Francesco

1. What I mean is moving _all_ the helper methods inside `Layer`. The `get_output` could traverse the list of `Layer`s recursively even being a `Layer` method, see e.g. the [code...

Thank you for your answers. I am unfortunately forced to call `get_output(l)` on each individual layer because I need to know their symbolic output shape to select the parameters of...

> But did you fully appreciate the impact of what happens if you do [l.get_output() for l in my_output_layers] instead of get_output(my_output_layers)? (#97, #104) Having get_output() as a Layer method...

> The only solution we saw when discussing #104 was to make the cache keys weak references to the input variables (the inputs passed to get_output, not the outputs of...

I am sorry for the long time it took me to reply: I didn't have time to keep this conversation going lately, but it's a topic I really care about....

> > The current interface hides these debug tools from the user > > I don't see how. Setting the test_value on the input is easy, accessing it on the...

I like the optional `cache` dictionary idea!! It doesn't break the API and solves all the use-cases that I was hoping to address at the very small price to take...

Let me revive this conversation! I was wondering if there is any update on this issue. Do you plan to add it to the 0.2 or 0.3 milestone by any...

> What you currently have is basically a macro function [...] It does not encapsulate the layers. I see your point, I will fix it! Thanks!

@f0k I am looking into your suggestion. I cannot apply [your code](https://github.com/Lasagne/Lasagne/issues/659#issuecomment-216241685) exactly as is, because my layer has just _one_ input layer. I modified it like this: ``` python...