Exp: Basic setup and DiskANN experiments

.gitignore

@@ -0,0 +1,2 @@
+.vscode
+build

algorithms/utils/check_nn_recall.h

@@ -127,7 +127,7 @@ nn_result checkRecall(const Graph<indexType> &G,
   return N;
 }
 
-void write_to_csv(std::string csv_filename, parlay::sequence<float> buckets,
+void write_to_csv(std::string csv_filename,
                   parlay::sequence<nn_result> results, Graph_ G) {
   csvfile csv(csv_filename);
   csv << "GRAPH"
@@ -140,7 +140,6 @@ void write_to_csv(std::string csv_filename, parlay::sequence<float> buckets,
       << endrow;
   csv << endrow;
   csv << "Num queries"
-      << "Target recall"
       << "Actual recall"
       << "QPS"
       << "Average Cmps"
@@ -152,7 +151,7 @@ void write_to_csv(std::string csv_filename, parlay::sequence<float> buckets,
       << "cut" << endrow;
   for (int i = 0; i < results.size(); i++) {
     nn_result N = results[i];
-    csv << N.num_queries << buckets[i] << N.recall << N.QPS << N.avg_cmps
+    csv << N.num_queries << N.recall << N.QPS << N.avg_cmps
         << N.tail_cmps << N.avg_visited << N.tail_visited << N.k << N.beamQ
         << N.cut << endrow;
   }
@@ -214,59 +213,27 @@ void search_and_parse(Graph_ G_,
   QP.limit = (long) G.size();
   QP.rerank_factor = rerank_factor;
   QP.degree_limit = (long) G.max_degree();
-  beams = {10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 22, 24, 26, 28, 30, 32,
-    34, 36, 38, 40, 45, 50, 55, 60, 65, 70, 80, 90, 100, 120, 140, 160,
-    180, 200, 225, 250, 275, 300, 375, 500, 750, 1000};
+  beams = {10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 125, 150, 200, 250, 275, 300, 375, 500, 750, 1000};
   if(k==0) allr = {10};
   else allr = {k};
   cuts = {1.35};
-
-  if (fixed_beam_width != 0) {
-    QP.k = allr[0];
-    QP.cut = cuts[0];
-    QP.beamSize = fixed_beam_width;
-    for (int i = 0; i < 5; i++)
-      check(QP.k, QP);
-  } else {
-    for (long r : allr) {
-      results.clear();
-      QP.k = r;
-      for (float cut : cuts){
-        QP.cut = cut;
-        for (float Q : beams){
-          QP.beamSize = Q;
-          if (Q >= r){
-            results.push_back(check(r, QP));
-          }
+
+  std::cout<<"Fixed beam width: "<<fixed_beam_width<<std::endl;
+  for (long r : allr) {
+    results.clear();
+    QP.k = r;
+    for (float cut : cuts){
+      QP.cut = cut;
+      for (float Q : beams){
+        QP.beamSize = Q;
+        if (Q >= r){
+          results.push_back(check(r, QP));
         }
       }
-
-      // check "limited accuracy"
-      // {10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 28, 30, 35}; //
-      parlay::sequence<long> limits = {10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 28, 30, 35};
-      //calculate_limits(results[0].avg_visited);
-      //parlay::sequence<long> degree_limits = calculate_limits(G.max_degree());
-      //degree_limits.push_back(G.max_degree());
-      QP = QueryParams(r, r, 1.35, (long) G.size(), (long) G.max_degree());
-      for(long l : limits){
-        QP.limit = l;
-        QP.beamSize = std::max<long>(l, r);
-        //for(long dl : degree_limits){
-        QP.degree_limit = std::min<int>(G.max_degree(), 5 * l);
-        results.push_back(check(r, QP));
-      }
-      // check "best accuracy"
-      QP = QueryParams((long) 100, (long) 1000, (double) 10.0, (long) G.size(), (long) G.max_degree());
-      results.push_back(check(r, QP));
-
-      parlay::sequence<float> buckets =  {.1, .2, .3,  .4,  .5,  .6, .7, .75,  .8, .85,
-        .9, .93, .95, .97, .98, .99, .995, .999, .9995,
-        .9999, .99995, .99999};
-      auto [res, ret_buckets] = parse_result(results, buckets);
-      std::cout << std::endl;
-      if (res_file != NULL)
-        write_to_csv(std::string(res_file), ret_buckets, res, G_);
     }
+    std::cout << std::endl;
+    if (res_file != NULL)
+      write_to_csv(std::string(res_file), results, G_);
   }
 }
 

