David Warde-Farley

Losing contact with Bokeh server causes training halt

7

@jbornschein just mentioned this to me and I think it's probably not desirable behaviour.

There are classes defined in `__main__`. We could fix this by moving them to a separate file.

Arrangement of files in the examples (__init__.py, __main__.py?)

6

I'm worried that people will be confused by all the example scripts being called `__init__.py`, and think this is required or something. Maybe we should mention this in the readme...

Examples meant to be instructive ought to be more thoroughly commented. There are some non-obvious bits even in the MNIST example.

Transformers never call their next_epoch

1

Apparently.

Allow user to preempt Theano import via environment variable

1

Fuel is supposed to be Theano-independent. This seemed to be [causing somebody trouble on the mailing list](https://groups.google.com/forum/?utm_medium=email&utm_source=footer#!msg/fuel-users/lXfXvrkA8zA/VcrouU3tCQAJ) so maybe we should just provide a short circuit.

It doesn't mention that Fuel tries to import Theano and use `theano.config.floatX`.

Someone should wrap this.

Things like load() and the difference between load() and open(), for starters

Threaded jpeg4py transformer

7

It would be really useful to be able to a) leverage libjpeg-turbo to do faster decodes, b) be able to parallelize a minibatch across CPU cores. [jpeg4py](https://pypi.python.org/pypi/jpeg4py/0.1.0) is a low-level...