What to go to from pylearn2

[Isaac Gerg]

So it seems that pylearn2 is without a developer.

Is there a recommended path forward?  I would like to contribute to a framework but want to start where the work will be fruitful long term.

Thanks,

Isaac

[Nicu Tofan]

I would be curious to know what was the experience of somebody who switched from Theano and PyLearn2 to using TensorFlow.

I don't have the time to look into that myself nowdays.

--

---
You received this message because you are subscribed to the Google Groups "pylearn-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pylearn-dev...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ian Goodfellow]

I'm using TensorFlow obviously, albeit the internal Google version. Is
there anything in particular you want to know?

[Nicu Tofan]

Thanks for offering. You have the developer view, while I try to get the perspective of a newcomer with similar background to build an estimation on how much time I should set aside for understanding TensorFlow well enough to use it in production and, maybe, contribute something back in the process.

I was looking for statements like: the concepts in TensorFlow are quite easy to grasp because the way it works is very similar to Theano or TensorFlow looks similar to Theano but only on the surface.

There are some happy people that can use libraries/projects effectively without delving into the project internals. I'm not one of those happy people. :)

[Ian Goodfellow]

Anything you know how to do in Theano is straightforward in
TensorFlow. That's why Keras was able to add TensorFlow support so
quickly.
TensorFlow also introduces a lot of new concepts to do things Theano
doesn't do, like run multiple threads, stream data from files, etc.
For example, see:
https://www.tensorflow.org/versions/master/how_tos/threading_and_queues/index.html