TensorFlow.FSharp is an open source project that develops a TensorFlow API for F #. In addition, a DSL is created for working with numerical models.
What is TensorFlow.FSharp?
TensorFlow.FSharp is an open source F# API for TensorFlow currently under development. It is provided together with FM, a Domain Specific Language (DSL) for writing numerical models in F#. FM is written in F#.
TensorFlow.FSharp is completely implemented in F # and is different from TensorFlowSharp, an existing F # and C # TensorFlow API. According to the description in the repository on GitHub , TensorFlow.FSharp should contain more functionality than TensorFlowSharp. Within the repository some features of TensorFlow.FSharp are presented as examples . One of these examples is the ImageClassifier. This can classify an image into one of 1000 categories, using Resnet50 in the background. Also AttGAN is one of the presented examples. With AttGAN the attributes of faces can be visibly changed. This feature uses Generative Adversarial Network (GAN) and is still described as work in progress, as well as TensorFlow.FSharp in total. Features of the API planned for future development include GPU and TPU execution.
FM: Write KI Models in F#
FM is a DSL for the use of F# in the AI environment. It aims to support the writing of numerical functions and AI models in F #. Here, neural networks are explicitly included. Models written in FM can be passed to algorithms for optimization and training using automatic differentiation without any changes to the model code, and can be executed by TensorFlow on GPUs and TPUs.
The makers describe FM as a proof of concept for the feasibility of AI models in F #. A complete description of TensorFlow.FSharp and FM, along with some code examples, can be found in the GitHub repository