It would be great to get some ANN performance benchmarks for SPTAG for single node and distributed. The repo doesn't say how well it scales.

My team wanted to use FAISS a few months ago but there was a problem with the License so we are now using HNSWlib even though it isn't quite as performant. We would love to know if this is a viable alternative to FAISS

https://github.com/erikbern/ann-benchmarks is a great repo for ANN benchmarks and it would be great if you could help the author add SPTAG benchmarks for single node and multiple node setups. The author may have some difficulties setting everything up.

If you can't help out in the above repo. It would still be useful to show your own performance benchmarks in the README.

I installed the latest version of TBB through anaconda (https://anaconda.org/intel/tbb), but for some reason, every time I try to build SPTAG, it keeps stopping with "CMake Error at CMakeLists.txt:109 (message): Could not find TBB!"

Does anyone know a work-around for this? In the case that this is due to a faulty TBB installation, does anyone have some insight on how to properly install it on their windows machine (Visual Studio 2019)? All the available TBB installation documentation that I found are very outdated.

I used the code in demo.

The main function is :

The error message is:

Thanks for open sourcing, however its written that it supports distributed serving, and its unclear in the documentation how to do it. Is it all about loading same index into multiple machines ?

Thanks

Describe the bug
Following #65 . The size of the hash table of visited nodes for NGQueue is set to 16394 ("two" hash table of size 8192). However, when searching in a large data-set and setting the "MaxCheck" to a relatively large number (i.e. 16394), the hash table would be full during the searching process. After that, every node that is not in the hash table (either visit for first time or not first time) would result in a -1 return value from _CheckAndSet function in WorkSpace.h. Since the CheckAndSet function only check if the return value from _CheckAndSet is equal to 0, all those nodes with -1 return value would be added to the NGQueue, causing duplicated search and inaccurate result (i.e. duplicate) not only during actual searching process, but also during the refine stage of the building process (which use searching result to update RNG).

To Reproduce
Steps to reproduce the behavior:

1. Build with a large data-set. (theoretically, a data-set larger than 16394 would be above to reproduce using extreme parameter, but a large data-set, i.e. 1M, would be easier to reproduce)
2. The behavior is already reproduced during refine stage of the building process.
3. Search with "MaxCheck" setting to a number larger than 16394.
4. The result of the search may contains duplicate result.

Expected behavior
Either:
Hash table is large enough to store all the visited node.
Or:
Proper handle of -1 return value from _CheckAndSet function or proper handle of the situation that hash table becomes full.

Desktop (please complete the following information):

• OS: Windows 10
• Version : Current master version

Additional context
Have tried to set the m_poolSize to a large number (i.e. 65535), does solve the problem ("MaxCheck" set to 16394), but the problem would still occur if using a much larger "MaxCheck". Since there is at least the memory space issue, and search efficiency, should not simply set the m_poolSize to a very large number.
Maybe set the m_poolSize according to "MaxCheckForRefineGraph" during build stage, and according to "MaxCheck" during search stage.

From GettingStarted.md I figured that we can use L2 distance by:
SPTAG.AnnIndex.SetBuildParam("DistCalcMethod", "L2")
What is the value for cosine similarity distance?

Describe the bug
Bad encoding in output of search

To Reproduce
Steps to reproduce the behavior:

1. Build an index with algorithm BKT using the cli
2. Build a search using a cli
3. Open output file of search
4. Contains bad encoding part way through the file

Expected behavior
Consistent encoding throughout the results

Output
Melanoma:[email protected]_Style_Transfer_for_Videos.pdf|1.005@^_O�:�^B�>�|1807.00273v1.Photorealistic_Style_Transfer_for_Videos.pdf��?�<:i��v3Y�$��<�n)8^\��=�J^R<J^?�<@hw���K;LV^A�FO~U;�^?

Desktop (please complete the following information):

• OS:Ubuntu 18.04

Is your feature request related to a problem? Please describe.
I want to use SPTAG to do kNN queries on vectors from a TF-IDF vectorizer, usually sparse vectors. Can I use sparse vectors in SPTAG or rather do you have an idea of how it might perform with sparsity?

