Getting started with Deep Learning for Coders I

After giving a try with more math oriented ML/DL offers I was attracted by the name of this course. While I am not against the explanations given by those in love with the math notations I feel more at home when looking at code.

I wanted to not depend on an Internet connection, and be able to study this stuff no matter where I’d happen to be, so I went shopping for a laptop. I wound up buying what I had never pictured myself doing, a game laptop.

So far quite pleased with it but setting it up for the course was a bit frustrating. I wanted to stay away from Python 2.7, so when when installing Anaconda 3 I went for the latest Python 3 version, 3.6. Long story short, there is no libpython3.6 so, after many frustrating attempts I discovered that there was a version for 3.5, so I downgraded python.

Another source of frustration was the installation of cuda, it does not install the compiler or the utils. After some searching found out how to do it.

Finally got it all working except that I was running out of memory, it seems that the GTX 1060 6GB memory is not big a enough for batches=64. Through trial and error found that it can handle batches=52. Yoohoo!

Thoroughly happy now. will complete the rest of of lesson 1 tasks.

1 Like

For this course you want python2, part II is python 3 and tensorflow (YEAH!) and I believe all future parts would be python3 as well.

Unless you already know what you are doing, I highly suggest just sucking it up and using python2. Trying to convert everything to python3 will distract you so much from the lesson at hand I don’t think you will come out ahead.

1060 you will likely need to lower batch size, it’s not a huge factor as the 1060 still greatly outperforms the AWS P2 instance at $0.90/hr.

@dradientgescent / @lucho The python version doesn’t matter as long as you’re willing to do a little tweaking. I’ve been doing Part I of the course using Python 3.5 in Anaconda without any issues other than a few minor google’able ports from Python 2.7 to 3.5.

All of the major libraries, and in particular Theano/Cuda are now compatible with 3.5.

Python isn’t so hard it’s the tensorflow that is more time consuming. But with future the differences with Python 2 are not that big of a deal and doesn’t really affect the lesson material (which is most important) but I wasn’t happy about using 2 either. Just felt time better spent on lesson material.

I just wanted to let anyone thinking of going the the laptop/desktop way that it is feasible, even with Python 3.5 these days. Just be prepared to pay the price in lots of time devoted to getting started. This certainly set me back nearly two weeks, but I am pleased with the setup.

Fortunately I am between assignments (actually recently retired, but planning to get busy in this field in the future :wink: so I can dedicate plenty of time to study.

Thanks for the tips and ideas. I will certainly need plenty of help as I move froward.

At least you didn’t buy a laptop with an AMD card :grin:

:sweat_smile:
I know nothing about game computers so it took me a few weeks to learn stuff like that, NVidia was the way to go, GPU memory is important, cpu not so much. I’m far from knowing enough but I’m satisfied so far. The unit I bought was an MSI GS43VR and I wound up paying much more than I had originally budgeted. I hope that it is useful for my training, given that short time span of things these days, at least 2 years.