diff --git a/.github/workflows/benchmarks.yml b/.github/workflows/benchmarks.yml
index b6a6412d..583a14e9 100644
--- a/.github/workflows/benchmarks.yml
+++ b/.github/workflows/benchmarks.yml
@@ -36,6 +36,12 @@ jobs:
           - algorithm: httpann_example
             dataset: random-range-xs
             library: httpann_example
+          - algorithm: ngt-t1
+            dataset: random-range-xs
+            library: ngt
+          - algorithm: ngt-t2
+            dataset: random-range-xs
+            library: ngt
           - algorithm : kst_ann_t1
             dataset: random-xs
             library: kst_ann_t1
diff --git a/algos.yaml b/algos.yaml
index ff406b7a..9c387639 100644
--- a/algos.yaml
+++ b/algos.yaml
@@ -228,6 +228,30 @@ random-xs:
           args:
             - [ 0.2, 0.8, 1.0 ]
           query-args: [ ]
+    ngt-t1:
+      # ngt-t1: random-xs
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t1
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 10, "s": 600, "b": 10000, "rs": 1000, "ri": 100, "rx": 10}]
+          query-args: |
+              [{"epsilon":0.10, "edge":0, "blob":500}]
+    ngt-t2:
+      # ngt-t2: random-xs
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t1
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 10, "s": 600, "b": 10000, "rs": 1000, "ri": 100, "rx": 10}]
+          query-args: |
+              [{"epsilon":0.10, "edge":0, "blob":500, "expansion":2.0}]
     bbann: 
       docker-tag: billion-scale-benchmark-bbann
       module: benchmark.algorithms.bbann
@@ -429,6 +453,49 @@ deep-1B:
               {"Ls":70, "BW":4, "T":16},
               {"Ls":80, "BW":4, "T":16},
               {"Ls":100, "BW":4, "T":16}]
+    ngt-t1:
+      # ngt-t1: deep-1B
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t1
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 1000000, "s": 100, "b": 1000000, "rs": 1000000, "ri": 600, "rx": 10,
+                "index": "https://ngtf-rlab.east.edge.storage-yahoo.jp/neurips21/indices/1128/Deep1BDataset-1000000000.tar"}]
+          query-args: |
+              [{"epsilon":0.10, "edge":600, "blob":200},
+               {"epsilon":0.15, "edge":600, "blob":200},
+               {"epsilon":0.20, "edge":600, "blob":200},
+               {"epsilon":0.24, "edge":600, "blob":200},
+               {"epsilon":0.26, "edge":600, "blob":200},
+               {"epsilon":0.28, "edge":600, "blob":200},
+               {"epsilon":0.30, "edge":600, "blob":200},
+               {"epsilon":0.32, "edge":600, "blob":200},
+               {"epsilon":0.35, "edge":600, "blob":200},
+               {"epsilon":0.40, "edge":600, "blob":200}]
+    ngt-t2:
+      # ngt-t2: deep-1B
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t2
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 1000000, "s": 100, "b": 1000000, "rs": 1000000, "ri": 600, "rx": 10,
+                "index": "https://ngtf-rlab.east.edge.storage-yahoo.jp/neurips21/indices/1128/Deep1BDataset-1000000000.tar"}]
+          query-args: |
+              [{"epsilon":0.16, "edge":100, "blob":500, "expansion":4.0},
+               {"epsilon":0.16, "edge":100, "blob":500, "expansion":5.0},
+               {"epsilon":0.16, "edge":100, "blob":500, "expansion":6.0},
+               {"epsilon":0.16, "edge":100, "blob":500, "expansion":7.0},
+               {"epsilon":0.16, "edge":100, "blob":500, "expansion":8.0},
+               {"epsilon":0.16, "edge":100, "blob":500, "expansion":9.0},
+               {"epsilon":0.16, "edge":100, "blob":500, "expansion":10.0},
+               {"epsilon":0.16, "edge":100, "blob":500, "expansion":11.0},
+               {"epsilon":0.16, "edge":100, "blob":500, "expansion":12.0}]
 msspacev-1B:
     kst_ann_t1:
       docker-tag: billion-scale-benchmark-kst_ann_t1
