Draft STDEV() aggregate function

src/engine/GroupBy.cpp

@@ -19,6 +19,7 @@
 #include "engine/sparqlExpressions/SampleExpression.h"
 #include "engine/sparqlExpressions/SparqlExpression.h"
 #include "engine/sparqlExpressions/SparqlExpressionGenerators.h"
+#include "engine/sparqlExpressions/StdevExpression.h"
 #include "global/RuntimeParameters.h"
 #include "index/Index.h"
 #include "index/IndexImpl.h"
@@ -1013,6 +1014,7 @@ GroupBy::isSupportedAggregate(sparqlExpression::SparqlExpression* expr) {
   if (auto val = dynamic_cast<GroupConcatExpression*>(expr)) {
     return H{GROUP_CONCAT, val->getSeparator()};
   }
+  if (dynamic_cast<StdevExpression*>(expr)) return H{STDEV};
   if (dynamic_cast<SampleExpression*>(expr)) return H{SAMPLE};
 
   // `expr` is an unsupported aggregate

src/engine/GroupBy.h

@@ -245,6 +245,7 @@ class GroupBy : public Operation {
     MAX,
     SUM,
     GROUP_CONCAT,
+    STDEV,
     SAMPLE
   };
 

src/engine/sparqlExpressions/AggregateExpression.cpp

@@ -6,6 +6,7 @@
 #include "engine/sparqlExpressions/AggregateExpression.h"
 
 #include "engine/sparqlExpressions/GroupConcatExpression.h"
+#include "engine/sparqlExpressions/StdevExpression.h"
 
 namespace sparqlExpression::detail {
 
@@ -180,6 +181,10 @@ AggregateExpression<AggregateOperation, FinalOperation>::getVariableForCount()
 // Explicit instantiation for the AVG expression.
 template class AggregateExpression<AvgOperation, decltype(avgFinalOperation)>;
 
+// Explicit instantiation for the STDEV expression.
+template class DeviationAggExpression<AvgOperation,
+                                      decltype(stdevFinalOperation)>;
+
 // Explicit instantiations for the other aggregate expressions.
 #define INSTANTIATE_AGG_EXP(Function, ValueGetter) \
   template class AggregateExpression<              \

