Installation and testing

This page guides you through installing the Tensor library and creating a skeleton project for experimentation.

System requirements

The following system requirements must be met.

• System architecture: x86-64 (AMD64 or Intel 64)
• Operating system: Linux, MacOS or Microsoft Windows
• Microsoft .NET Standard 2.0 implementation
  • Recommended platform is .NET Core >= 2.0
  • .NET Framework >= 4.7 is supported
  • Mono >= 5.10 is supported, but significantly slower
• For Linux
  • The library libgomp.so.1 must be installed. (install on Ubuntu by running sudo apt install libgomp1)
• For GPU acceleration (optional)
  • nVidia GPU supporting CUDA compute capability 3.5 or higher
  • nVidia GPU driver 387.92 or higher

Installation

The library is delivered in two NuGet packages. The Tensor NuGet package provides the Tensor<'T> type and all core functions. Additional algorithms and data exchange methods are provided in the Tensor.Algorithm NuGet package.

The packages can be installed into your project by installing the Tensor and Tensor.Algorithm packages using the NuGet package manager (either via command line or graphical interface).

Skeleton project for .NET Core

In the course of this tutorial you will use the following skeleton project for experimentation. We assume that you are using .NET Core 2.0 on either Linux or Windows for the rest of the tutorial.

To create the skeleton project run the following commands.

$ mkdir tutorial
$ cd tutorial
$ dotnet new console -lang F#

Then, run the following commands to install the Tensor library into your project.

$ dotnet add package Tensor
$ dotnet add package Tensor.Algorithm

Basic verificiation test

To verify that the installation was successful you can perform a basic test of the library. Place the following code into Program.fs.

open Tensor
[<EntryPoint>]
let main argv =
    let x = HostTensor.counting 6L
    printfn "x = %A" x
    0

If everything works fine, dotnet run automatically builds your project and produces the following output.

$ dotnet run
x = [   0    1    2    3    4    5]

GPU acceleration verification test

By changing HostTensor to CudaTensor inside Program.fs and executing dotnet run, you can test if GPU acceleration works properly.

Source code and issues

The source code of the Tensor library is available at https://github.com/DeepMLNet/DeepNet.

You can also directly reference the Tensor.fsproj and Tensor.Algorithm.fsproj projects inside the source tree from your project by using dotnet add reference <path>. This is useful if you want to modify the Tensor library itself or for debugging.

Please report issues via https://github.com/DeepMLNet/DeepNet/issues and submit your pull requests via https://github.com/DeepMLNet/DeepNet/pulls.