relationalai 0.12.13__py3-none-any.whl → 0.13.0.dev0__py3-none-any.whl

v0/relationalai/early_access/dsl/core/constraints/predicate/atomic.py

@@ -0,0 +1,455 @@
+from v0.relationalai.early_access.dsl.core.constraints import Constraint
+from v0.relationalai.early_access.dsl.core.constraints.scalar import ScalarConstraint
+from v0.relationalai.early_access.dsl.core.exprs import Expr, Wildcard
+
+# An AtomicConstraint is a Constraint that will ultimately realize as a
+# literal in the head or body of a datalog rule. The concrete subclass
+# Atom allows users to declare these directly, while other subclasses
+# provide a slightly higher-level means for doing so by treating
+# relations as sets.
+#
+class AtomicConstraint(Constraint):
+
+    relcomp = None
+
+    def component(self): return self.relcomp # type: ignore
+
+    def decorated(self) -> bool:
+        return self.relcomp.decorated() # type: ignore
+
+    def undecorated(self) -> bool:
+        return self.relcomp.undecorated() # type: ignore
+
+
+# An Atom is a constraint that can be used as a literal in the
+# head or body of a datalog rule, i.e.:
+#
+#   R(x1, x2, ..., xn)
+#
+# or:
+#
+#   not R(x1, x2, ..., xn)
+#
+# where x1, x2, ..., xn are arithmetic expressions.
+#
+class Atom(AtomicConstraint):
+
+    def __init__(self, r, args):
+        self.relcomp = r
+        self.args = args
+        self.negated = False
+
+    def entityid(self):
+        hvals = [a.entityid() for a in self.args]
+        hvals.append(self.relcomp.entityid())
+        return hash(tuple(hvals))
+
+    def grounds(self, v, groundings):
+        for a in self.args:
+            if a.entityid() == v.entityid():
+                return True
+        return False
+
+    # Returns a new Atom that duplicates self after mapping
+    # the Relation component through relmap
+    #
+    def map(self, relmap):
+        newrel = relmap[self.relcomp.entityid()]
+        newatom = Atom(newrel, self.args)
+        newatom.negated = self.negated
+        return newatom
+
+    # Returns a new Atom that is constructed by the given
+    # builder function
+    #
+    def map_builder(self, builder):
+        new_atom = builder(self.relcomp, self.args)
+        new_atom.negated = self.negated
+        return new_atom
+
+    def __invert__(self):
+        self.negated = not self.negated
+        return self
+
+    def relation(self):
+        return self.relcomp.typeof() if hasattr(self.relcomp, 'typeof') else self.relcomp
+
+    # Emit this schema to a textual output using the Z display style
+    #
+    def pprint(self):
+        relname = self.relcomp.display()
+        head_str = ", ".join([v.display() for v in self.args])
+        patom = f"{relname}({head_str})"
+        if self.negated:
+            return "not " + patom
+        else:
+            return patom
+
+    def rel_formula(self):
+        relname = self.relcomp.display()
+        head_str = ", ".join([v.display() for v in self.args])
+
+        if self.relation().arity() == len(self.args):
+            patom = f"{relname}({head_str})"
+        else:
+            patom = f"{relname}[{head_str}]"
+
+        if self.negated:
+            return "not " + patom
+        else:
+            return patom
+
+    def scalar_refs(self):
+        dic = {}
+        for a in self.args:
+            if a.variable():
+                dic[a.display()] = a
+            else:
+                cdic = a.scalar_refs()
+                for v in cdic:
+                    if v not in dic:
+                        dic[v] = cdic[v]
+        return dic
+
+    def refersto(self, varname: str) -> bool:
+        if self.relcomp.refersto(varname):
+            return True
+        for a in self.args:
+            if a.refersto(varname):
+                return True
+        return False
+
+    def rename(self, renaming):
+        #
+        # `self.relcomp` can either be a `RelationalComponent` or another type such as `ExternalRelation`.
+        # Note that `RelationalComponent` and `ExternalRelation` do not have a common supertype.
+        # Only instances of `RelationalComponent` support renaming, so we need to check for that here.
+        # We need to import `RelationalComponent` locally to avoid circular dependencies.
+        #
+        # TODO: This requires a proper fix, which would possibly introduce a common supertype for the types which can
+        #  occur here and separate the relation and variable renamings into distinct maps (as is done elsewhere, e.g.,
+        #  in callers of this method).