src/engine/sparqlExpressions/StdevExpression.h

@@ -0,0 +1,163 @@
+// Copyright 2024, University of Freiburg,
+// Chair of Algorithms and Data Structures.
+// Author: Christoph Ullinger <[email protected]>
+
+#ifndef QLEVER_STDEVEXPRESSION_H
+#define QLEVER_STDEVEXPRESSION_H
+
+#include <cmath>
+#include <functional>
+#include <memory>
+#include <variant>
+
+#include "engine/sparqlExpressions/AggregateExpression.h"
+#include "engine/sparqlExpressions/LiteralExpression.h"
+#include "engine/sparqlExpressions/NaryExpression.h"
+#include "engine/sparqlExpressions/SparqlExpression.h"
+#include "engine/sparqlExpressions/SparqlExpressionTypes.h"
+#include "engine/sparqlExpressions/SparqlExpressionValueGetters.h"
+#include "global/ValueId.h"
+
+namespace sparqlExpression {
+
+namespace detail {
+
+/// The STDEV Expression
+
+// Helper function to extract a double from a NumericValue variant
+auto inline numValToDouble =
+    []<typename T>(T&& value) -> std::optional<double> {
+  if constexpr (ad_utility::isSimilar<T, double> ||
+                ad_utility::isSimilar<T, int64_t>) {
+    return static_cast<double>(value);
+  } else {
+    return std::nullopt;
+  }
+};
+
+// Helper expression: The individual deviation squares. A DeviationExpression
+// over X corresponds to the value (X - AVG(X))^2.
+class DeviationExpression : public SparqlExpression {
+ private:
+  Ptr child_;
+  bool distinct_;
+
+ public:
+  DeviationExpression(bool distinct, Ptr&& child)
+      : child_{std::move(child)}, distinct_{distinct} {}
+
+  // __________________________________________________________________________
+  ExpressionResult evaluate(EvaluationContext* context) const override {
+    auto impl =
+        [this, context](SingleExpressionResult auto&& el) -> ExpressionResult {
+      // Prepare space for result
+      VectorWithMemoryLimit<IdOrLiteralOrIri> exprResult{context->_allocator};
+      std::fill_n(std::back_inserter(exprResult), context->size(),
+                  IdOrLiteralOrIri{Id::makeUndefined()});
+
+      auto devImpl = [&exprResult, context](auto generator) {
+        double sum = 0.0;
+        std::vector<double> childResults = {};
+
+        // Collect values as doubles
+        for (auto& inp : generator) {
+          const auto& s = detail::NumericValueGetter{}(std::move(inp), context);
+          auto v = std::visit(numValToDouble, s);
+          if (v.has_value()) {
+            childResults.push_back(v.value());
+            sum += v.value();
+          }
+          context->cancellationHandle_->throwIfCancelled();
+        }
+
+        // Calculate squared deviation and save for result
+        double avg = sum / static_cast<double>(context->size());
+        for (size_t i = 0; i < childResults.size(); i++) {
+          exprResult.at(i) = IdOrLiteralOrIri{
+              ValueId::makeFromDouble(std::pow(childResults.at(i) - avg, 2))};
+        }
+      };
+
+      auto generator =
+          detail::makeGenerator(AD_FWD(el), context->size(), context);
+      if (distinct_) {
+        context->cancellationHandle_->throwIfCancelled();
+        devImpl(detail::getUniqueElements(context, context->size(),
+                                          std::move(generator)));
+      } else {
+        devImpl(std::move(generator));
+      }
+
+      return exprResult;
+    };
+    auto childRes = child_->evaluate(context);
+    return std::visit(impl, std::move(childRes));
+  };
+
+  // __________________________________________________________________________
+  AggregateStatus isAggregate() const override {
+    return SparqlExpression::AggregateStatus::NoAggregate;
+  }
+
+  // __________________________________________________________________________
+  [[nodiscard]] string getCacheKey(
+      const VariableToColumnMap& varColMap) const override {
+    return absl::StrCat("[ SQ.DEVIATION ", distinct_ ? " DISTINCT " : "", "]",
+                        child_->getCacheKey(varColMap));
+  }
+
+ private:
+  // _________________________________________________________________________
+  std::span<SparqlExpression::Ptr> childrenImpl() override {
+    return {&child_, 1};
+  }
+};
+
+// Separate subclass of AggregateOperation, that replaces its child with a
+// DeviationExpression of this child. Everything else is left untouched.
+template <typename AggregateOperation,
+          typename FinalOperation = decltype(identity)>
+class DeviationAggExpression
+    : public AggregateExpression<AggregateOperation, FinalOperation> {
+ public:
+  // __________________________________________________________________________
+  DeviationAggExpression(bool distinct, SparqlExpression::Ptr&& child,
+                         AggregateOperation aggregateOp = AggregateOperation{})
+      : AggregateExpression<AggregateOperation, FinalOperation>(
+            distinct,
+            std::make_unique<DeviationExpression>(distinct, std::move(child)),
+            aggregateOp){};
+};
+
+// The final operation for dividing by degrees of freedom and calculation square
+// root after summing up the squared deviation
+inline auto stdevFinalOperation = [](const NumericValue& aggregation,
+                                     size_t numElements) {
+  auto divAndRoot = [](double value, double degreesOfFreedom) {
+    if (degreesOfFreedom <= 0) {
+      return 0.0;
+    } else {
+      return std::sqrt(value / degreesOfFreedom);
+    }
+  };
+  return makeNumericExpressionForAggregate<decltype(divAndRoot)>()(
+      aggregation, NumericValue{static_cast<double>(numElements) - 1});
+};
+
+// The actual Standard Deviation Expression
+// Mind the explicit instantiation of StdevExpressionBase in
+// AggregateExpression.cpp
+using StdevExpressionBase =
+    DeviationAggExpression<AvgOperation, decltype(stdevFinalOperation)>;
+class StdevExpression : public StdevExpressionBase {
+  using StdevExpressionBase::StdevExpressionBase;
+  ValueId resultForEmptyGroup() const override { return Id::makeFromInt(0); }
+};
+
+}  // namespace detail
+
+using detail::StdevExpression;
+
+}  // namespace sparqlExpression
+
+#endif  // QLEVER_STDEVEXPRESSION_H

src/parser/sparqlParser/SparqlQleverVisitor.cpp

@@ -23,6 +23,7 @@
 #include "engine/sparqlExpressions/RegexExpression.h"
 #include "engine/sparqlExpressions/RelationalExpressions.h"
 #include "engine/sparqlExpressions/SampleExpression.h"
+#include "engine/sparqlExpressions/StdevExpression.h"
 #include "engine/sparqlExpressions/UuidExpressions.h"
 #include "parser/RdfParser.h"
 #include "parser/SparqlParser.h"
@@ -2227,6 +2228,8 @@ ExpressionPtr Visitor::visit(Parser::AggregateContext* ctx) {
     }
 