DescriptionI get an error when compiling. The error seems related to copying of files from the Wrappers/inc folder to the Release folder. When I compile I get the following error.

Error copying file "/home/junior/programming/builds/SPTAG/Wrappers/inc/SPTAGClient.py" to "/home/junior/programming/builds/SPTAG/Release/". make[2]: *** [Wrappers/CMakeFiles/_SPTAGClient.dir/build.make:246: ../Release/_SPTAGClient.so] Error 1 make[2]: *** Deleting file '../Release/_SPTAGClient.so' make[1]: *** [CMakeFiles/Makefile2:386: Wrappers/CMakeFiles/_SPTAGClient.dir/all] Error 2 make: *** [Makefile:130: all] Error 2

To Reproduce
Steps to reproduce the behavior:

1. git clone
2. cd SPTAG
3. mkdir build
4. cd ubild && cmake .. && make

Desktop (please complete the following information):

• OS: Fedora 30
• Python 2.7

To benchmark against other python ann search methods, would be nice to have some pip installable package that would make it easier to set-up and run these search methods.

Excellent package. Thanks!

When there are too many vectors, such as 5 million （n = 1024×1024×5）， index build will fail to complete.
The program stops at this place, prompting:
"Save Data To xxxxx\vectors.bin"
And I noticed that the file size of vectors.bin reach to 300G! that is not normal，the The correct file size should be 10G.

The code that caused the error is as follows：

import SPTAG
import numpy as np
n = 1024*1024*5 #this szie will cause the file size of vectors.bin to reach 300G!
k = 3
r = 3
Dimension=512 #the size of vectors.bin will be 1024×1024×512×4=2G  n×Dimension×4
def testBuild(algo, distmethod, x, out):
   i = SPTAG.AnnIndex(algo, 'Float', x.shape[1])
   i.SetBuildParam("NumberOfThreads", '4')
   i.SetBuildParam("DistCalcMethod", distmethod)
   ret = i.Build(x.tobytes(), x.shape[0])
   i.Save(out)
def Test(algo, distmethod):
   x = np.ones((n, Dimension), dtype=np.float32) * np.reshape(np.arange(n, dtype=np.float32), (n, 1))
   q = np.ones((r, Dimension), dtype=np.float32) * np.reshape(np.arange(r, dtype=np.float32), (r, 1)) * 2
   print ("Build.............................")
   testBuild(algo, distmethod, x, 'testindices')

if __name__ == '__main__':
   Test('BKT', 'L2')

How can I solve this error?
Please Help Me！

I was trying to use the C# wrapper and faced some issues. Here are the things I tried:

• Built the CSHARPSPTAG.dll compiling the CSHARPSPTAG project in VS17. But I was unable to add reference to this dll directly to a C# project. It said

A reference to 'path\to\CSHARPSPTAG.dll' could not be added. Please make sure that the file is accessible, and that it is a valid assembly or COM component.

• Following this and this, I created a new C# DLL project with the swig generated *.cs files to make a .NET dll that I can add reference to. I then added SPTAGLib.dll and CSHARPSPTAG.dll in the \bin\Release\ dir of my project. This time, I was able to get it compiled, but when I ran it, it threw an exception

System.TypeInitializationException was unhandled
Message: An unhandled exception of type 'System.TypeInitializationException' occurred in ConsoleTest.exe
Additional information: The type initializer for 'CSHARPSPTAGPINVOKE' threw an exception.

• I also tried adding all the TBB related dlls into the bin of my C# project but none of it seems to work.

Can any of you share how you managed to make it work?

When i use the np.float64, I turned the SPTAG.Annindex() datatype parameter from 'Float' (used in the test demo) to 'Double', 'Float64','float64', 'double'， but all failed.

When we use delete function, the model reports OverflowError. The data type is already float32.

