Don't move complex to the GPU in the new back-end.

Lower the number of iteration for local_add_mul_fusion

* Adding issue and PR template to Theano repo

* making URL into text + removing a line

* adding the theano tag to stackoverflow comment

Jenkins buildbot

Fix to use new GPU backend op for multinomial

Use types that exist for floats.

[WIP] Scan doc internal

Don't crash when pydot isn't there.

Mixed

Fix d3viz multi-level/inheritance tags

fix conv2d_grad_wrt_weights bug

clearer name the properties.

Pad descriptors to 3d if the input tensor is less than that and up the limit to the one supported by cudnn (8d).

Documentation update: conv2d_grad_wrt_inputs moved to theano.tensor.nnet.abstract_conv

Make nose-paremetrised optional for theano.sparse