algorithms/vamana/parlay_vamana_results.csv

@@ -0,0 +1,140 @@
+"GRAPH","Parameters","Size","Build time","Avg degree","Max degree"
+"Vamana","R = 64, L = 100",1010812,306.005,47.5557,64
+
+"Num queries","Target recall","Actual recall","QPS","Average Cmps","Tail Cmps","Average Visited","Tail Visited","k","Q","cut"
+13265,0.1,0.144109,107123,469,490,10,10,10,10,1.35
+13265,0.2,0.236932,72406,670,703,12,12,10,12,1.35
+13265,0.3,0.305435,57168.8,828,871,14,14,10,14,1.35
+13265,0.4,0.414489,41645.3,1110,1175,19,19,10,19,1.35
+13265,0.5,0.509242,33372,1376,2327,23,40,10,14,1.35
+13265,0.6,0.602314,23899,1882,3145,32,54,10,22,1.35
+13265,0.7,0.703777,15759.5,2823,4520,49,78,10,38,1.35
+13265,0.75,0.760671,11730.2,3784,5660,66,98,10,55,1.35
+13265,0.8,0.812439,8598.29,5064,7278,92,129,10,80,1.35
+13265,0.85,0.858093,6123.03,6994,9707,133,174,10,120,1.35
+13265,0.9,0.902676,3913.39,10516,14071,218,261,10,200,1.35
+13265,0.93,0.930418,2714.22,14857,19754,324,372,10,300,1.35
+13265,0.95,0.95663,1717.45,23216,30743,537,596,10,500,1.35
+13265,0.97,0.971964,1182.04,32974,43910,816,899,10,750,1.35
+
+
+"GRAPH","Parameters","Size","Build time","Avg degree","Max degree"
+"Vamana","R = 32, L = 50",1010812,70.5576,22.9389,32
+
+"Num queries","Target recall","Actual recall","QPS","Average Cmps","Tail Cmps","Average Visited","Tail Visited","k","Q","cut"
+13265,0.1,0.103468,114707,397,427,14,14,10,14,1.35
+13265,0.2,0.224109,80493.9,588,1039,20,37,10,10,1.35
+13265,0.3,0.301357,59723.5,784,1373,28,50,10,15,1.35
+13265,0.4,0.406672,33859.8,1150,2002,42,73,10,24,1.35
+13265,0.5,0.50634,23632.8,1666,2867,61,104,10,40,1.35
+13265,0.6,0.602133,15665,2474,4190,95,159,10,70,1.35
+13265,0.7,0.701153,8997.7,4163,6616,174,265,10,140,1.35
+13265,0.75,0.755665,6040.98,6092,8989,267,372,10,225,1.35
+13265,0.8,0.806385,3849.66,9292,13149,437,548,10,375,1.35
+13265,0.85,0.861772,2048.92,16941,23427,861,999,10,750,1.35
+
+
+"GRAPH","Parameters","Size","Build time","Avg degree","Max degree"
+"Vamana","R = 32, L = 50",1010812,0,22.9389,32
+
+"Num queries","Target recall","Actual recall","QPS","Average Cmps","Tail Cmps","Average Visited","Tail Visited","k","Q","cut"
+13265,0.1,0.103468,114800,397,427,14,14,10,14,1.35
+13265,0.2,0.208194,68717.4,615,661,22,22,10,22,1.35
+13265,0.3,0.339299,42290.3,961,1038,35,35,10,35,1.35
+13265,0.4,0.406672,31805.1,1150,2002,42,73,10,24,1.35
+13265,0.5,0.50634,17573.4,1666,2867,61,104,10,40,1.35
+13265,0.6,0.602133,12943.3,2474,4190,95,159,10,70,1.35
+13265,0.7,0.701153,7510.01,4163,6616,174,265,10,140,1.35
+13265,0.75,0.766174,5438.07,6659,9652,294,396,10,250,1.35
+13265,0.8,0.806385,3811.94,9292,13149,437,548,10,375,1.35
+13265,0.85,0.861772,2021.44,16941,23427,861,999,10,750,1.35
+
+
+"GRAPH","Parameters","Size","Build time","Avg degree","Max degree"
+"Vamana","R = 32, L = 50",1010812,0,22.9389,32
+
+"Num queries","Target recall","Actual recall","QPS","Average Cmps","Tail Cmps","Average Visited","Tail Visited","k","Q","cut"
+13265,0.1,0.103468,124846,397,427,14,14,10,14,1.35
+13265,0.2,0.208194,74976.1,615,661,22,22,10,22,1.35
+13265,0.3,0.339299,46272.7,961,1038,35,35,10,35,1.35
+13265,0.4,0.406672,30427.5,1150,2002,42,73,10,24,1.35
+13265,0.5,0.50634,21626.9,1666,2867,61,104,10,40,1.35
+13265,0.6,0.602133,14680.3,2474,4190,95,159,10,70,1.35
+13265,0.7,0.701153,8501.88,4163,6616,174,265,10,140,1.35
+13265,0.75,0.755665,5717.47,6092,8989,267,372,10,225,1.35
+13265,0.8,0.806385,3848.63,9292,13149,437,548,10,375,1.35
+13265,0.85,0.861772,2068.67,16941,23427,861,999,10,750,1.35
+
+