@@ -537,6 +604,51 @@ msspacev-1B:
               {"Ls":110, "BW":4, "T":16},
               {"Ls":120, "BW":4, "T":16},
               {"Ls":130, "BW":4, "T":16}]
+    ngt-t1:
+      # ngt-t1: msspacev-1B
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t1
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 400000, "s": 500, "b": 1000000, "rs": 1000000, "ri": 600, "rx": 10,
+                "index": "https://ngtf-rlab.east.edge.storage-yahoo.jp/neurips21/indices/1115/MSSPACEV1B-1000000000.tar"}]
+          query-args: |
+              [{"epsilon":0.10, "edge":0, "blob":500},
+               {"epsilon":0.15, "edge":0, "blob":500},
+               {"epsilon":0.18, "edge":0, "blob":500},
+               {"epsilon":0.20, "edge":0, "blob":500},
+               {"epsilon":0.22, "edge":0, "blob":500},
+               {"epsilon":0.24, "edge":0, "blob":500},
+               {"epsilon":0.26, "edge":0, "blob":500},
+               {"epsilon":0.30, "edge":0, "blob":500},
+               {"epsilon":0.32, "edge":0, "blob":500},
+               {"epsilon":0.34, "edge":0, "blob":500},
+               {"epsilon":0.36, "edge":0, "blob":500}]
+    ngt-t2:
+      # ngt-t2: msspacev-1B
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t2
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 400000, "s": 500, "b": 1000000, "rs": 1000000, "ri": 600, "rx": 10,
+                "index": "https://ngtf-rlab.east.edge.storage-yahoo.jp/neurips21/indices/1115/MSSPACEV1B-1000000000.tar"}]
+          query-args: |
+              [{"epsilon":0.18, "edge":0, "blob":500, "expansion":2},
+               {"epsilon":0.18, "edge":0, "blob":500, "expansion":3},
+               {"epsilon":0.18, "edge":0, "blob":500, "expansion":4},
+               {"epsilon":0.18, "edge":0, "blob":500, "expansion":5},
+               {"epsilon":0.18, "edge":0, "blob":500, "expansion":6},
+               {"epsilon":0.18, "edge":0, "blob":500, "expansion":7},
+               {"epsilon":0.18, "edge":0, "blob":500, "expansion":8},
+               {"epsilon":0.18, "edge":0, "blob":500, "expansion":9},
+               {"epsilon":0.18, "edge":0, "blob":500, "expansion":10}]
+
     bbann: 
       docker-tag: billion-scale-benchmark-bbann
       module: benchmark.algorithms.bbann
@@ -691,6 +803,49 @@ msturing-1B:
               {"Ls":70, "BW":4, "T":16},
               {"Ls":80, "BW":4, "T":16},
               {"Ls":100, "BW":4, "T":16}]
+    ngt-t1:
+      # ngt-t1: msturing-1B:
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t1
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 400000, "s": 200, "b": 1000000, "rs": 1000000, "ri": 600, "rx": 10,
+                "index": "https://ngtf-rlab.east.edge.storage-yahoo.jp/neurips21/indices/1112/MSTuringANNS-1000000000.tar"}]
+          query-args: |
+              [{"epsilon":0.10, "edge":0, "blob":500},
+               {"epsilon":0.12, "edge":0, "blob":500},
+               {"epsilon":0.14, "edge":0, "blob":500},
+               {"epsilon":0.16, "edge":0, "blob":500},
+               {"epsilon":0.18, "edge":0, "blob":500},
+               {"epsilon":0.20, "edge":0, "blob":500},
+               {"epsilon":0.22, "edge":0, "blob":500},
+               {"epsilon":0.24, "edge":0, "blob":500},
+               {"epsilon":0.26, "edge":0, "blob":500},
+               {"epsilon":0.30, "edge":0, "blob":500}]
+    ngt-t2:
+      # ngt-t2: msturing-1B:
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t2
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 400000, "s": 200, "b": 1000000, "rs": 1000000, "ri": 600, "rx": 10,
+                "index": "https://ngtf-rlab.east.edge.storage-yahoo.jp/neurips21/indices/1112/MSTuringANNS-1000000000.tar"}]
+          query-args: |
+              [{"epsilon":0.20, "edge":0, "blob":500, "expansion":1.0},
