We just released LostTech.TensorFlow 2.4 Preview 1 on NuGet. This is the first preview of our TensorFlow binding for .NET, that targets TensorFlow 2.x.

Take a look at TensorFlow official documentation to see what’s new in TensorFlow 2.x.

We also updated Samples to work with the Preview.

TensorFlow 2 by default works in eager execution mode. Most samples, that use TF 1.x style require legacy Session-based mode instead. It can be enabled at the start of the program by calling v1.disable_eager_exection(). Legacy TF 1.x APIs are available in v1 class under tensorflow.compat.v1 namespace.

Today LostTech.TensorFlow, our C# binding to TensorFlow, gets its first Release Candidate available to the public. It brings a stable API, that mirrors TensorFlow 1.15, and includes tensors, graphs, and sessions, along with Keras, an optional eager execution mode, and much more.

Check out the product webpage for the overview of available features. See our samples or just install the NuGet package and start your journey into deep learning with .NET!

This Release Candidate comes with a go-live license, meaning unlike the previews, it has no expiration date and is fully supported.

LostTech.TensorFlow is completely free for individuals and small businesses, while larger enterprises can get a pay-as-you-go license through their Azure Subscription, which includes some trial credits. Please, contact Sales if you need the license for disconnected use cases.

A couple of years ago Unity started working on a framework, that would enable training machine learning algorithms in virtual environments: ML-Agents Toolkit. It is used by a number of companies across the world, including famous DeepMind, to aid research in computer vision and robotics in such tasks as autonomous driving.

In this post we present an example bot built with C# and TensorFlow framework, that learns to play a game in a simple Unity-based virtual environment using one of the state of the art reinforcement learning algorithms: soft actor-critic.

As you might have heard, today Apple has acquired our Seattle neighbor Xnor.ai for $200M. The company’s main product is a mechanism to run neural networks on low-power devices, and its core is just 50 lines of code. It achieves then efficiency by performing operations en masse on individual bits instead of the normal 32- and recently 16-bit floating point numbers.

For the last few months we have been working to bring bitwise operations to Gradient, and yesterday we finally got the first relatively stable build based on the latest TensorFlow from 1.x family: 1.15 (previous versions of Gradient up to Preview 6.4 were based off TensorFlow 1.10). The new version brings support for many new features, among them tf.bitwise and gen_bitwise_ops modules. In the light of the Xnor.ai acquisition news I decided to publish these bits with a simple sample code to a work-in-progress branch, so you could start trying them early. You can view the new sample code for bitwise ops in Gradient-Samples repository, but it is as easy as

Tensor xor = tf.bitwise.bitwise_xor(x, y);
Tensor bitcount = gen_bitwise_ops.population_count_dyn(xor);

Stay tuned for the official release with TensorFlow 1.15 support. It is coming soon!

We released Gradient Preview 6.4 on Oct 15, 2019. It brings several new features and bug fixes:

• feature: inheriting from TensorFlow classes enables defining custom Keras layers and models
• feature: improved automatic conversion .NET types to TensorFlow
• feature: fast marshalling from .NET arrays to NumPy arrays
• bug fix: it is now possible to modify collections belonging to TensorFlow objects
• bug fix: enumerating TensorFlow collections could crash in multithreaded environment
• new samples: ResNetBlock and C# or Not
• train models in Jupyter F# notebook in your browser hosted for free by Microsoft Azure
• preview expiration: extended to March 2020

In Silicon Valley season 4, Jian-Yang builds an AI app that identifies pictures of hotdogs. Today I am going to make a similar app to identify C# from code fragments (in case you forget how C# looks like!).

Look at this Python trying to pretend to be your favorite language:

var = await.add(item)
switch(hello)

Everything from building and training a deep convolutional network with custom layers to the cross-platform UI will be written in C#.

Tutorial on participating in a Kaggle competition with a simple neural network built with LostTech.TensorFlow.