PyBrain so slow When I train a simple neural net using all the defaults it takes very long (I have played around with changing some of the defaults to little success). 
Are there any tricks I am missing with PyBrain, or implementation of neural networks in general?
My dataset has has 100,000 observations, with one feature (covariate) and a binomial response ([0,1]). 
example code:    
    net = buildNetwork(1, 1, 1, bias = True)

    ds_train = SupervisedDataSet(1, 1)
    ds_train.addSample(x,y)

    trainer = BackpropTrainer(net, ds_train)

    trainer.train()

I was actually using this as a test case for the larger regression I am looking to do which will have 20 features. Right now it seems that pybrain is too slow and I'm currently looking into Theano, is this a good idea?
 A: Former developer of PyBrain here. I programmed arac a few years back and have moved away from using PyBrain about 3 years ago.
PyBrain does not make use of mini batches; if your data set is of size 100'000, there will be a python for loop doing 100'000 iterations--and that is quite slow.
I do recommend to use Theano instead. Arac will make it better, but Theano will probably always be faster.
A: I started writing an answer suggesting that you might be having convergence problems. If your one feature is only weakly predictive of the outcome variable, then it might take many iterations for the network to converge (or it might not converge at all, and training would only stop due to some some other criterion).  However, if you're actually calling BackPropTrainer.train(), then that only performs a single epoch of training, so that is probably not the issue. 
There are some things in Python that I find surprisingly slow (like the new Collections!), so you may want to profile your code (see Python docs), or ask for suggestions on StackOverflow.
PyBrain also has a "high-performance" module called arac that moves a lot of the computation into C++. I've never used it, but they claim to get considerable speed-ups and it might involve minimal changes to your existing code. 
