeName, mapFromCategoricalTypeNameToSizeString, additionalTypes, isUniquelyNamedConstantPredicate, theory);
    if (solverListener != null) {
        solverListener.apply(null);
    }
    return result;
}

Example 65 with Expression

use of com.sri.ai.expresso.api.Expression in project aic-expresso by aic-sri-international.

the class ExpressoAPIExamples method main.

public static void main(String[] args) {
    // Symbols are expressions representing Java values, such as a string,
    // a number, a boolean, and even any other objects (Expressions themselves will throw an error, though, to prevent common mistakes)
    Expression a = DefaultSymbol.createSymbol("a");
    Expression ten = DefaultSymbol.createSymbol(10);
    Expression trueValue = DefaultSymbol.createSymbol(true);
    Expression object = DefaultSymbol.createSymbol(Car.ferrari);
    // Below, we use Util.println to avoid having to write System.out.println.
    // Util has lots and lots and lots of very useful methods that eliminate boring Java chores.
    // We definitely recommend studying it carefully!
    // To easily write just "println" and have it work, write "println" and use Ctrl-1 to be offered the option
    // of statically importing Util.println.
    // You can statically import an identifier in Eclipse by placing the cursor on it and pressing Shift-Ctrl-M in Windows.
    println("a        : " + a);
    println("ten      : " + ten);
    println("trueValue: " + trueValue);
    println("object   : " + object);
    // It is easier to remember to make symbols using Expressions.makeSymbol:
    // 'Expressions' is a class with lots of useful static methods for working with expressions.
    // This of it as the counterpart of 'Util', but for expressions.
    // Having many useful expression classes in a single place makes it easier to remember them and access them through code completion.
    // As with 'Util', we often import its methods statically.
    a = makeSymbol("a");
    ten = makeSymbol(10);
    trueValue = makeSymbol(true);
    object = makeSymbol(Car.ferrari);
    println("a        : " + a);
    println("ten      : " + ten);
    println("trueValue: " + trueValue);
    println("object   : " + object);
    // The second most important type of expressions are function applications.
    // They consist of a functor (an expression representing a function, typically a symbol) applied to zero or more arguments:
    Expression f = makeSymbol("f");
    Expression g = makeSymbol("g");
    Expression fATen = new DefaultFunctionApplication(f, list(a, ten));
    Expression gOnNothing = new DefaultFunctionApplication(g, list());
    println("function f applied to a and ten: " + fATen);
    println("function g applied to nothing: " + gOnNothing);
    // It is much easier to use Expressions.apply:
    fATen = apply(f, a, ten);
    gOnNothing = apply(g);
    println("function f applied to a and ten: " + fATen);
    println("function g applied to nothing: " + gOnNothing);
    // Naturally, function applications can be applied to any expression, including other function applications:
    Expression gFATen = apply(g, apply(f, a, ten));
    println("function g applied to function f applied to a and ten: " + gFATen);
    // If we do not create symbols for Java values, apply does it automatically for us:
    gFATen = apply("g", apply("f", "a", 10));
    println("function g applied to function f applied to a and ten: " + gFATen);
    // Some operators are output in special infix notation for readability:
    // again not creating symbols first
    Expression twoPlusTwoPlusThree = apply("+", 2, 2, 3);
    println("two plus two plus three: " + twoPlusTwoPlusThree);
    Expression arithmetic1 = apply("*", 2, apply("+", 2, 3));
    Expression arithmetic2 = apply("+", 2, apply("*", 2, 3));
    println("Arithmetic gets printed while respecting usual precedence rules by using parentheses: " + arithmetic1);
    println("Arithmetic gets printed while respecting usual precedence rules by using parentheses: " + arithmetic2);
    // Same for logic:
    Expression logic1 = apply("and", "p", apply("or", "q", "r"));
    Expression logic2 = apply("or", "p", apply("and", "q", "r"));
    println("Same for logic: " + logic1);
    println("Same for logic: " + logic2);
    // It is not good practice to use separate strings for referring to the same operators.
    // In the future, we may decide to change the string associated to an operator
    // and have that string used in many places would make that hard to effect.
