Searching for tutorials and software about Deep Learning and Neural Nets? Be sure to look at my Resource Page!
Looking for Octave? Go to my Easy Octave on Mac page!

Monday, January 25, 2016

TensorFlow DeepLearning course on Udacity

As usual I'm blogging about what I'm viewing. There's a lot of needless detail because I'm an idiot, and I need a notebook or I forget everything.

Google has put up a new free course on Deep Learning on Udacity, which I'm auditing.  Credit where credit is due, the instructors is Vincent VanHoucke and the course was developed by Arpan Chakraborty.  TensorFlow is used for the practical work. 

Contrary to Professor Hinton's Coursera materials, this course is  practical, and seriously quick-paced; the video immediately starts off with AX+b, Softmax and CrossEntropy, so you should have some knowledge of linear algebra and Python.

To give you an idea of the course difficulty, here is the github repo with the assignments. You hit the first one after about an hour in. 

For rusty old geeks like me, I recommend finding a Cheatsheet for Python and Numpy. Google is your friend.It seems that Python will be the unavoidable scripting language workhorse for the rest of the decade, superseding things like Matlab and Octave.  Actually I quite like iPython's notebook interface. 

Readers may be surprised that programming fluency in whatever language is being used is not a requisite for AI work, but IMHO it really isn't. Getting a model that works is usually the hard part. 

I will blog my experiences later. For now, you can get a headstart by reading some of my old posts on TensorFlow, and especially the one detailing  TensorFlow installation with Anaconda. I just followed my own advice and it works well. BTW, the Anaconda graphical launcher is great, but it writes notebooks directly into your home directory at the top level. 

BTW, here is a trick to figure out where TensorFlow lives on your comp. And btw, the base install contains MNIST data.

$ python -c 'import os; import inspect; import tensorflow; print(os.path.dirname(inspect.getfile(tensorflow)))'
I copied the trick from this install and getting started page, linked by VanHoucke, but I still recommend conda ...unless you want to run one of the supplied Docker images. 

No comments:

Post a Comment

Hey, let me know what you think of my blog, and what material I should add!