+               {"epsilon":0.20, "edge":0, "blob":500, "expansion":1.2},
+               {"epsilon":0.20, "edge":0, "blob":500, "expansion":1.4},
+               {"epsilon":0.20, "edge":0, "blob":500, "expansion":1.6},
+               {"epsilon":0.20, "edge":0, "blob":500, "expansion":1.8},
+               {"epsilon":0.20, "edge":0, "blob":500, "expansion":2.0},
+               {"epsilon":0.20, "edge":0, "blob":500, "expansion":2.2},
+               {"epsilon":0.20, "edge":0, "blob":500, "expansion":2.4},
+               {"epsilon":0.20, "edge":0, "blob":500, "expansion":2.6}]
 bigann-1B:
     kst_ann_t1:
       docker-tag: billion-scale-benchmark-kst_ann_t1
@@ -847,6 +1002,56 @@ bigann-1B:
               {"Ls":70, "BW":4, "T":16},
               {"Ls":80, "BW":4, "T":16},
               {"Ls":100, "BW":4, "T":16}]
+    ngt-t1:
+      # ngt-t1: bigann-1B
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t1
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 500000, "s": 500, "b": 1000000, "rs": 1000000, "ri": 600, "rx": 10,
+                "index": "https://ngtf-rlab.east.edge.storage-yahoo.jp/neurips21/indices/1119/BigANNDataset-1000000000.tar"}]
+          query-args: |
+              [{"epsilon":0.10, "edge":600, "blob":200},
+               {"epsilon":0.20, "edge":600, "blob":200},
+               {"epsilon":0.25, "edge":600, "blob":200},
+               {"epsilon":0.27, "edge":600, "blob":200},
+               {"epsilon":0.29, "edge":600, "blob":200},
+               {"epsilon":0.30, "edge":600, "blob":200},
+               {"epsilon":0.31, "edge":600, "blob":200},
+               {"epsilon":0.32, "edge":600, "blob":200},
+               {"epsilon":0.33, "edge":600, "blob":200},
+               {"epsilon":0.34, "edge":600, "blob":200},
+               {"epsilon":0.35, "edge":600, "blob":200},
+               {"epsilon":0.40, "edge":600, "blob":200},
+               {"epsilon":0.45, "edge":600, "blob":200},
+               {"epsilon":0.50, "edge":600, "blob":200}]
+    ngt-t2:
+      # ngt-t2: bigann-1B
+      docker-tag: billion-scale-benchmark-ngt
+      module: benchmark.algorithms.ngt_t2
+      constructor: NGT
+      base-args: ["@metric"]
+      run-groups:
+        base:
+          args: |
+              [{"q": 500000, "s": 500, "b": 1000000, "rs": 1000000, "ri": 600, "rx": 10,
+                "index": "https://ngtf-rlab.east.edge.storage-yahoo.jp/neurips21/indices/1119/BigANNDataset-1000000000.tar"}]
+          query-args: |
+              [{"epsilon":0.18, "edge":600, "blob":1000, "expansion":2},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":3},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":4},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":5},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":6},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":7},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":8},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":9},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":10},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":11},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":12},
+               {"epsilon":0.18, "edge":600, "blob":1000, "expansion":13}]
 ssnpp-1B:
     kota-t2:
       docker-tag: billion-scale-benchmark-kota
diff --git a/benchmark/algorithms/ngt_t1.py b/benchmark/algorithms/ngt_t1.py
new file mode 100644
index 00000000..16a55c5d
--- /dev/null
+++ b/benchmark/algorithms/ngt_t1.py
@@ -0,0 +1,122 @@
+from __future__ import absolute_import
+import numpy as np
+import os
+import time
+import subprocess
+from benchmark.algorithms.base import BaseANN
+from benchmark.datasets import DATASETS, download_accelerated
+
+import ngtpy
+
+class NGT(BaseANN):
+    def __init__(self, metric, params):