    // FunctorConstants is a class with lots of static fields for operator strings:
    arithmetic1 = apply(FunctorConstants.TIMES, 2, apply(FunctorConstants.PLUS, 2, 3));
    logic1 = apply(FunctorConstants.AND, "p", apply(FunctorConstants.OR, "q", "r"));
    // Functor constants can also be statically imported:
    arithmetic1 = apply(TIMES, 2, apply(PLUS, 2, 3));
    logic1 = apply(AND, "p", apply(OR, "q", "r"));
    // We can access the functor and arguments of a function application:
    println("The functor of " + gFATen + " is " + gFATen.getFunctor());
    println("The second argument of " + fATen + " is " + fATen.get(1));
    // returns a List<Expression>
    println("All arguments of " + fATen + " are " + fATen.getArguments());
    // We can also set new functors or arguments.
    // IMPORTANT: expressions are IMMUTABLE, so this creates a new expression,
    // although it does re-use the unchanged parts.
    Expression newArgument = gFATen.set(0, a);
    println("Changed first argument of " + gFATen + " to " + a + " and obtained " + newArgument);
    println("Original expression continues the same, since they are immutable: " + gFATen);
    // getFunctor returns a symbol, so to check if it is, for example, "f", we need
    // to write expression.getFunctor().equals(makeSymbol("f"))
    // which is too long.
    // Instead, we can use 'hasFunctor'
    println(fATen + " has functor \"f\": " + fATen.hasFunctor("f"));
    // Finally, we can parse expressions for strings.
    // BUT we should never use that to construct expressions if we have the sub-expressions already represented as Java objects.
    // For example, don't do this: parse("f(" + a + ", " + b + ")");
    // Instead, use apply("f", a, b);
    ten = parse("10");
    trueValue = parse("true");
    fATen = parse("f(a,10)");
    arithmetic1 = parse("2*(2 + 3)");
    arithmetic2 = parse("2+(2 * 3)");
    println(ten);
    println(trueValue);
    println(fATen);
    println(arithmetic1);
    println(arithmetic2);
    // Another important type of expression is sets.
    // There are two dimensions for sets: they can be uni- or multi-sets, and they can be extensionally or intensionally defined.
    //
    // A uni-set has at most one instance of each element in it. This is the typical mathematical set.
    // A multi-set may have multiple instances of the same element in it.
    // For example, the multi-set {1,2,2,3} is distinct from multi-set {1,2,2,2,3}.
    // In Expresso, we use double-brackets for denoting multi-sets:  {{ 1, 2, 2, 3 }}.
    // {{ }} denotes the empty multi-set.
    // The singleton uni-set with an empty set in it is denoted { {} }.
    // You need a space between the brackets to avoid them being parsed as a double bracket.
    //
    // An extensionally defined set is an explicit enumeration of its elements: {1, 2, 3}, {{1, 2, 2, 3}}, {}, {{ }}.
    // An intensionally defined set is defined by a condition: { (on I in Integer)  I^2 : I > 3 and I <= 100 }, for example,
    // which is equal to { 16, 25, 36, ..., 10000 }.
    // The general form of an intensionally defined set (or, less precisely but more succinctly, an intensional set) is
    // { (on Index1 in Index1Domain, Index2 in Index2Domain, ..., Index_n in Index_nDomain)   Head   :  Condition }
    // We can also have intensionally defined multi-sets using double brackets.
    // Here are some ways of constructing sets:
    a = makeSymbol("a");
    Expression b = makeSymbol("b");
    Expression c = makeSymbol("c");
    Expression d = makeSymbol("d");
    Expression extensionalUniSet = ExtensionalSets.makeUniSet(a, b, c, d);
    Expression extensionalMultiSet = ExtensionalSets.makeMultiSet(a, b, c, d);
    println(extensionalUniSet);
    println(extensionalMultiSet);
    // Creating an intensionally defined set programmatically (as opposed to parsing a string description of it)
    // is a bit of work (this will be shown below).
    // Here's an example of parsing one:
    Expression intensionalUniSet = parse("{ ( on P in People, F in Foods ) eats(P, F) : not (P = Rodrigo and F = shrimp) }");
    println(intensionalUniSet);
    // Here's how to do it from scratch, but see next the way we typically actually do it.