Describe the bug
OverflowError: in method 'AnnIndex_AddWithMetaData', argument 4 of type 'int'
when running the Python test script from "Get Started"

To Reproduce
Create a new Python file in the Release directory, copy&paste the Python test script from "Get Started".

Additional context
My environment:
Ubuntu 18.04, Kernel 4.15.0 x86_64
Tried swig 3.0 & 4.0
cmake 3.14.4
gcc 7.4.0
python 3.6

A well-documented drawback of LSH, in general, is the 'false-negative' issue, whereby you do not know truly whether all nearest neighbors were actually considered at query time.

Approaches to solving this in other libraries have made good headway - how does SPTAG handle this specific issue?

Would you like to wrap any pointer data members with the class template “std::unique_ptr”?

I used SPTAG in face searching. There're more than a million faces in my database.

1. When I change k value from 28000 to 30000, the searching time raised from 0.4s to several minutes.

2. I got many duplicate indices from j.Search() when I set k to a value bigger than 10000.

Looking forward to your reply.

My environment:

Python 3.7
Ubuntu 18.04

Is your feature request related to a problem? Please describe.
A clear and concise description of what the problem is. Ex. I'm always frustrated when [...]

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

Describe the bug
When I test SPTAG in python by your python test code. I find when I delete some data from Index, SaveIndex always get segmentation fault.

This is Delete and Save code:

def testDelete(index, x, out):
   i = SPTAG.AnnIndex.Load(index)
   ret = i.Delete(x.tobytes(), x.shape[0])
   print (ret)
   print("Del finished...................")
   i.Save(out)

After save tree.bin finished, segmentation fault.
Thanks for your favor.

Describe the bug

Test SPTag wrapper on python 2.7, ubuntu 18.04
TRY x is FLOAT data. Got error Segmentation fault (core dumped)

import SPTAG
import numpy as np

n = 100
k = 3
r = 3

def testBuild(algo, distmethod, x, out):
    i = SPTAG.AnnIndex(algo, 'Float', x.shape[1])
    i.SetBuildParam("NumberOfThreads", '4')
    i.SetBuildParam("DistCalcMethod", distmethod)
    ret = i.Build(x.tobytes(), x.shape[0])
    i.Save(out)

def testSearch(index, q, k):
    j = SPTAG.AnnIndex.Load(index)
    for t in range(q.shape[0]):
        result = j.Search(q[t].tobytes(), k)
        print(result[0])  # ids
        print(result[1])  # distances

def Test(algo, distmethod):
    x = np.ones((n, 10), dtype=np.float32) * np.reshape(np.random.uniform(low=0.5, high=13.3, size=(n,)), (n,1))

    print('x.shape: ', str(x.shape))
    print(x[1,])

    print("Build.............................")
    testBuild(algo, distmethod, x, 'testindices')
    testSearch('testindices', q, k)

if __name__ == '__main__':
    Test('BKT', 'L2')
    # Test('KDT', 'L2')

Log console:

('x.shape: ', '(100, 10)')
[ 0.60697413 0.60697413 0.60697413 0.60697413 0.60697413 0.60697413
0.60697413 0.60697413 0.60697413 0.60697413]
Build.............................
Setting NumberOfThreads with value 4
Setting DistCalcMethod with value L2
Save Data To testindices/vectors.bin
Save Data (0, 1) Finish!
Save BKT to testindices/tree.bin
Segmentation fault (core dumped)

Most ANN libraries do not have a clear method for deployment. Docker and Kubernetes are widely used.

I would appreciate it if you added some documentation on how to deploy SPTAG onto a Kubernetes cluster.

Especially, will the performance compromise if the added points are imbalanced?

Hey, thanks for this great project.
I wanna use this algorithm in face feature search, I tried another algorithm called hnswlib, this is also fast, but static, cannot delete.
So I compare three methods: SPTAG.AnnIndex("BKT"), hnswlib and brutoforce search (np.argmax).
Following table is each method's cost time. All are querying 1000 features' 3 most similar features in 26458 fetures.