     return makePtr.operator()<GroupConcatExpression>(std::move(separator));
+  } else if (functionName == "stdev") {
+    return makePtr.operator()<StdevExpression>();
   } else {
     AD_CORRECTNESS_CHECK(functionName == "sample");
     return makePtr.operator()<SampleExpression>();

src/parser/sparqlParser/SparqlQleverVisitor.h

@@ -10,6 +10,7 @@
 
 #include "engine/sparqlExpressions/AggregateExpression.h"
 #include "engine/sparqlExpressions/NaryExpression.h"
+#include "engine/sparqlExpressions/StdevExpression.h"
 #include "parser/data/GraphRef.h"
 #undef EOF
 #include "parser/sparqlParser/generated/SparqlAutomaticVisitor.h"

src/parser/sparqlParser/generated/SparqlAutomatic.g4

@@ -582,6 +582,7 @@ aggregate :  COUNT '(' DISTINCT? ( '*' | expression ) ')'
            | MIN '(' DISTINCT? expression ')'
            | MAX '(' DISTINCT? expression ')'
            | AVG '(' DISTINCT? expression ')'
+           | STDEV '(' DISTINCT? expression ')'
            | SAMPLE '(' DISTINCT? expression ')'
            | GROUP_CONCAT '(' DISTINCT? expression ( ';' SEPARATOR '=' string )? ')'  ;
 
@@ -763,6 +764,7 @@ SUM : S U M;
 MIN : M I N;
 MAX : M A X;
 AVG : A V G;
+STDEV : S T D E V ;
 SAMPLE : S A M P L E;
 SEPARATOR : S E P A R A T O R;
 

src/parser/sparqlParser/generated/SparqlAutomatic.tokens

@@ -136,43 +136,44 @@ SUM=135
 MIN=136
 MAX=137
 AVG=138
-SAMPLE=139
-SEPARATOR=140
-IRI_REF=141
-PNAME_NS=142
-PNAME_LN=143
-BLANK_NODE_LABEL=144
-VAR1=145
-VAR2=146
-LANGTAG=147
-PREFIX_LANGTAG=148
-INTEGER=149
-DECIMAL=150
-DOUBLE=151
-INTEGER_POSITIVE=152
-DECIMAL_POSITIVE=153
-DOUBLE_POSITIVE=154
-INTEGER_NEGATIVE=155
-DECIMAL_NEGATIVE=156
-DOUBLE_NEGATIVE=157
-EXPONENT=158
-STRING_LITERAL1=159
-STRING_LITERAL2=160
-STRING_LITERAL_LONG1=161
-STRING_LITERAL_LONG2=162
-ECHAR=163
-NIL=164
-ANON=165
-PN_CHARS_U=166
-VARNAME=167
-PN_PREFIX=168
-PN_LOCAL=169
-PLX=170
-PERCENT=171
-HEX=172
-PN_LOCAL_ESC=173
-WS=174
-COMMENTS=175
+STDEV=139
+SAMPLE=140
+SEPARATOR=141
+IRI_REF=142
+PNAME_NS=143
+PNAME_LN=144
+BLANK_NODE_LABEL=145
+VAR1=146
+VAR2=147
+LANGTAG=148
+PREFIX_LANGTAG=149
+INTEGER=150
+DECIMAL=151
+DOUBLE=152
+INTEGER_POSITIVE=153
+DECIMAL_POSITIVE=154
+DOUBLE_POSITIVE=155
+INTEGER_NEGATIVE=156
+DECIMAL_NEGATIVE=157
+DOUBLE_NEGATIVE=158
+EXPONENT=159
+STRING_LITERAL1=160
+STRING_LITERAL2=161
+STRING_LITERAL_LONG1=162
+STRING_LITERAL_LONG2=163
+ECHAR=164
+NIL=165
+ANON=166
+PN_CHARS_U=167
+VARNAME=168
+PN_PREFIX=169
+PN_LOCAL=170
+PLX=171
+PERCENT=172
+HEX=173
+PN_LOCAL_ESC=174
+WS=175
+COMMENTS=176
 '*'=1
 '('=2
 ')'=3