    Expression p = makeSymbol("P");
    Expression people = makeSymbol("People");
    f = makeSymbol("F");
    Expression foods = makeSymbol("Foods");
    IndexExpressionsSet indices = new ExtensionalIndexExpressionsSet(apply(IN, p, people), apply(IN, f, foods));
    // The "extensional" in ExtensionalIndexExpressionsSet means that the list/set of indices is extensionally defined,
    // even though they will be the indices of an intensionally defined set.
    intensionalUniSet = // IntensionalSet.intensionalUniSet, or simply intensionalUniSet, also works
    IntensionalSet.makeUniSet(indices, apply("eats", p, f), apply(NOT, apply(AND, Equality.make(p, "Rodrigo"), Equality.make(f, "shrimp"))));
    // Note that Equality.make(p, "Rodrigo") is the same as apply(FunctorConstants.EQUAL, p, "Rodrigo").
    // We often have 'make' methods for many operators: And.make, Or.make and so on.
    // packages in com.sri.ai.expresso.grinder.sgdpllt.library have many such operator-specific classes.
    println(intensionalUniSet);
    // When writing code on sets, we typically are modifying an existing set expression, so we can re-use its parts,
    // by using special part-replacement methods.
    // This requires the variable to implement the IntensionalSet interface, though.
    // IMPORTANT: expressions are IMMUTABLE, so setCondition and other part-replacement methods return a NEW expression,
    // although the parts not replaced are re-used.
    IntensionalSet intensionalSetCast = (IntensionalSet) intensionalUniSet;
    Expression noCondition = intensionalSetCast.setCondition(makeSymbol(true));
    println("Set with no condition: " + noCondition);
    Expression headSaysLoveInsteadOfEats = intensionalSetCast.setHead(apply("loves", p, f));
    println("Set with new head: " + headSaysLoveInsteadOfEats);
    Expression withNewIndices = intensionalSetCast.setIndexExpressions(new ExtensionalIndexExpressionsSet(apply(IN, p, people), apply(IN, f, foods), apply(IN, "D", "Days")));
    println("Set with new indices: " + withNewIndices);
    // summations and products are just function applications of FunctorConstants.SUM and FunctorConstants.PRODUCT on intensional multi-sets.
    // sum( {{ (on Indices)  Head  : Condition }} ) represents the summation (in Latex notation) sum_{Indices : Condition} Head
    Expression summation = apply(SUM, intensionalSetCast);
    println(summation);
    Expression product = apply(PRODUCT, intensionalSetCast);
    println(product);
    ///// Evaluating expressions
    // The above code shows how to deal with the syntax of expressions.
    // Evaluating expressions requires knowing about the semantics, that is, to what functions each operator corresponds to ("+" to addition, etc).
    // This is provided by a theory, which for now it suffices to know is a collection of methods for evaluating expressions
    // according to an interpretation to some symbols.
    Theory theory = new CompoundTheory(new EqualityTheory(false, true), new DifferenceArithmeticTheory(false, false), new LinearRealArithmeticTheory(false, false), new TupleTheory(), new PropositionalTheory());
    // Because this evaluation is symbolic, evaluated expressions may involve free variables.
    // In this case, the result of the evaluation will be a simplified expression that
    // is equivalent to the original expression for all possible assignments to the free variables.
    // For example, X + 0*Y is evaluate to X because, for any assignment to (X,Y), X + 0*Y = X.
    // true context: all assignments to free variables are of interest
    Context context = new TrueContext(theory);
    // We will later see how we can use contexts that restrict the free variable assignments of interest.
    // Now that we have a theory and a context, we can evaluate expressions:
    println("1 + 0*X + 1  =  " + theory.evaluate(parse("1 + 1"), context));
    evaluate(new String[] { "1 + 1", "2", "X + 1 + 1", "X + 2", "sum({{ (on I in 1..10) I }})", "55", "product({{ (on I in 1..5) 2 : I != 3 and I != 5 }})", "8" }, theory, context);
    // now let us assume we have a free variable J which is an integer
    // Contexts are, like expressions, also IMMUTABLE:
    Context context2 = context.extendWithSymbolsAndTypes("J", "Integer");
    // However, here we just want to use the same variable 'context' all along, so we keep the updated context in it:
    context = context2;
    // Because we store the reference to the modified context in the same variable, we lose the reference to the original one,
    // but, if we wanted, we could keep contexts in a stack, for example,
    // so that we could always easily revert back to a previous context if needed.
