Samples for Gradient, TensorFlow binding for .NET

Billion Songs a separate repository with deep learning-powered song lyrics generator in an ASP.NET Core web site.

See also: Writing billion songs with C# and Deep Learning

BasicMath- creates two constant tensors and performs simple algebraic operations on them

CharRNN - generates semi-sensical text in the style of input. For example (Shakespeare):

LOST: What is your name beautiful princess?

COMPUTER: Gradient your horn of exemity.

GLOUCESTER: how one sick up scourish, thither place and thronds the grow for these bowing.

MARIANA: He shall speak not reverbering injurance.

CSharpOrNot - a mini-ResNet convolutional network, that guesses programming language, given a rectangular text block from a code file. Has a cross-platform UI demo. Get pretrained model here: https://github.com/losttech/Gradient-Samples/releases/tag/csharp-or-not%2Fv1

GPT-2 - latest English language model from OpenAI (smaller version, pretrained). Added fine-tuning from https://github.com/nshepperd/gpt-2.

FashionMnistClassification - standard TensorFlow example, that classifies small pictures of clothes.

LinearSVM - incomplete port due to a bug with tf.flags

ResNetBlock - same as FashionMnistClassification above, but shows Model subclassing to implement ResNet block.

RL-MLAgents - reinforcement learning agent, that learns to play Unity 3D based games using Soft Actor-Critic algorithm, and Unity ML agents library. More details in the blog post.

SimpleApproximation - uses a simple 1 hidden layer neural network to approximate an arbitrary function.

All models can be modified and trained.

LICENSE - MIT, individual samples might have different licenses (clearing that up, see individual sample folders).

SIREN - neural representation for any kind of signal (image, video, audio).