+        self._params = params
+        self._metric = metric
+        self._quantization = params.get("q", 1024)
+        self._quantization_sample = params.get("s", 100)
+        self._blob = params.get("b", 10000)
+        self._num_of_r_samples = params.get("rs", 1000)
+        self._num_of_r_iterations = params.get("ri", 10)
+        self._r_step = params.get("rx", 2)
+        #self._ngt_root = "data/ngt"  # debug
+        self._ngt_root = "ngt"
+        self._ngt_index_root = self._ngt_root + "/indexes/"
+        self._is_open = False
+
+    def setIndexPath(self, dataset):
+        self._path = f"data/indices/trackT1/algo.NGT:q{self._quantization}-s{self._quantization_sample}-b{self._blob}-rs{self._num_of_r_samples}-ri{self._num_of_r_iterations}"
+        os.makedirs(self._path, exist_ok=True)
+        self._index_path = os.path.join(self._path, DATASETS[dataset]().short_name())
+        
+    def fit(self, dataset):
+        index_params = self._params
+        ds = DATASETS[dataset]()
+        if ds.d <= 128:
+            pseudo_dimension = 128
+            subvector_dimension = 2
+        elif ds.d <= 256:
+            pseudo_dimension = 256
+            subvector_dimension = 4
+        self.setIndexPath(dataset)
+        print("NGT dataset:", dataset)
+        print("NGT dataset str:", ds.__str__())
+        print("NGT distance:", ds.distance())
+        print("NGT dimension:", ds.d)
+        print("NGT type:", ds.dtype)
+        print("NGT nb:", ds.nb)
+        print("NGT dataset file name:", ds.get_dataset_fn())
+        print("NGT quantization (q):", self._quantization)
+        print("NGT quantization sample (s):", self._quantization_sample)
+        print("NGT blob (b):", self._blob)
+        print("NGT # of r samples (rs):", self._num_of_r_samples)
+        print("NGT # of r iterations (rs):", self._num_of_r_iterations)
+        print("NGT ngt root:", self._ngt_root)
+        print("NGT index path:", self._index_path)
+        print("NGT build.sh:", self._ngt_root + '/build.sh')
+        args = ['/bin/bash', self._ngt_root + '/build.sh',
+                #'--search',
+                '--root=' + self._ngt_root, 
+                '--object=' + ds.get_dataset_fn(),
+                '--benchmark=' + self._index_path, '-d=' + str(pseudo_dimension),
+                '-D=' + str(subvector_dimension),
+                '-f', '-m=KMEANS_QUANTIZATION', '-n=' + str(ds.nb),
+                '-q=' + str(self._quantization), '-E=' + str(self._quantization_sample),
+                '-b=' + str(self._blob),
+                '-r=' + str(self._num_of_r_samples), '-R=' + str(self._num_of_r_iterations),
+                '-X=' + str(self._r_step)]
+        print(args)
+        subprocess.call(args)
+
+        if not self._is_open:
+            print("NGT: opening the index...")
+            self._index = ngtpy.QuantizedBlobIndex(self._index_path)
+            self._is_open = True
+
+    def load_index(self, dataset):
+        self.setIndexPath(dataset)
+
+        if not os.path.exists(self._index_path + "/grp"):
+            if "index" not in self._params:
+                return False
+            if not os.path.exists(self._index_path + ".tar"):
+                print(f"NGT: dowinloading the index... index={self._params['index']}->{self._index_path}")
+                download_accelerated(self._params["index"], self._index_path + ".tar", quiet=True)
+            args = ['tar', 'xf', self._index_path + ".tar", "-C", self._path]
+            print(args)
+            subprocess.call(args)
+            args = ['rm', '-r', self._index_path + ".tar"]
+            print(args)
+            subprocess.call(args)
+            
+        if not self._is_open:
+            print("NGT: opening the index...")
