Not all the files can be unzipped correctly.
Check that the size of the original zip file is correct (about 1 GB).
Some unzip applications cannot handle large zip files containing a large number of files. If problems occur, try a different unzip application.
“Access denied” is displayed, and the program will not start.
The folder in which Neural Network Console was unzipped may not be writable from Neural Network Console. This occurs such as when the unzipped Neural Network Console is copied into the Program Files folder (folders under the Program Files folder cannot be written to from applications). Install Neural Network Console in a folder that can be written to freely (immediately under drive C, on the desktop, in the document folder, etc.).
“The ordinal 4540 could not be located in the dynamic link library LIBEAY32.dll” is displayed when the program is started.
The installed OpenSSL may be of an older version. Install the latest OpenSSL.
“DLL load failed: The specified module could not be found” is displayed, and training cannot be executed.
Visual Studio 2015 Visual C++ Redistribution Package may not be installed in your PC. Download it from the following page, and install it.
https://www.microsoft.com/en-us/download/details.aspx?id=52685
“CUDA driver version is insufficient for CUDA runtime version” is displayed, and training cannot be executed.
The graphic driver installed in your PC may not be the latest version. Down load the latest graphic driver from the NVIDIA page, and install it.
An error occurs when training is executed with a sample project.
If Neural Network Console is installed in an environment that already has Python, or the like installed, training may not be performed correctly. Temporarily uninstall Python, or the like, or remove the PATH to them in the environment variables to isolate the problem.
Reference
Neural Network Console calls Python contained in the zip file to run Python code. The Python used in this instance is specified dynamically through a PATH environment variable specified on the setup window on the Neural Network Console GUI, but when Python is available in a folder specified by the OS PATH, it may be used with higher priority.
I cannot perform training using a GPU.
Check that the GPU board driver is the latest version.
Neural Network Console currently only supports NVIDIA (CUDA compatible) GPUs. GPUs in AMD and Intel chipsets are not supported.
Training is slow. Is there a way to know whether the GPU is working properly?
Tools for monitoring the GPU usage can be used.
If the GPU load is 0% most of the time, it may be that the GPU is not being used properly. Check the settings on the setup window and whether the GPU is supported by the layers that you are using.
If the GPU load is not maintained at 50% or higher, loading training data from the disk may be falling behind, the batch size may be too small, a network structure with small array size may be in use, many layers not suitable for parallel computing are in use, the data size may be too small for effective parallel computing to begin with (many of the included sample projects fall under this description), and so on.
Further, training speed has been found to slow down due to the GPU’s used memory size reaching 100% depending on the environment. For this phenomenon, it is reported that the problem can be resolved by updating the display driver to the latest version.
The GPU environment cannot be set up or training using a GPU does not work on a remote desktop environment.
By design, Windows may disable the GPU when remote desktop connection is in use depending on the environment. Perform the operation on a real machine, or try a non Windows standard remote desktop that does not disable the GPU.