+        #
+        from v0.relationalai.early_access.dsl.schemas import RelationalComponent
+        new_relcomp = self.relcomp.rename(renaming) if isinstance(self.relcomp, RelationalComponent) else self.relcomp
+        renamed = Atom(new_relcomp,
+                       [a.rename(renaming) for a in self.args])
+        renamed.negated = self.negated
+        return renamed
+
+    def simplify(self):
+        return self
+
+    def substitute(self, bindings):
+        return Atom(self.relcomp.substitute(bindings),
+                    [a.substitute(bindings)
+                     for a in self.args])
+
+    # [REKS] Need a better name. The purpose of this method is to undash an atom
+    #        that involves relation R when we are generating a recursive Rel rule
+    #        from an operation Schema involving relational components that are
+    #        dashed and undashed occurrences of R. The "undashed" atom will be
+    #        used to form the head of this rule.
+    #
+    def rule_head(self):
+        return Atom(self.relcomp.undash(), self.args)
+
+    def value_column(self):
+        rel = self.relation()
+        pos = rel.arity() - 1  # Position of value component
+        return self.args[pos]
+
+
+class ElementOf(AtomicConstraint):
+
+    def __init__(self, x, y):
+        self.element = x
+        self.relcomp = y
+
+        # Emit this schema to a textual output using the Z display style
+
+    #
+    def pprint(self):
+        return self.element.display() + " IN " + self.relcomp.display()
+
+    def rel_formula(self):
+        elem = self.element
+        if elem.relational():
+            raise Exception(f"Cannot make expr {elem.display()} an argument of an AtomicConstraint")
+
+        return self.relcomp.project(elem).rel_formula()
+
+    def grounded(self):
+        return self.element.grounded()
+
+    def grounds(self, v, groundings):
+        return self.element.entityid() == v.entityid()
+
+    def scalar_refs(self):
+        dic = self.relcomp.scalar_refs()
+        elem = self.element
+        if elem.variable():
+            dic[elem.display()] = elem
+        else:
+            edic = elem.scalar_refs()
+            for v in edic:
+                if v not in dic:
+                    dic[v] = edic[v]
+        return dic
+
+    def refersto(self, varname: str) -> bool:
+        return self.element.refersto(varname) or self.relcomp.refersto(varname)
+
+    def rename(self, renaming):
+        return ElementOf(self.element.rename(renaming),
+                         self.relcomp.rename(renaming))
+
+    def simplify(self):
+        return ElementOf(self.element, self.relcomp.simplify())
+
+    def substitute(self, bindings):
+        e = self.element.substitute(bindings)
+        p = self.relcomp.substitute(bindings)
+        return ElementOf(e, p)
+
+    def entityid(self):
+        return hash((ElementOf,
+                     self.element.entityid(),
+                     self.relcomp.entityid()))
+
+
+class NotElementOf(AtomicConstraint):
+
+    def __init__(self, x, y):
+        self.element = x
+        self.relcomp = y
+
+    # Emit this schema to a textual output using the Z display style
+    #
+    def pprint(self):
+        return self.element.display() + " NOT IN " + self.relcomp.display()
+
+    def rel_formula(self):
+        return "not " + self.relcomp.project(self.element).rel_formula()
+
+    def grounded(self):
+        return self.element.grounded()
+
+    def grounds(self, v, groundings):
+        return False
+
+    def scalar_refs(self):
+        dic = self.relcomp.scalar_refs()
+        elem = self.element
+        if elem.variable():
+            dic[elem.display()] = elem
+        else:
+            edic = elem.scalar_refs()
+            for v in edic:
+                if v not in dic:
+                    dic[v] = edic[v]
+        return dic
+
+    def refersto(self, varname: str) -> bool:
+        return self.element.refersto(varname) or self.relcomp.refersto(varname)
+
+    def rename(self, renaming):
+        return NotElementOf(self.element.rename(renaming),
+                            self.relcomp.rename(renaming))
+
+    def simplify(self):
+        return ElementOf(self.element, self.relcomp.simplify())
+
+    def substitute(self, bindings):
+        e = self.element.substitute(bindings)
+        p = self.relcomp.substitute(bindings)
+        return NotElementOf(e, p)
+
+    def entityid(self):
+        return hash((NotElementOf,
+                     self.element.entityid(),
+                     self.relcomp.entityid()))
+
+
+# This module declares operations that range over sets of scalar and
+# relational constraints.