As the table shows, SPTAG.AnnIndex("BKT") is not faster than brutoforce search, don't get spped up.
This is my SPTAG.AnnIndex("BKT") test code:

@calc_time
def testSearch(index, q, k):
   j = SPTAG.AnnIndex.Load(index)
   _t0 = datetime.now()
   for t in range(q.shape[0]):
       result = j.Search(q[t].tobytes(), k)
   _t1 = datetime.now()
   print("Search time is {} ms".format(1000*(_t1-_t0).total_seconds()))

I don't know if this is normal, or my test code is wrong?
Thank you.

Below is the code that will cause the error：

import SPTAG
import numpy as np
n = 1024*1024 #this szie will encounted a fatal error
k = 3
r = 3
Dimension=512 #the size of vectors.bin will be 1024×1024×512×4=2G  n×Dimension×4
def testBuild(algo, distmethod, x, out):
   i = SPTAG.AnnIndex(algo, 'Float', x.shape[1])
   i.SetBuildParam("NumberOfThreads", '4')
   i.SetBuildParam("DistCalcMethod", distmethod)
   ret = i.Build(x.tobytes(), x.shape[0])
   i.Save(out)
def Test(algo, distmethod):
   x = np.ones((n, Dimension), dtype=np.float32) * np.reshape(np.arange(n, dtype=np.float32), (n, 1))
   q = np.ones((r, Dimension), dtype=np.float32) * np.reshape(np.arange(r, dtype=np.float32), (r, 1)) * 2
   print ("Build.............................")
   testBuild(algo, distmethod, x, 'testindices')
   testSearch('testindices', q, k)

if __name__ == '__main__':
   Test('BKT', 'L2')

And here is the code that can run successfully:

import SPTAG
import numpy as np
n = 1024*1024 -1 #Note the difference here, 1 less than the parameters of the above code.
k = 3
r = 3
Dimension=512 #the size of vectors.bin will be 1024×1024×512×4=2G  n×Dimension×4
def testBuild(algo, distmethod, x, out):
   i = SPTAG.AnnIndex(algo, 'Float', x.shape[1])
   i.SetBuildParam("NumberOfThreads", '4')
   i.SetBuildParam("DistCalcMethod", distmethod)
   ret = i.Build(x.tobytes(), x.shape[0])
   i.Save(out)
def Test(algo, distmethod):
   x = np.ones((n, Dimension), dtype=np.float32) * np.reshape(np.arange(n, dtype=np.float32), (n, 1))
   q = np.ones((r, Dimension), dtype=np.float32) * np.reshape(np.arange(r, dtype=np.float32), (r, 1)) * 2
   print ("Build.............................")
   testBuild(algo, distmethod, x, 'testindices')
   testSearch('testindices', q, k)

if __name__ == '__main__':
   Test('BKT', 'L2')

How can I solve this error?
Please Help Me！

Describe the solution you'd like
I'm using more and more ML.net framework so my primary language is C#. I hope that a C# wrapper will be produced so it will be very useful to encompass it in my business applications.

I would like to point out that an identifier like “_SPTAG_CORE_COMMONDEFS_H_” does eventually not fit to the expected naming convention of the C++ language standard.
Would you like to adjust your selection for unique names?

In this sentence:
"SPTAG provides two methods: kd-tree and relative neighborhood graph (SPTAG-KDT) and balanced k-means tree and relatrive neighborhood graph (SPTAG-BKT)," "relatrive" should be "relative."

Thanks for open-sourcing!

Describe the bug
Ubuntu 18.04 installed SPTAG and trying to run the sample code here.
Also added this snippet based on a different issue:

import sys
sys.path.append('./Release')
import SPTAG
import numpy as np

n = 100
k = 3
r = 3
: 
:

Additional context
I get this error message:

  File "test_ann.py", line 3, in <module>
    import SPTAG
  File "./Release/SPTAG.py", line 15, in <module>
    import _SPTAG
ImportError: dynamic module does not define module export function (PyInit__SPTAG)

what am I doing wrong?