+"GRAPH","Parameters","Size","Build time","Avg degree","Max degree"
+"Vamana","R = 32, L = 50",1010812,0,22.9389,32
+
+"Num queries","Target recall","Actual recall","QPS","Average Cmps","Tail Cmps","Average Visited","Tail Visited","k","Q","cut"
+13265,0.1,0.103468,127823,397,427,14,14,10,14,1.35
+13265,0.2,0.208194,76946.7,615,661,22,22,10,22,1.35
+13265,0.3,0.339299,48126.8,961,1038,35,35,10,35,1.35
+13265,0.4,0.406672,29702.7,1150,2002,42,73,10,24,1.35
+13265,0.5,0.50634,21150.6,1666,2867,61,104,10,40,1.35
+13265,0.6,0.602133,15453.9,2474,4190,95,159,10,70,1.35
+13265,0.7,0.701153,9316.64,4163,6616,174,265,10,140,1.35
+13265,0.75,0.755665,6396.81,6092,8989,267,372,10,225,1.35
+13265,0.8,0.806385,4473.85,9292,13149,437,548,10,375,1.35
+13265,0.85,0.861772,2390.13,16941,23427,861,999,10,750,1.35
+
+
+"GRAPH","Parameters","Size","Build time","Avg degree","Max degree"
+"Vamana","R = 32, L = 50",1010812,0,22.9389,32
+
+"Num queries","Target recall","Actual recall","QPS","Average Cmps","Tail Cmps","Average Visited","Tail Visited","k","Q","cut"
+13265,0.1,0.110939,100768,452,486,16,16,25,25,1.35
+13265,0.2,0.200127,64993.9,669,720,24,24,25,25,1.35
+13265,0.3,0.373823,32257,1211,2104,44,77,25,26,1.35
+13265,0.4,0.414583,22561.9,1417,2463,52,90,25,32,1.35
+13265,0.5,0.508722,17233.9,2084,3506,77,130,25,55,1.35
+13265,0.6,0.607846,11025,3228,5351,129,209,25,100,1.35
+13265,0.7,0.701066,6274.72,5530,8388,241,345,25,200,1.35
+13265,0.75,0.770651,3945.84,9296,13163,437,550,25,375,1.35
+13265,0.8,0.834961,2099.69,16991,23380,864,1000,25,750,1.35
+13265,0.85,0.857399,1699.7,21957,30131,1145,1296,25,1000,10
+
+
+"GRAPH","Parameters","Size","Build time","Avg degree","Max degree"
+"Vamana","R = 32, L = 50",1010812,0,22.9389,32
+
+"Num queries","Actual recall","QPS","Average Cmps","Tail Cmps","Average Visited","Tail Visited","k","Q","cut"
+13265,0.367098,31378.3,1181,2051,43,74,25,25,1.35
+13265,0.399421,28504.7,1326,2288,48,83,25,30,1.35
+13265,0.45543,24822.8,1666,2867,61,104,25,40,1.35
+13265,0.493859,21947,1955,3318,73,122,25,50,1.35
+13265,0.561873,16572.1,2617,4448,101,168,25,75,1.35
+13265,0.607846,13324.7,3228,5351,129,209,25,100,1.35
+13265,0.638568,11353.4,3841,6162,156,244,25,125,1.35
+13265,0.663786,9875.83,4392,6908,184,278,25,150,1.35
+13265,0.701066,7719.68,5530,8388,241,345,25,200,1.35
+13265,0.725877,6385.63,6659,9652,294,396,25,250,1.35
+13265,0.746491,5468.08,7724,11043,352,456,25,300,1.35
+13265,0.777294,4250.6,9841,13975,465,580,25,400,1.35
+13265,0.798254,3486.96,11936,16702,575,692,25,500,1.35
+13265,0.0436457,192979,285,308,10,10,25,25,1.35
+13265,0.0538832,170661,313,338,11,11,25,25,1.35
+13265,0.0649167,154255,341,368,12,12,25,25,1.35
+13265,0.0760588,139679,369,398,13,13,25,25,1.35
+13265,0.087611,129670,397,427,14,14,25,25,1.35
+13265,0.0992145,121226,425,456,15,15,25,25,1.35
+13265,0.110939,111702,452,486,16,16,25,25,1.35
+13265,0.122934,104539,479,515,17,17,25,25,1.35
+13265,0.134423,99122.7,507,545,18,18,25,25,1.35
+13265,0.145833,92399.7,534,574,19,19,25,25,1.35
+13265,0.157141,87719.9,561,603,20,20,25,25,1.35
+13265,0.168271,82704.7,588,632,21,21,25,25,1.35
+13265,0.179175,79105.7,615,661,22,22,25,25,1.35
+13265,0.189735,74748,642,690,23,23,25,25,1.35
+13265,0.200127,72427,669,720,24,24,25,25,1.35
+13265,0.21034,67823.6,696,749,25,25,25,25,1.35
+13265,0.21989,66788.8,722,778,26,26,25,26,1.35
+13265,0.238779,61644.1,776,835,28,28,25,28,1.35
+13265,0.256733,56312.1,829,893,30,30,25,30,1.35
+13265,0.297538,48058.5,961,1038,35,35,25,35,1.35
+13265,0.857399,1816.09,21957,30131,1145,1296,25,1000,10
+
+