+#
+# By convention, we model constraint sets as dictionaries that map
+# a hash of the Constraint to the Constraint itself to aid in
+# deduplication.
+#
+
+
+# AtomicConstraint set optimizations follow
+
+
+# Given a set A of AtomicConstraints:
+#
+#   - Remove any obviously redundant atoms, and
+#   - Unify any pair { c1, c2 } subseteq A, if c1 and c2 refer to the
+#     same Relation and that Relation is functional, then if c2 and c2
+#     can be shown to refer to the same fact, add new equality
+#     constraints to record this knowledge.
+#
+# Returns the set A less any redundant atoms and extended by any equality
+# constraints that can be discovered through unification.
+#
+# We peform unification by constructing a UnificationTree (Utree), which is
+# a recursive dictionary of dictionaries ultimately terminating in lists of
+# atoms from A. Each dictionary at depth k implements the decision to extend
+# the tuple formed by decisions made at depths > k with a decision (column
+# value) at depth k. For instance, were A to contain the following atoms:
+#
+#    a1 = foo(x, y, 2)
+#    a2 = foo(x, y, v)
+#    a3 = foo(a, y, c)
+#    a4 = foo(a, z, d)
+#
+# then the Utree would look like this:
+#
+#    { a: { y: [a3],
+#           z: [a4] },
+#      x: { y: [a1, a2] } }
+#
+# where {a, x, y, z} in these keys are the hashes of those expressions
+#
+def reduce_relational_atoms(A):
+
+    # Given a set of AtomicConstraints A and a Relation hash r, returns those
+    # atoms in A that reference the Relation whose hash is r
+    #
+    def references(A, r):
+        return {a.entityid(): a
+                for a in A.values()
+                if a.relcomp.entityid() == r}
+
+    # Given a set of AtomicConstraints A, return the minimal set of relations
+    # such that each Relation in the return set is referenced in at least
+    # one atom in A
+    #
+    def rel_refs_of(A):
+        rels = {}
+        for a in A.values():
+            rel = a.relcomp
+            rv = rel.entityid()
+            if rv not in rels:
+                rels[rv] = rel
+        return rels
+
+    # A UnificationTree (Utree) is a decision-tree representation (ordered
+    # by column number) of one or more facts that are asserted by a set of
+    # AtomicConstraints that all reference the same Relation
+    #
+    def utree_digest(utree, depth, args, atom):
+        cval = args[0]
+        v = cval.entityid() if isinstance(cval, Expr) else cval
+        if v in utree:
+            if depth == 1:
+                utree[v].append(atom)
+            else:
+                utree_digest(utree[v], depth - 1, args[1:], atom)
+        else:
+            if depth == 1:
+                utree[v] = [atom]
+            else:
+                utree[v] = {}
+                utree_digest(utree[v], depth - 1, args[1:], atom)
+
+    # Flatten a Utree by recording all of the atoms at its leaves
+    # in the set atoms
+    #
+    def utree_export(utree, depth, atoms):
+        for v in utree.values():
+            if depth == 1:
+                atoms[a.entityid()] = a
+            else:
+                utree_export(v, depth - 1, atoms)
+        return atoms
+
+    # Mutate utree to remove atoms with Wildcard values when we can show
+    # those atoms are redundant
+    #
+    def utree_remove_wildcards(utree, depth):
+        if depth == 1:
+            for n in utree.values():
+                if len(n) > 1:
+                    # If a functional relation has an n-1 tuple prefix that maps
+                    # to multiple values, then any atom with a Wildcard value
+                    # is redundant and can be removed
+                    #
+                    for a in n:
+                        if isinstance(a.value_column(), Wildcard):
+                            n.remove(a)
+        else:
+            for pref in utree:
+                utree_remove_wildcards(utree[pref], depth - 1)
+
+    # Given a Utree, whose paths represent a set of referent_atoms that refer to some
+    # functional relation, walk that tree to unify referent_atoms that differ only in
+    # the last column so as to produce two sets:
+    #
+    #  - unique_atoms, which is a minimal set of referent atoms such that no two
+    #    atoms will reference the same key tuple; and
+    #
+    #  - unifying_constraints, which is a set of equality constraints that equate
+    #    any expressions that that arise from unifying two or more paths in the tree
+    #    that reference the same key tuple.
