As someone who had to go through pain of using caffe, struggled with steep learning curve of Lua/Torch and frustrated by lack of simple features (train on gpu/

As someone who had to go through pain of using caffe, struggled with steep learning curve of Lua/Torch and frustrated by lack of simple features (train on gpu/test on cpu) of Theano. You could not be more wrong. Having a well written, consistent and portable system is a huge plus. Tensor Flow is likely to do to deep learning what Hadoop did to Big Data.