Add C++ implementation of the Count Sketch algorithm

commit.sh

@@ -1,4 +1,4 @@
-BRANCH=NIGHTLY_TEST
+BRANCH=count-sketch-cpp
 git init
 git checkout -b $BRANCH
 git add .

commit_info

@@ -1,3 +1 @@
-1. add pure c++ algorithms
-2. add normalized error and error bound evaluation at benchmark
-3. update scripts
+1. FIX CONFLICTS WITH MAIN

include/AMMBench.h

@@ -100,6 +100,7 @@
 #include <CPPAlgos/CPPAlgoTable.h>
 #include <CPPAlgos/CRSCPPAlgo.h>
 #include <CPPAlgos/CRSV2CPPAlgo.h>
+#include <CPPAlgos/CountSketchCPPAlgo.h>
 #include <CPPAlgos/BCRSCPPAlgo.h>
 #include <CPPAlgos/EWSCPPAlgo.h>
 #include <CPPAlgos/CoOccurringFDCPPAlgo.h>

include/CPPAlgos/CountSketchCPPAlgo.h

@@ -0,0 +1,58 @@
+//
+// Created by luv on 5/28/23.
+//
+
+#ifndef INTELLISTREAM_COUNTSKETCHCPPALGO_H
+#define INTELLISTREAM_COUNTSKETCHCPPALGO_H
+
+#include <CPPAlgos/AbstractCPPAlgo.h>
+
+namespace AMMBench {
+/**
+ * @ingroup AMMBENCH_CppAlgos The algorithms writtrn in c++
+ * @{
+ */
+/**
+ * @class CountSketchCPPAlgo CPPAlgos/CountSketchCPPAlgo.h
+ * @brief The cloumn row sampling (CRS) class of c++ algos
+ *
+ */
+    class CountSketchCPPAlgo : public AMMBench::AbstractCPPAlgo {
+    public:
+        CountSketchCPPAlgo() {
+
+        }
+
+        ~CountSketchCPPAlgo() {
+
+        }
+
+        /**
+         * @brief the virtual function provided for outside callers, rewrite in children classes
+         * @param A the A matrix
+         * @param B the B matrix
+         * @param sketchSize the size of sketc or sampling
+         * @return the output c matrix
+         */
+        virtual torch::Tensor amm(torch::Tensor A, torch::Tensor B, int sketchSize);
+
+    };
+
+/**
+ * @ingroup AMMBENCH_CppAlgos
+ * @typedef AbstractMatrixCppAlgoPtr
+ * @brief The class to describe a shared pointer to @ref CountSketchCPPAlgo
+
+ */
+    typedef std::shared_ptr<class AMMBench::CountSketchCPPAlgo> CountSketchCPPAlgoPtr;
+/**
+ * @ingroup AMMBENCH_CppAlgos
+ * @def newCRSV2CppAlgo
+ * @brief (Macro) To creat a new @ref  CRSV2CppAlgounder shared pointer.
+ */
+#define newCountSketchCPPAlgo std::make_shared<AMMBench::CountSketchCPPAlgo>
+}
+/**
+ * @}
+ */
+#endif //INTELLISTREAM_COUNTSKETCHCPPALGO_H

src/CPPAlgos/CMakeLists.txt

@@ -7,4 +7,6 @@ add_sources(
         EWSCPPAlgo.cpp
         CoOccurringFDCPPAlgo.cpp
         BetaCoOFDCPPAlgo.cpp
-)
+        CountSketchCPPAlgo.cpp
+)
+