I have been trying to figure SPTAG out to use in my ANN searches to no avail. I am not getting an responses from the contributors, so am trying my luck with other uses. Could someone provide a simple walkthrough of how to use this tool? This is what I am stuck on:

1. If i want to create a 300 dimension store, I need to a "build" first? and is this done only once?

n = 300
k = 5
r = 3
def testBuild(algo, distmethod, x, out):
   i = SPTAG.AnnIndex(algo, 'Float', x.shape[1])
   i.SetBuildParam("NumberOfThreads", '4')
   i.SetBuildParam("DistCalcMethod", distmethod)
   ret = i.Build(x, x.shape[0])
   i.Save(out)

2. For incremental adds, do I just use this method without overwriting previous ones:

def testAdd(index, x, out, algo, distmethod):
   if index != None:
       i = SPTAG.AnnIndex.Load(index)
   else:
       i = SPTAG.AnnIndex(algo, 'Float', x.shape[1])
   i.SetBuildParam("NumberOfThreads", '4')
   i.SetBuildParam("DistCalcMethod", distmethod)
   if i.Add(x, x.shape[0]):
       i.Save(out)

3. what does the testAddWithMetaData do? does it return say an identifier with each result it returns? Is there a type of metadata that is supported i.e. can i save text snippets with it?

Could someone from microsoft please take a moment to provide a simple guide to using this powerful tool?

In python demo code , I added a line in testBuild function:
i.SetBuildParam("RefineIterations","3")
But it seems useless , when the index building , the RefineIterations still is 2.

thanks for this great project, I like this project so much!
When I run cd build && cmake .. && make, get an error at 86%.

As the error messge said: this is caused by the SPTAG.py missed. And I found SPTAG.py is in .gitignore
I dont know how to solve this, do you have any ideas?
thank you so much.

ModuleNotFoundError: No module named 'SPTAG'

-a, --algo Index Algorithm type, required.
such as?

Describe the bug
A clear and concise description of what the bug is.

To Reproduce
Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

Expected behavior
A clear and concise description of what you expected to happen.

Screenshots
If applicable, add screenshots to help explain your problem.

Desktop (please complete the following information):

• OS: [e.g. iOS]
• Browser [e.g. chrome, safari]
• Version [e.g. 22]

Smartphone (please complete the following information):

• Device: [e.g. iPhone6]
• OS: [e.g. iOS8.1]
• Browser [e.g. stock browser, safari]
• Version [e.g. 22]

Additional context
Add any other context about the problem here.

Inside the KmeansAssign function during the BKT building stage, it seems like it tries to parallel the job amount the threads (if set the NumberOfThreads to more than 1)

However, as the omp_get_num_threads() would only return 1 outside the "omp parallel for" loop, the function actually never parallel the job.
I'm not sure if this is done on purpose (since the BKT build wouldn't take much time, so no need to parallel?) or just a small bug.
I've also tried to use omp_get_max_thread() instead of omp_get_num_threads() to make use of the parallel. However, there is some kind of a bug that would just end the build process without any error message.

I am still trying to figure out SPTAG, but unclear on what the metadata methods provide.
for example in the Test() method in sample:

the metadata is generated thus:

for i in range(n):
       m += str(i) + '\n'
   m = m.encode()

does mean each line is associated with each row in the vectors?
can someone provide a simple example of storing metadata with the vectors?

can you give me demos of QueryFile and TruthFile?
[email protected]
I want the files.

