I’ve been so busy the last few days, that I didn’t even notice that Microsoft CNTK version 2.0 deep learning library was released on June 1, 2017. Very cool! See https://docs.microsoft.com/en-us/cognitive-toolkit/.
The CNTK library contains very sophisticated, very powerful, machine learning functions. CNTK is written in C++ so that it can take advantage of GPUs, if available. But I prefer to use the CNTK Python language API rather than wrestling with C++.
To install CNTK with Python, you need an Anaconda distribution. I like the Anaconda version (4.1.1) that has Python 3.5. Once you have Anaconda/Python, you can use the Python PIP utility to install CNTK using a .whl file. In the image below, I use the –upgrade and –no-deps flags because I had an earlier version of CNTK installed. Note that I installed the CPU-only version because my desktop machine doesn’t have a GPU.
After I installed the 2.0 version of CNTK, I refactored one of the logistic regression examples I found in the CNTK documentation. CNTK is intended mostly for deep neural networks, but CNTK can do logistic regression too.
I’m very excited to explore CNTK version 2.0. I use both Google’s TensorFlow and Microsoft’s CNTK. I prefer CNTK — it just feels better to me. I know that’s subjective, and TensorFlow has a big lead over CNTK in terms of usage, but I’m going to place my bet on CNTK.