+            self._index = ngtpy.QuantizedBlobIndex(self._index_path)
+            self._is_open = True
+
+    def set_query_arguments(self, query_args):
+        self._epsilon = query_args.get("epsilon", 0.1)
+        self._edge_size = query_args.get("edge", 0)
+        self._exploration_size = query_args.get("blob", 120)
+        # only this part is different between t1 and t2
+        #self._exact_result_expansion = query_args.get("expansion", 2.0)
+        self._exact_result_expansion = 0.0
+        self._index.set(epsilon=self._epsilon, blob_epsilon=0.0, edge_size=self._edge_size,
+                        exploration_size=self._exploration_size,
+                        exact_result_expansion=self._exact_result_expansion)
+        print(f"NGT: epsilon={self._epsilon} edge={self._edge_size} blob={self._exploration_size}")
+
+    def query(self, X, n):
+        self._results = ngtpy.BatchResults()
+        self._index.batchSearch(X, self._results, n)
+
+    def range_query(self, X, radius):
+        print("NGT: range_query")
+
+    def get_results(self):
+        return self._results.getIDs()
+
+    def get_range_results(self):
+        return self._results.getIndex(), self._results.getIndexedIDs(), self._results.getIndexedDistances()
+
+    def __str__(self):
+        return f"NGT-T1:q{self._quantization}-b{self._blob}-rs{self._num_of_r_samples}-ri{self._num_of_r_iterations}-e{self._epsilon:.3f}-b{self._exploration_size}"
diff --git a/benchmark/algorithms/ngt_t2.py b/benchmark/algorithms/ngt_t2.py
new file mode 100644
index 00000000..7fac62c1
--- /dev/null
+++ b/benchmark/algorithms/ngt_t2.py
@@ -0,0 +1,122 @@
+from __future__ import absolute_import
+import numpy as np
+import os
+import time
+import subprocess
+from benchmark.algorithms.base import BaseANN
+from benchmark.datasets import DATASETS, download_accelerated
+
+import ngtpy
+
+class NGT(BaseANN):
+    def __init__(self, metric, params):
+        self._params = params
+        self._metric = metric
+        self._quantization = params.get("q", 1024)
+        self._quantization_sample = params.get("s", 100)
+        self._blob = params.get("b", 10000)
+        self._num_of_r_samples = params.get("rs", 1000)
+        self._num_of_r_iterations = params.get("ri", 10)
+        self._r_step = params.get("rx", 2)
+        #self._ngt_root = "data/ngt"  # debug
+        self._ngt_root = "ngt"
+        self._ngt_index_root = self._ngt_root + "/indexes/"
+        self._is_open = False
+
+    def setIndexPath(self, dataset):
+        self._path = f"data/indices/trackT2/algo.NGT:q{self._quantization}-s{self._quantization_sample}-b{self._blob}-rs{self._num_of_r_samples}-ri{self._num_of_r_iterations}"
+        os.makedirs(self._path, exist_ok=True)
+        self._index_path = os.path.join(self._path, DATASETS[dataset]().short_name())
+        
+    def fit(self, dataset):
+        index_params = self._params
+        ds = DATASETS[dataset]()
+        if ds.d <= 128:
+            pseudo_dimension = 128
+            subvector_dimension = 2
+        elif ds.d <= 256:
+            pseudo_dimension = 256
+            subvector_dimension = 4
+        self.setIndexPath(dataset)
+        print("NGT dataset:", dataset)
+        print("NGT dataset str:", ds.__str__())
+        print("NGT distance:", ds.distance())
+        print("NGT dimension:", ds.d)
+        print("NGT type:", ds.dtype)
+        print("NGT nb:", ds.nb)
+        print("NGT dataset file name:", ds.get_dataset_fn())
+        print("NGT quantization (q):", self._quantization)
+        print("NGT quantization sample (s):", self._quantization_sample)
+        print("NGT blob (b):", self._blob)
+        print("NGT # of r samples (rs):", self._num_of_r_samples)
+        print("NGT # of r iterations (rs):", self._num_of_r_iterations)
+        print("NGT ngt root:", self._ngt_root)
+        print("NGT index path:", self._index_path)
+        print("NGT build.sh:", self._ngt_root + '/build.sh')
+        args = ['/bin/bash', self._ngt_root + '/build.sh',
+                #'--search',
+                '--root=' + self._ngt_root, 
+                '--object=' + ds.get_dataset_fn(),
+                '--benchmark=' + self._index_path, '-d=' + str(pseudo_dimension),
+                '-D=' + str(subvector_dimension),
+                '-f', '-m=KMEANS_QUANTIZATION', '-n=' + str(ds.nb),
+                '-q=' + str(self._quantization), '-E=' + str(self._quantization_sample),
+                '-b=' + str(self._blob),
+                '-r=' + str(self._num_of_r_samples), '-R=' + str(self._num_of_r_iterations),
+                '-X=' + str(self._r_step)]
+        print(args)
+        subprocess.call(args)
+
+        if not self._is_open:
+            print("NGT: opening the index...")
