Share good ideas and conversation.   Login or Take a Tour!
comment by ButterflyEffect
ButterflyEffect  ·  1440 days ago  ·  link  ·    ·  parent  ·  post: Google open sources their deep learning library

This seems cool, but I don't fully understand what this means. Obviously open source is good, but what are the benefits of this and does anyone have ideas on how to use it?




thundara  ·  1440 days ago  ·  link  ·  

I can't tell if it implements any novel algorithms not available through other libraries, but it does wrap a number of separate functionalities and provide them with a nice pretty UX. It also lets you express functions symbolically so that it can quickly calculate their derivatives for the purpose of optimization problems. A lower the barrier to non-coders and a streamlined experience are pretty useful things to have.

ButterflyEffect  ·  1440 days ago  ·  link  ·  

Any recommendations for where to start with this? I'm going to give the tutorials a shot and see what happens, trying to learn Python in my free time and this seems like it would tie in nicely with that.

thundara  ·  1439 days ago  ·  link  ·  

Oh scratch what I said before, it looks like there isn't an actual GUI associated with this library, despite the graphics on the front page... That makes the pluses of this library a bit less, as the competing packages (scikit-learn, sage, scipy, etc) already cover similar functionalities, albeit not all in one integrated system.

If you are starting with Python, I would recommend first following the basic tutorials they have, then pick a basic project or task you'd like to try (i.e. digitize my handwriting), find a few examples of what approaches other people have used (i.e. RNNs / CNNs ) and work through the basics of loading your images, and training / testing whatever algorithm you choose to try.

ButterflyEffect  ·  1439 days ago  ·  link  ·  

Also holy shit the whole Virtualenv thing, which I still don't fully understand how it works beyond creating a "container" with individual installs/packages/etc. so you can better manage projects is so much easier to figure out in their instructions than on new-coder.io, where errors galore appeared while trying to figure out their instructions.

Going to have to read up on RNNs and CNNs (not the tv channel?!), thanks thundara!

thundara  ·  1438 days ago  ·  link  ·  

Virtualenv is pretty simple! Instead of installing packages to a system-wide folder, it places them locally and tells python to search that folder instead of the system's one:

    $ virtualenv .

    $ source bin/activate

    $ pip install pkg1 pkg2 ...

    $ python run_program...

The underlying implementation has changed a bit over the years, but from a user's point of view, it's been extremely easy to work with.

Also spotted today