colorfulsoft / system.ai

Machine Learning and Data Analysis stack for .NET ecosystem

What is System.AI

System.AI is a stack of machine learning and data analysis technologies for the .NET ecosystem.

Supported platforms

System.AI is supported by any software and hardware platforms that support the Microsoft .NET Framework or its implementations, such as Microsoft .NET Core, Mono, or Xamarin. The core of the System.AI is written in pure C# and is exclusively managed code.

Getting started

Detailed instructions for launching on different platforms: https://github.com/ColorfulSoft/System.AI/blob/master/Docs/GettingStarted.md

Libraries included in the System.AI and architecture of the project

Library	Root .NET namespace	Status	Version	Docs
System.FloatingPoint	System	stable	1.1	https://github.com/ColorfulSoft/System.AI/blob/master/Docs/System.FloatingPoint.md
imageio.NET	System.IO	stable	1.2	https://github.com/ColorfulSoft/System.AI/blob/master/Docs/imageio.NET.md
warnings.NET	System	stable	1.0	https://github.com/ColorfulSoft/System.AI/blob/master/Docs/warnings.NET.md
System.Complex	System	stable	1.0	https://github.com/ColorfulSoft/System.AI/blob/master/Docs/System.Complex.md
PyType.NET	System	stable	1.0	https://github.com/ColorfulSoft/System.AI/blob/master/Docs/PyType.NET.md
torchfile.NET	System.IO	stable	1.0	https://github.com/ColorfulSoft/System.AI/blob/master/Docs/torchfile.NET.md
micrograd.NET	System.AI	stable	1.0	https://github.com/ColorfulSoft/System.AI/blob/master/Docs/micrograd.NET.md
DotnetExtensions	System	stable	1.0	https://github.com/ColorfulSoft/System.AI/blob/master/Docs/DotnetExtensions.md

Warning!

Note that the libraries of System.AI basically repeats the interface and behavior of the corresponding analogues from the Python ecosystem, but they do not do it 100%. This is largely due to the difference in topologies .NET and Python. For example, the torch.jit module that exists in PyTorch probably won't be implemented in Torch.NET, since it will not make sense in the case of working in the ecosystem .NET (programs with Torch.NET by themselves are portable and there is no point in making them even more portable with torchscript).

System.AI is a set of libraries, each of which can be used independently of the others as a component of your product. By default, System.AI is compiled into a set of dynamic libraries, however, it is also possible to compile into a single System.AI.dll library

The status of libraries in the System.AI

• ready - Development is complete. The library is ready to use. Expansion of the functionality is not planned(but not excluded).
• stable - Stable version, ready to use.
• dev - Development is underway, and the code will be added soon.
• prev - The code is posted for informational purposes and is currently not ready for use in production.

Update history

• [22.11.2022] DotnetExtensions library is now available after reconstruction.
• [31.10.2022] micrograd.NET library is now available.
• [31.10.2022] V2.0 is now the main branch.
• [27.10.2022] torchfile.NET library is now available.
• [20.07.2022] Support for VisualStudio/SharpDevelop/MonoDevelop via handwritten *.sln and *.csproj files. Tested using SharpDevelop 5.1.
• [20.04.2022] Active work is underway on the torch.NET and torchvision.NET modules. Some codes are placed in the repository for demonstration purposes.
• [14.04.2022] Added documentation for warnings.NET.
• [11.04.2022] The current documentation for the System.FloatingPoint and imageio.NET libraries is posted.
• [10.04.2022] A new version of the PyType.NET library is available.
• [10.04.2022] A new version of imageio.NET has been released.
• [10.04.2022] The System.Complex module is implemented, which contains CQuarter, CHalf, CBFloat16, CFloat and CDouble types of complex numbers. Added missing type conversions to the System.FloatingPoint module.
• [05.03.2022] Implemented a warnings.NET warning module, an analogue of the standard warnings module of the Python ecosystem.
• [05.03.2022] The generation of the debug version has been added, the imageio.NET library has been optimized.
• [03.03.2022] An updated version of imageio.NET is now available in System.AI 2.0
• [13.12.2021] Version 1.1 of the System.FloatingPoint component has been released. The BFloat16 data type is implemented.
• [12.12.2021] The restructuring of the project has been launched.
• [06.02.2021] The code of torch.NET and MKL is placed for demonstration purposes in the form in which they are at this stage of development. Added the PyType.NET library.
• [19.01.2021] The new version of imageIO.NET library is ready to use! In the new version: data format has been changed to HWC, as in imageio for Python, performance of imageio.imread and imageio.imwrite has been improved.
• [27.10.2020] The new version of imageIO.NET library is ready to use! Arrays of all standard numeric types are now supported for images. Also, imageIO.NET now supports reading and writing images from / to the stream.

system.ai's Issues

Implement reduction ops

Musthave

Implement broadcasting operations

Implement +, -, *, /, div, mod and other broadcasting operations.

Add the printing of five-dimensional tensors.

Implement cross-platform version of ImageIO.NET

Implement multithreading for nn.Functional.max_pool2d.

Implement Conv2d layer

Implement function torch.nn.Functional.conv2d and class torch.nn.Conv2d

Implement tensor initialization from arrays of all numeric types .NET.

Transfer parallelism from 0 dimension to 1 dimension in layers working with BCHW tensors.

Parallelism at the batch level does not give profit in many tasks.

Implement imageio.NET package and torchvision.transforms class.

macOS compilation and execute instructions

Musthave