algorithms/vamana/results.csv

@@ -0,0 +1,24 @@
+"GRAPH","Parameters","Size","Build time","Avg degree","Max degree"
+"Vamana","R = 32, L = 50",1010812,0,22.9389,32
+
+"Num queries","Actual recall","QPS","Average Cmps","Tail Cmps","Average Visited","Tail Visited","k","Q","cut"
+13265,0.399421,28989.3,1326,2288,48,83,25,30,1.35
+13265,0.45543,24088.3,1666,2867,61,104,25,40,1.35
+13265,0.493859,21535,1955,3318,73,122,25,50,1.35
+13265,0.524381,19881.5,2240,3807,83,140,25,60,1.35
+13265,0.551089,18115,2474,4190,95,159,25,70,1.35
+13265,0.573012,16326.2,2761,4644,107,175,25,80,1.35
+13265,0.590266,14933.4,3018,5017,117,192,25,90,1.35
+13265,0.607846,13831.8,3228,5351,129,209,25,100,1.35
+13265,0.638568,11610.9,3841,6162,156,244,25,125,1.35
+13265,0.663786,9950.91,4392,6908,184,278,25,150,1.35
+13265,0.701066,7042.62,5530,8388,241,345,25,200,1.35
+13265,0.725877,6171.78,6659,9652,294,396,25,250,1.35
+13265,0.737336,5694.58,7184,10438,325,431,25,275,1.35
+13265,0.746491,5303.66,7724,11043,352,456,25,300,1.35
+13265,0.770651,4327.17,9296,13163,437,550,25,375,1.35
+13265,0.798254,3454.53,11936,16702,575,692,25,500,1.35
+13265,0.834961,2360.3,16991,23380,864,1000,25,750,1.35
+13265,0.857487,1824.55,21959,30128,1145,1295,25,1000,1.35
+
+