When I run the official test python code, I get an error：
Traceback (most recent call last):
File "testSPTAG.py", line 4, in
import SPTAG
File "SPTAG/Release/SPTAG.py", line 17, in
_SPTAG = swig_import_helper()
File "SPTAG/Release/SPTAG.py", line 16, in swig_import_helper
return importlib.import_module('_SPTAG')
File "/usr/lib/python2.7/importlib/ __ init __.py", line 37, in import_module
__import __(name)
ImportError: dynamic module does not define init function (init_SPTAG)

The SPTAG module caused Python to stop working.
Fault Description: I am using SPTAG source code compiled with python3.6 64-bit VS2017, test.exe test passed.
Under Windows 7, Python 3.6 can import the SPTAG module. I have successfully indexed 1 million data, but when my data volume exceeds 1 million, an error occurs. The specific tips are:
PYTHON has stopped working,
Problem Event Name: BEX64
Application name: python.exe
Application version: 3.6.150.1013
Fault module name: ucrtbase.DLL
Exception code: C0000417
I went to the Internet to find a solution. Some people said that it is the reason for the VC runtime library, but I still installed the VC2015 and VC2017 runtime libraries.

Also, I noticed that when the program fails, the word "Start to build BKTree" does not appear after ""Setting DistCalcMethod", but it appears directly:
"Save Data To xxxxx\vectors.bin"
Save Data(0,1) Finish!
Save BKT to xxxxx\tree.bin
An error window will pop up here, prompting "Python has stopped working"

Describe the bug
I'm not expert in C++ compilation but seems very hard to get the solution built.

My questions are :

• Did someone test the build for VS 2019 ? is it working ?
• How can I build the solution for C# wrapper by using VS 2019 instead of cmake
• Seems that boost 1.70 is not working ? am I wrong ?

Desktop (please complete the following information):

• OS: windows 10
• Visual Studio version : VS 2019 16.1.6

Thanks in advance for your feedback

Describe the bug
When I install SPTAG with boost 1.70.0 version cause cmake below error message on Ubuntu machine.

/usr/local/include/boost/asio/detail/io_object_impl.hpp:88:53: error: ‘class boost::asio::execution_context’ has no member named ‘get_executor’

Removed 1.70.0 and Installed boost 1.67.0, solved the build issue.

Here is the detail error message.

/usr/local/include/boost/asio/io_context_strand.hpp:89:19: note:   no known conversion for argument 1 from ‘boost::asio::execution_context’ to ‘const boost::asio::io_context::strand&’

In file included from /usr/local/include/boost/asio/basic_socket.hpp:21:0,
                 from /usr/local/include/boost/asio/basic_socket_acceptor.hpp:19,
                 from /usr/local/include/boost/asio/ip/tcp.hpp:19,
                 from /home/USERNAME/project/SPTAG/AnnService/inc/Socket/Connection.h:13,
                 from /home/USERNAME/project/SPTAG/AnnService/src/Socket/Connection.cpp:4:

/usr/local/include/boost/asio/detail/io_object_impl.hpp: In instantiation of ‘boost::asio::detail::io_object_impl<IoObjectService, Executor>::io_object_impl(ExecutionContext&, typename std::enable_if<std::is_convertible<ExecutionContext&, boost::asio::execution_context&>::value>::type*) [with ExecutionContext = boost::asio::execution_context; IoObjectService = boost::asio::detail::deadline_timer_service<boost::asio::time_traits<boost::posix_time::ptime> >; Executor = boost::asio::executor; typename std::enable_if<std::is_convertible<ExecutionContext&, boost::asio::execution_context&>::value>::type = void]’:

/usr/local/include/boost/asio/basic_deadline_timer.hpp:174:20:   required from ‘boost::asio::basic_deadline_timer<Time, TimeTraits, Executor>::basic_deadline_timer(ExecutionContext&, typename std::enable_if<std::is_convertible<ExecutionContext&, boost::asio::execution_context&>::value>::type*) [with ExecutionContext = boost::asio::execution_context; Time = boost::posix_time::ptime; TimeTraits = boost::asio::time_traits<boost::posix_time::ptime>; Executor = boost::asio::executor; typename std::enable_if<std::is_convertible<ExecutionContext&, boost::asio::execution_context&>::value>::type = void]’

/home/USERNAME/project/SPTAG/AnnService/src/Socket/Connection.cpp:29:31:   required from here

