hi david, I'm working on porting theano to android using python for android. As you've said, to make sure it doesnt compile c++ lib at runtime, I will have to disable the cxx flag. However, that directs me to cross-compile scipy. I did that as well but it turns out that the computation is very slow using conv2d based on python. My testing failed within 60 seconds. I hope it helps.
Wang On Thursday, July 16, 2015 at 11:34:50 AM UTC-7, David Liebman wrote: > > hi, > > I thought that if I loaded theano with THEANO_FLAGS set to cxx='' and > blas.ldflags='' that theano would not compile c or c++ code as it ran. Is > this so? > > I have a conv neural net that I'm trying to run on an android device (!). > I have done some training with my desktop computer and I have saved weights > and biases. I would now like to try to run the nn on an android device > using python-for-android and buildozer. There is a restriction on the > mobile device that no compilation be done but I thought that if I kept it > all to python that theano might work. It's very much a proof of concept > thing. > > Has anyone ever done anything like this? I am just starting out and I have > trouble with even the simplest code. The following crashes my test app: > > import theano ; print theano.config ; > > thanks for your time. > -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