src/CPPAlgos/CPPAlgoTable.cpp

@@ -5,19 +5,24 @@
 #include <CPPAlgos/CPPAlgoTable.h>
 #include <CPPAlgos/CRSCPPAlgo.h>
 #include <CPPAlgos/CRSV2CPPAlgo.h>
+
+#include <CPPAlgos/CountSketchCPPAlgo.h>
+
 #include <CPPAlgos/BCRSCPPAlgo.h>
 #include <CPPAlgos/EWSCPPAlgo.h>
 #include <CPPAlgos/CoOccurringFDCPPAlgo.h>
 #include <CPPAlgos/BetaCoOFDCPPAlgo.h>
+
 namespace AMMBench {
 AMMBench::CPPAlgoTable::CPPAlgoTable() {
   algoMap["mm"] = newAbstractCPPAlgo();
   algoMap["crs"] = newCRSCPPAlgo();
   algoMap["crsV2"] = newCRSV2CPPAlgo();
+  algoMap["count-sketch"] = newCountSketchCPPAlgo();
   algoMap["bcrs"] = newBCRSCPPAlgo();
   algoMap["ews"] = newEWSCPPAlgo();
   algoMap["CoOFD"] = newCoOccurringFDCPPAlgo();
   algoMap["bcoofd"] = newBetaCoOFDCPPAlgo();
 }
 
-} // AMMBench
+} // AMMBench

src/CPPAlgos/CRSV2CPPAlgo.cpp

@@ -37,4 +37,4 @@ namespace AMMBench {
 
         return torch::matmul(b, B);
     }
-} // AMMBench
+} // AMMBench

src/CPPAlgos/CountSketchCPPAlgo.cpp

@@ -0,0 +1,33 @@
+//
+// Created by haolan on 5/26/23.
+//
+
+#include <CPPAlgos/CountSketchCPPAlgo.h>
+
+namespace AMMBench {
+    torch::Tensor AMMBench::CountSketchCPPAlgo::amm(torch::Tensor A, torch::Tensor B, int k) {
+        int64_t m1 = A.size(0);
+        int64_t n = A.size(1);
+        int64_t m2 = B.size(1);
+
+        // Initialize sketch matrices
+        torch::Tensor Ca = torch::zeros({m1, k});
+        torch::Tensor Cb = torch::zeros({k, m2});
+
+        torch::Tensor L = torch::randint(k, {n});
+        torch::Tensor G = torch::randint(2, {n});
+
+        // Modify the random column with random sign
+        for (int64_t i = 0; i < n; ++i) {
+            if (G[i].item<int>() == 1) {
+		Ca.index({torch::indexing::Slice(), L[i]}).add_(A.index({torch::indexing::Slice(), i}));
+		Cb.index({L[i], torch::indexing::Slice()}).add_(B.index({i, torch::indexing::Slice()}));
+            } else {
+                Ca.index({torch::indexing::Slice(), L[i]}).sub_(A.index({torch::indexing::Slice(), i}));
+                Cb.index({L[i], torch::indexing::Slice()}).sub_(B.index({i, torch::indexing::Slice()}));
+            }
+        }
+
+        return torch::matmul(Ca, Cb);
+    }
+} // AMMBench

test/SystemTest/SketchTest.cpp

@@ -50,10 +50,22 @@ auto B = torch::rand({(long) aCol, (long) bCol});*/
 TEST_CASE("Test the counter sketch", "[short]")
 {
   int a = 0;
-  runSingleThreadTest("scripts/config_counterSketch.csv");
+  runSingleThreadTest("scripts/config_countSketch.csv");
   // place your test here
   REQUIRE(a == 0);
 }
+
+TEST_CASE("Test Count Sketch in cpp", "[short]")
+{
+  torch::manual_seed(114514);
+  AMMBench::CountSketchCPPAlgo cs;
+  auto A = torch::rand({400, 400});
+  auto B = torch::rand({400, 400});
+  auto realC = torch::matmul(A, B);
+  auto ammC = cs.amm(A, B, 20);
+  double froError = INTELLI::UtilityFunctions::relativeFrobeniusNorm(realC, ammC);
+  REQUIRE(froError < 0.5);
+}
 TEST_CASE("Test Co-Occurring FD in cpp", "[short]")
 {
     torch::manual_seed(114514);
@@ -75,4 +87,5 @@ TEST_CASE("Test Beta-Co-Occurring FD in cpp", "[short]")
     auto ammC = bcoofd.amm(A, B, 20);
     double froError = INTELLI::UtilityFunctions::relativeFrobeniusNorm(realC, ammC);
     REQUIRE(froError < 0.2);
-}
+}
+