Hi,
I had brought up the issue of Theano library on this forum once before and I was told It is being considered.
But I haven't seen any progress on this front.
Some recent threads seem to be refering to Theano's GPU capability Only.
I would like to point out that Theano can work with CPU Or GPU.
When working with CPU, it leverages the BLAS library for performance.
I see an approach, where
1. I train my Theano algorithms on GPU on Zipline on my own hardware and save the weights
2. And would like to use Theano on CPU/BLAS library in Quantopian with those saved wieghts.
This approach would not require Quantopian to support Theano/GPU if that is logistically more complex.
When it comes to neural network training, Traning is computationally expensive and can be done with GPU on our own hardware.
The application of trained networks or incremental optimizations can be done in plain Theano/CPU.
So could we have support for Theano/CPU as early as possible.
Thanks
Sarvi