+            self._index = ngtpy.QuantizedBlobIndex(self._index_path)
+            self._is_open = True
+
+    def load_index(self, dataset):
+        self.setIndexPath(dataset)
+
+        if not os.path.exists(self._index_path + "/grp"):
+            if "index" not in self._params:
+                return False
+            if not os.path.exists(self._index_path + ".tar"):
+                print(f"NGT: dowinloading the index... index={self._params['index']}->{self._index_path}")
+                download_accelerated(self._params["index"], self._index_path + ".tar", quiet=True)
+            args = ['tar', 'xf', self._index_path + ".tar", "-C", self._path]
+            print(args)
+            subprocess.call(args)
+            args = ['rm', '-r', self._index_path + ".tar"]
+            print(args)
+            subprocess.call(args)
+
+        if not self._is_open:
+            print("NGT: opening the index...")
+            self._index = ngtpy.QuantizedBlobIndex(self._index_path)
+            self._is_open = True
+
+    def set_query_arguments(self, query_args):
+        self._epsilon = query_args.get("epsilon", 0.1)
+        self._edge_size = query_args.get("edge", 0)
+        self._exploration_size = query_args.get("blob", 120)
+        # only this part is different between t1 and t2
+        self._exact_result_expansion = query_args.get("expansion", 2.0)
+        #self._exact_result_expansion = 0.0
+        self._index.set(epsilon=self._epsilon, blob_epsilon=0.0, edge_size=self._edge_size,
+                        exploration_size=self._exploration_size,
+                        exact_result_expansion=self._exact_result_expansion)
+        print(f"NGT: epsilon={self._epsilon} edge={self._edge_size} blob={self._exploration_size}")
+
+    def query(self, X, n):
+        self._results = ngtpy.BatchResults()
+        self._index.batchSearch(X, self._results, n)
+
+    def range_query(self, X, radius):
+        print("NGT: range_query")
+
+    def get_results(self):
+        return self._results.getIDs()
+
+    def get_range_results(self):
+        return self._results.getIndex(), self._results.getIndexedIDs(), self._results.getIndexedDistances()
+
+    def __str__(self):
+        return f"NGT-T2:q{self._quantization}-b{self._blob}-rs{self._num_of_r_samples}-ri{self._num_of_r_iterations}-e{self._epsilon:.3f}-b{self._exploration_size}-x{self._exact_result_expansion}"
diff --git a/install/Dockerfile.ngt b/install/Dockerfile.ngt
new file mode 100644
index 00000000..25679fe8
--- /dev/null
+++ b/install/Dockerfile.ngt
@@ -0,0 +1,13 @@
+FROM billion-scale-benchmark
+
+RUN apt-get update
+RUN apt-get install -y git cmake liblapack-dev bc
+RUN pip3 install wheel pybind11
+RUN git clone -b neurips21 https://github.com/yahoojapan/NGT.git
+RUN cd NGT && mkdir -p build && cd build && cmake .. 
+RUN cd NGT/build && make -j 8 && make install
+RUN mkdir ngt; cp NGT/bin/build.sh ngt/
+RUN ldconfig
+RUN cd NGT/python && python3 setup.py bdist_wheel
+RUN pip3 install NGT/python/dist/ngt-*-linux_x86_64.whl
+