Gradient
TensorFlow for .NET/C#
After almost 3 year journey our TensorFlow binding for C# is ready for prime time!
The API is stable, samples rock, and we are ready to get the first commercial customers via Azure Marketplace!
The last bit turned out to be the hardest part recently, but it is the one I am excited about as the product owner.
How do you sell a software framework? Any non-zero price is a huge barrier to entry, because in many companies even slightest expenses have to go through a long approval process.
Well - turns out Azure lets you to exactly that in the best way possible: you can publish it as a SaaS, which gives you ability to charge for it 1) on a Pay-As-You-Go basis 2) under client's existing Azure subscription. And the best part is that the product does not even have to always run on Azure to take advantage of that.
So from the client perspective: he hits "Get" button on the website, which redirects to Azure, where he can tie the product to his Azure Subscription. In return he gets a secret key, which he then uses to unlock the license. The framework then simply track how much time the app built on it is running and reports to Azure. At the end of the month the client gets his monthly Azure bill, that already includes the charges for your framework, proportional to the actual usage.
Somehow I overlooked sharing this cool demo on Indie Hackers back in February. If you are interested in AI robotics, or a Unity game developer curious about deep learning, check it out!
It is an implementation of a state-of-the-art reinforcement learning algorithm in C# + TensorFlow, that uses recently released Unity ML Agents framework.
Some new features include: custom neural network layers and models, more performant marshaling, support for F# Jupyter Notebooks on Azure, and more samples.
Now you can train neural networks in browser: compute is provided by Microsoft Azure.
Also, I am hungry!
Published an article about making an open source song lyrics generator with ASP.NET Core + Gradient.
-
1
I am throwing this out there, but could this be used as a content generator to get traffic from google. How would google know it was written by a human or not
Just made the source code for a deep learning-powered song lyrics generator Billion Songs, that I demoed on
#GlobalAI available on GitHub: https://github.com/losttech/BillionSongs
The repository still lacks training and running instructions. I will work on them over the next day or two.
#deeplearning #ai #lyrics #tensorflow #gradient #gpt2 #tfworld
Added voting, and a chart of top 10 best songs to the new sample project, built on top of Gradient: http://billion.dev.losttech.software:2095/top
It generates random lyrics, and now also lets you vote on the best songs!
Built a demo on top of Gradient: a website, that uses AI to generate random song lyrics: http://billion.dev.losttech.software:2095 to promote this project.
This is the first dev version, so don't expect texts to stay. Later ones will have more features, and, possibly, generate better lyrics :)
This version is fully cross-platform: you can now create deep learning models with Gradient on #MacOS and #Linux.
Also, we bundled XML documentation into the NuGet package. Which means you will get tooltips in the IDEs about TensorFlow APIs.
#AI #ML #TensorFlow #CSharp #dotnet #dev #programming
-
1
Best of luck on the release.
Did you want to try deep learning, but never liked dynamic languages like Python? Try a preview of Gradient, fully-featured TensorFlow to C# binding.
#ai #ml #tensorflow
The goal of the 3rd preview was to ensure Gradient is alive, and can kick as hard as anything else. So I joined Kaggle, and started hacking.
Kaggle - is a giant playground, owned by Google, where data scientists compete for fame and coins (but mostly fame). Kaggle provides data, and participants need to build a machine learning model, that can predict, guess, classify, etc based on that data.
In less than 4 hours, I was able to make my first submission with an average score without any prior experience in the matter. That dogfooding helped to uncover multiple rough edges, which were documented and or fixed in Preview 3.
A more detailed story with the detailed competition description, the whole process of the initial data analysis, implementation with complete code samples, and submission is covered in the blog post: https://habr.com/post/437174/
1 Comment