+    #
+    # This function should be called with a root Utree, its depth, and two empty
+    # dictionaries that represent the sets unique_atoms and unifying_constraints,
+    # and the function will recursively walk the tree to fill in these sets.
+    #
+    # If the set referent_atoms already has the unique key tuple property, then on
+    # return:
+    #
+    #    unique_atoms == referent_atoms and unifying_constraints == {}
+    #
+    # Otherwise, on return:
+    #
+    #    unique_atoms \subset referent_atoms and unifying_constraints != {}
+    #
+    def utree_unify_and_export(utree, depth, unique_atoms, unifying_constraints):
+        for v in utree.values():
+            if depth == 1:
+                prev_value = None
+                for idx, a in enumerate(v):
+                    last_col = a.value_column()
+                    if idx == 0:
+                        prev_value = last_col
+                        unique_atoms[a.entityid()] = a
+                    else:
+                        eq = ScalarConstraint(prev_value, "=", last_col)
+                        unifying_constraints[eq.entityid()] = eq
+            else:
+                utree_unify_and_export(v, depth - 1, unique_atoms, unifying_constraints)
+
+    equality_constraints = {}
+    atoms = {}
+    relations = rel_refs_of(A)
+    for r in relations:
+
+        # Depth to which we need to build interior nodes (lists) in the
+        # unification tree
+        #
+        rel = relations[r].typeof()
+        depth = rel.arity() - 1
+
+        referents = references(A, r)
+
+        if depth == 0:
+            for a in referents:
+                atoms[a] = referents[a]
+        else:
+            # Assemble a Utree for the atoms that reference Relation r
+            #
+            r_utree = {}
+            for a in referents.values():
+                utree_digest(r_utree, depth, a.args, a)
+
+            utree_remove_wildcards(r_utree, depth)
+            if rel.functional:
+                utree_unify_and_export(r_utree, depth, atoms, equality_constraints)
+            else:
+                utree_export(r_utree, depth, atoms)
+
+    return (equality_constraints, atoms)

v0/relationalai/early_access/dsl/core/constraints/predicate/universal.py

@@ -0,0 +1,73 @@
+from abc import abstractmethod
+
+from v0.relationalai.early_access.dsl.core.constraints import Constraint
+
+
+# A universal predciate constraint is one that can be formalized
+# over one or more Relations using variables that are universally
+# quantified
+#
+class UniversalConstraint(Constraint):
+
+    def __init__(self, left, right):
+        self.left = left
+        self.right = right
+
+    def rename(self, bdgs):
+        return UniversalConstraint(self.left.rename(bdgs), self.right.rename(bdgs))
+
+    @abstractmethod
+    def xi(self) -> bool:
+        pass
+
+    # If this constraint is a Xi constraint, returns it in a form that can be
+    # applied as substitution of the more heavily decorated variable with the
+    # less decorated one
+    #
+    def xiSubstitution(self):
+        if self.xi():
+            lexpr = self.left
+            rexpr = self.right
+            if (lexpr.var.name == rexpr.var.name) and (lexpr.var.dashes < rexpr.var.dashes):
+                return RelationalEqualityConstraint(rexpr, lexpr)
+            else:
+                return self
+        else:
+            return None
+
+
+class RelationalEqualityConstraint(UniversalConstraint):
+
+    def __init__(self, left, right):
+        super().__init__(left, right)
+
+    def pprint(self):
+        return self.left.display() + " == " + self.right.display()
+
+    def rename(self, bdgs):
+        return RelationalEqualityConstraint( self.left.rename(bdgs),
+                                             self.right.rename(bdgs) )
+
+    def relational(self) -> bool:
+        return True
+
+    def entityid(self):
+        return hash((RelationalEqualityConstraint,
+                     self.left.entityid(),
+                     self.right.entityid()))
+
+    # A Xi constraint is one that equates two schema variables of
+    # compatible relational type. These are so-called because they
+    # typically arise through application of the Xi schema operator.
+    #
+    def xi(self):
+
+        lexpr = self.left
+        if lexpr.variable():
+            rexpr = self.right
+            if rexpr.variable():
+                return lexpr.typeof().entityid() == rexpr.typeof().entityid()
+            else:
+                return False
+        else:
+            return False