build/_deps/parlaylib-subbuild/CMakeLists.txt

@@ -1,7 +1,7 @@
 # Distributed under the OSI-approved BSD 3-Clause License.  See accompanying
 # file Copyright.txt or https://cmake.org/licensing for details.
 
-cmake_minimum_required(VERSION 3.25.1)
+cmake_minimum_required(VERSION 3.22.1)
 
 # We name the project and the target for the ExternalProject_Add() call
 # to something that will highlight to the user what we are working on if
@@ -13,24 +13,23 @@ project(parlaylib-populate NONE)
 # Pass through things we've already detected in the main project to avoid
 # paying the cost of redetecting them again in ExternalProject_Add()
 set(GIT_EXECUTABLE [==[/usr/bin/git]==])
-set(GIT_VERSION_STRING [==[2.39.2]==])
+set(GIT_VERSION_STRING [==[2.34.1]==])
 set_property(GLOBAL PROPERTY _CMAKE_FindGit_GIT_EXECUTABLE_VERSION
-  [==[/usr/bin/git;2.39.2]==]
+  [==[/usr/bin/git;2.34.1]==]
 )
 
 
 include(ExternalProject)
 ExternalProject_Add(parlaylib-populate
                      "UPDATE_DISCONNECTED" "False" "GIT_REPOSITORY" "https://github.com/cmuparlay/parlaylib.git" "GIT_TAG" "master"
-                    SOURCE_DIR          "/home/laxmand_google_com/ParlayANN/build/_deps/parlaylib-src"
-                    BINARY_DIR          "/home/laxmand_google_com/ParlayANN/build/_deps/parlaylib-build"
+                    SOURCE_DIR          "/home/rakri/avarhade/ParlayANN/build/_deps/parlaylib-src"
+                    BINARY_DIR          "/home/rakri/avarhade/ParlayANN/build/_deps/parlaylib-build"
                     CONFIGURE_COMMAND   ""
                     BUILD_COMMAND       ""
                     INSTALL_COMMAND     ""
                     TEST_COMMAND        ""
                     USES_TERMINAL_DOWNLOAD  YES
                     USES_TERMINAL_UPDATE    YES
-                    USES_TERMINAL_PATCH     YES
 )
 
 

hybrid-ann.sh

@@ -0,0 +1,9 @@
+# cd algorithms/vamana
+# make
+# ./neighbors -R 32 -L 50 -alpha 1.2 -graph_outfile /nvmessd1/fbv4/avarhade/parlay_prec1M_32_50 -query_path /nvmessd1/fbv4/queries.fbin -gt_path /nvmessd1/fbv4/prec1M_gt100.bin -res_path parlay_vamana_res.csv -data_type float -dist_func Euclidian -base_path /nvmessd1/fbv4/prec1M.fbin
+# ./neighbors -R 32 -L 50 -k 25 -alpha 1.2 -graph_path /nvmessd1/fbv4/avarhade/parlay_prec1M_32_50 -query_path /nvmessd1/fbv4/queries.fbin -gt_path /nvmessd1/fbv4/prec1M_gt100.bin -res_path results.csv -data_type float -dist_func Euclidian -base_path /nvmessd1/fbv4/prec1M.fbin
+
+cd algorithms/HNSW
+make 
+
+./neighbors -m 32 -efc 50 -alpha 1.2 -ml 0.0 -graph_outfile /nvmessd1/fbv4/avarhade/parlay_hnsw_prec1M_32_50 -query_path /nvmessd1/fbv4/queries.fbin -gt_path /nvmessd1/fbv4/prec1M_gt100.bin -res_path hnsw_results.csv -data_type float -dist_func Euclidian -base_path /nvmessd1/fbv4/prec1M.fbin