CNTK Library Evaluation on Linux

The CNTK Library on Linux is provided as C++ and Python library.

IMPORTANT! Before reading further you may want to get familiar with the naming scheme of CNTK Shared Library files.

Using C++

The usage pattern on Linux is the same as that on Windows.

The evaluation library, libCntk.Core, can be found under cntk/lib in the CNTK binary package. If you build CNTK from source code, the libCntk.Core is available in the lib folder of the build directory.

Any program using the evaluation library needs to link libCntk.Core and libCntk.Math when compiling the program, for example by specifying

-lCntk.Core-2.0rc1 -lCntk.Math-2.0rc1

as well as appropriate search path for both libraries. (Ensure you use the correct file names! See the beginning of this article). Please use the same compiler version as that used to build libraries. The Examples/Evaluation/CNTKLibraryCPPEvalCPUOnlyExamples and Examples/Evaluation/CNTKLibraryCPPEvalGPUExamples in the CNTK source code illustrates the usage pattern in Linux. The Makefile contains the target CNTKLIBRARY_CPP_EVAL_EXAMPLES showing how to build the example.

For details on the CNTK Library for evaluation, please refer to the CNTK Library Evaluation Interface page in this wiki.

Using Python

You can use Python to evaluate a pre-trained model. Examples can be found here.