/usr/local/include/boost/asio/detail/io_object_impl.hpp:88:53: error: ‘class boost::asio::execution_context’ has no member named ‘get_executor’
         is_same<ExecutionContext, io_context>::value)
                                                     ^

AnnService/CMakeFiles/server.dir/build.make:348: recipe for target 'AnnService/CMakeFiles/server.dir/src/Socket/Connection.cpp.o' failed
make[2]: *** [AnnService/CMakeFiles/server.dir/src/Socket/Connection.cpp.o] Error 1
CMakeFiles/Makefile2:176: recipe for target 'AnnService/CMakeFiles/server.dir/all' failed
make[1]: *** [AnnService/CMakeFiles/server.dir/all] Error 2
Makefile:129: recipe for target 'all' failed
make: *** [all] Error 2

To Reproduce
Steps to reproduce the behavior:
Install boost 1.70.0 and run install instruction
on https://github.com/microsoft/SPTAG

mkdir build
cd build && cmake .. && make

Test OS version information:
Microsoft Azure default Ubuntu 14 distribution
OS: Linux dw-ubuntu14 4.15.0-1045-azure #49-Ubuntu SMP Mon May 13 16:30:09 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

Boost install steps I followed
https://stackoverflow.com/questions/12578499/how-to-install-boost-on-ubuntu/24086375#24086375

Removed 1.70.0 and Installed boost 1.67.0, solved the build issue.

I tested it with a GIST data set，The size of the query vector is (1000*960)，Return 100 neighbors，The time I built the index is 20562.084 s，The query time is 35.082 s，recall is 0.92669，The time to add 100 vectors is 22.281 s. I feel this time is very long, am I using it correctly? Its index construction and query time are longer than NSG.

Can anyone help me? Thanks

Describe the bug
I haven't been able to figure out how to run SPTAG in a server/client mode on Ubunt 18.04. I want to host SPTAG on a cloud server and run a python client to perform kNN lookups.

For example, the documentations states thus:

./Server [options]
Options: 
  -m, --mode <value>              Service mode, interactive or socket.
  -c, --config <value>            Configure file of the index
:  

But I can't seem to find the Server binary. I see a lowercase server binary in the Release folder, but it doesn't accept these arguments and gives an error:

Command 'server' not found, but can be installed with:
apt install rsplib-legacy-wrapper

Can someone post a simpler guide to using SPTAG in client/server mode for us noobs?
Thanks for such a wonderful tool!

I suppose there are about 19 parameters of BKT index in the code, which may influence the performance/accuracy tradeoff seriously. Can someone post a simpler guide to demonstrate how to tune these parameters?

Hi just wondering if there are plans to speed up computation via GPUs?

I have built the project successfully，but when I import SPTAG, I met this error:
`

import sys
sys.path.append(r"D:\library\github_repositories\SPTAG-2\build\Release")
import SPTAG
Traceback (most recent call last):
File "", line 1, in
File "D:\library\github_repositories\SPTAG-2\build\Release\SPTAG.py", line 15, in
import _SPTAG
ImportError: DLL load failed: 找不到指定的模块。`

The files in folder build/Release are:

_SPTAG.exp
_SPTAG.lib
_SPTAG.pyd
_SPTAGClient.exp
_SPTAGClient.lib
_SPTAGClient.pyd
aggregator.exe
client.exe
indexbuilder.exe
search.exe
server.exe
SPTAG.py
SPTAGClient.py
SPTAGLib.dll
SPTAGLib.lib
test.exe

my environments are :

vs2015
python3.6
windows7

Could anyone tell me the solution for this error?

SPTAG has around 10 hyperparameters. We try to use automl algorithms to find best-performing hyper-parameters for SPTAG. Neural Network Intelligence toolkit (NNI) is a perfect tool for this task. It is a toolkit to help users design and tune machine learning models (e.g., hyperparameters), neural network architectures, or complex system’s parameters, in an efficient and automatic way. NNI has many tuning algorithms built in, and can be executed in a distributed way on a local machine, a remote server, or a large scale training platform such as OpenPAI or Kubernetes.