I am doing a researach project and I am able to compile and run the theano function I created. We hope to ship a precompiled package for model selection on Neural Network, instead of spend tons of time doing cross validation.
Since we need to give different kinds of input to the model, and thus the model structure is different for different input. It is a pain to compile all possible structures each time I need them, right? So I natrually wonder if it is possible to pickle the theano function to the disk. In fact, we don't train the model on theano. The parameters, data and labels are all treated as input from other sources(trained on other platform). Therefore, we dont have shared variable just pure symbolic tensorvariable in the model. The reason we need Theano is its exact gradient capability which is a neccessary step of our algorithm. When I try to pickle the function, it said it can't pickle fortran objects. I guess the compiled theano function is related to fortran. I tried the theano.misc.pkl_utils.dump but it does not work either. I really appreciate any help on this. -- --- 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.
