I've been fiddling around with tensorflow for a while now. I got it to work and trained a simple neural network. It's not much to look at, but it can tell a pet from a car with 99% accuracy.
I build this by installing tensorflow and training one of the sample networks on datasets of pictures of cars and of pets (cats and dogs in this case)
Lord Almighty tensorflow did not want to work on my computer. I had to install it like three different times to get it to even output a test and the documentation is not very accessible to a beginner.
I am most proud of the fact that I got it to work. I put data into a convolutional neural network and it puts out answers just like that. It's a thing of beauty.
I've learned a little about machine learning and a lot about python. I didn't end up going with it, but I figured out a way to apply convolutional maps to images by the hundreds. It's like ten lines in python. Amazing. Also I figured out a way to include my cat in my project, so mission accomplished.
I want to break the whole thing down and build my own neural network from the ground up. I'll do it to, in the fullness of time.