    evaluate(new String[] { "X + 1 + 1 + J", "X + 2 + J", "sum({{ (on I in 1..10) I : I != J }})", "if J > 0 then if J <= 10 then -1 * J + 55 else 55 else 55" }, theory, context);
    // now let us assume we have a free variable J which is an integer
    // The context is also a boolean formula (a constraint)
    // Current, its value is "true", but we can conjoin it with a literal J < 0
    context = context.conjoin(parse("J < 0"));
    evaluate(new String[] { "J < 1", "true", // J is irrelevant because it is out of the range of I
    "sum({{ (on I in 1..1000) I : I != J }})", // J is irrelevant because it is out of the range of I
    "500500" }, theory, context);
    // we now add another symbol and constraint
    context = context.extendWithSymbolsAndTypes("K", "Integer");
    context = context.conjoin(parse("K > 0"));
    evaluate(new String[] { "J < K", "true" }, theory, context);
    // Obtaining all free variables in an expression.
    // In order to obtain all free variables appearing in an expression
    // (and therefore excluding quantified variables (for all X, there exists X) and set indices ({ (on Z in Real) Z }))
    // we must traverse the expression and select its sub-expressions that are variables.
    // However, we need to know what a variable is.
    // It is not enough to say that any symbol is a variable, because "1" and "true" are symbols, but not variables.
    // It is not enough to say that any symbols that is an identifier (starting with an alphabet letter) is a variable,
    // because we may have uniquely named constants such as "john" and "bob" that are not to be treated as variables
    // (we want "john = bob" to be evaluated to "false", and if they were variable, this would not happen.
    // The way Expresso deals with this question is by letting the user define a predicate in the context that
    // encoded what a uniquely named constant is, and considering any other symbol to be considered a variable.
    // By default, Expresso follows the Prolog convention of capitalized variables: X is a variable, x is not.
    // Note how this takes "Real" to be a variable!
    context = new TrueContext();
    Expression expression = parse("X + f(g(x, Y, 1, true, false, 10, bob, john, there exists Z in Real : 10, { (on W in Real) 1 } ))");
    Set<Expression> variablesInExpression = Expressions.freeVariables(expression, context);
    // outputs [X, Y]
    println("variables in " + expression + " by Prolog standard: " + variablesInExpression);
    // More recently, we have adopted the practice of not caring about capitalization.
    // This means that we may, for example, defined uniquely named constants to be any symbols that are not in a given set of variables.
    Set<Expression> allVariables = set(parse("x"), parse("X"), parse("Y"), parse("Z"), parse("W"));
    context = context.setIsUniquelyNamedConstantPredicate(new UniquelyNamedConstantAreAllSymbolsNotIn(allVariables));
    variablesInExpression = Expressions.freeVariables(expression, context);
    println("variables in " + expression + " if all variables is " + allVariables + ": " + // outputs [x, X, Y]
    variablesInExpression);
    // Sometimes, it is useful to replace subexpressions in a given expression by another subexpression:
    expression = parse("f(f(f(X))) + X");
    Expression valueOfX = parse("10");
    Expression replacementOfFirstOccurrenceOnly = expression.replaceFirstOccurrence(parse("X"), valueOfX, context);
    println("Replacing only the first occurrence of X by its value gives " + replacementOfFirstOccurrenceOnly);
    Expression replacementOfAllOccurrences = expression.replaceAllOccurrences(parse("X"), valueOfX, context);
    println("Replacing all occurrences of X by its value gives " + replacementOfAllOccurrences);
    // There are a LOT of variants of these functions in Expression (not Expressions),
    // including some very flexible ones that allow the user to provide a function for determining the replacement.
    // BUG: need to debug
    // Here's how to decide if a point is in the convex hull of other two points:
    Context convexityContext = context.extendWithSymbolsAndTypes("p", "Real", "p1", "Real", "p2", "Real");
    //convexityContext = convexityContext.conjoin(parse("p  = 4"));
    //convexityContext = convexityContext.conjoin(parse("p1 = 3"));
    //convexityContext = convexityContext.conjoin(parse("p2 = 5"));
    Expression isInConvexHull = parse("there exists c1 in Real : there exists c2 in Real : " + "0 <= c1 and c1 <= 1 and 0 <= c2 and c2 <= 1 and 4 = c1*3 + c2*5");
    Expression result = theory.evaluate(isInConvexHull, convexityContext);
    println("4 is in the convex hull of 3 and 5: " + result);
}