Hacker News Reader

Comment by krat0sprakhar | Hacker News Reader

The best part about this ML course is that all assignments are in Torch (a deep learning framework in Lua) for which Andrej Karpathy has good things to say on his blog[0]

> "Brief digression. The code is written in Torch 7, which has recently become my favorite deep learning framework. I've only started working with Torch/LUA over the last few months and it hasn't been easy (I spent a good amount of time digging through the raw Torch code on Github and asking questions on their gitter to get things done), but once you get a hang of things it offers a lot of flexibility and speed. I've also worked with Caffe and Theano in the past and I believe Torch, while not perfect, gets its levels of abstraction and philosophy right better than others."

[0] - http://karpathy.github.io/2015/05/21/rnn-effectiveness/

View on HN

Just to promote a heterodox(ford) opinion: while Torch is great, learning Theano might be a better use of a budding data scientist's time because python is used much more widely in the industry.

I think learning a new language like Lua along with Torch is probably useful if someone is doing cutting edge neural network research.

And to promote an even heterodoxer opinion, learning Deeplearning4j might be an even better use because Java is the language of corporate IT, the language of Hadoop (and therefore big data), and because Spark is based on the JVM. If budding data scientists want their neural networks to scale, they should think about distributed systems from day 1. http://deeplearning4j.org

Full disclosure: I helped create DL4J, and it is a younger framework than both Theano and Torch.

Definitely, but I think Oxford University is more interested in teaching the underlying principles than tracking what's popular in industry.

The approach generally taken by Oxford and similar is that a scripting language like Python or Lua is simple enough (or at least the bits of it needed for the course are) that anyone taking the course can pick it up with no specific learning necessary.

Here's a radical opinion: obsessing over frameworks is counterproductive; you most certainly should not be evaluating a class on a subject that is so math/concept-intensive by the class's choice of implementation coding framework. Especially since many of these frameworks are more alike than different, in the grand scheme of things.

I've found using nolearn and lasagne on top of Theano made Theano much easier to use while still being in Python with access to familiar graphics routines.

lasagne gives you ways of constructing neural network layers (implemented as Theano functions).

nolearn sits on top of lasagne and gives a Scikit learn style interface that makes it trivial to set up a standard deep network to predict values from given input data.

Using nolearn was a very similar experience for me to using the Torch7 framework.

Reply on news.ycombinator.com