Microsoft's CNTK is a useful open source toolkit for developing deep learning trading systems in Windows/Linux that can leverage both cpu/gpu processors. Model architectures MLP, CNN, RNN, LSTM and even Reinforcement Learning (RL) models can be trained using native "BrainScript" or APIs for Python 3.4 or C++. Models can be evaluated directly from within NinjaTrader 8 via an evaluation dll provided. A C#.Net API is planned so models can be built directly from an NT8 add-on as an alternative to running BrainScripts or using the C++ API.
Here are some useful links:
Source code:
CNTK Wiki:
CNTK V2 Beta 5 Review:
CNTK Deep Dive and Hands-On Tutorial
CNTK has been my first choice of toolkits since it runs natively on Windows and why I passed up TensorFlow (although a windows option is around the corner). The CNTK Wiki has some excellent tutorials but lacks third-party training resources like Udacity that favor TensorFlow. Were fortunate to have two tech giants battling for toolkit supremacy with a commitment to democratizing AI by delivering outstanding tools.
Kudos to NJAMC, rlepae and others for sharing their ML/DL links, experiences and insights on this forum. I wanted to start a thread to collect the same regarding the use of CNTK with the caveat I really don't have the bandwidth to support technical questions.