Conversation
…CTORs to init(Ops) and added call(Ops, ...).
|
@karllessard the |
|
@deansher , any chance you can take a look at this work? BTW @JimClarke5 , you'll need to rebase it. For the ops.pb, normally it should have no impact that you moved |
|
Thank you, @karllessard, for inviting me to review. I have lately been exploring different aspects of deep learning that don't leave me the time to participate in tensorflow/java for now. |
|
@JimClarke5 there is a lot of conflict in this PR, I think it's because we have merged a few of your other PRs before it, can you please rebase so I can start to take a look at it? |
3 participants
These are the first set of model layers, fairly simple ones without back propagation.
I have also revamped the
TestSession,GraphTestSessionandEagerTestSessionto use lambdas for common operations between theGraphandEagermodes. This will need to be updated with #272, "Tensor.dataToString" when that is merged.This PR supersedes #65, "Adding layers (based on keras) supporting multiple outputs"
This PR is dependent #191 "TensorFormat enum". I temporary created a copy of TensorFormat until #191 is merged.
The layers framework supports multiple inputs and outputs, but some layers are restricted to single input/output by design,
(e.g. Dense).
This revision moved the
Opsparameter out of the CTORs, and creates a new method,init(Ops), and adds anOpsparameter to the call method [e.g.call(tf, ...)]. see issue #202.SequentialLayersTest tests chains layers together like in the Sequential Model.