Re: [pylearn-dev] a convolutional net with regression output

[David Warde-Farley]

As noted in the repository README, pylearn2 is not maintained anymore.
You'd be better off looking into another framework like Blocks,
Lasagne, Keras, or TensorFlow.

On Tue, Apr 19, 2016 at 1:40 AM, Stone <esoft...@gmail.com> wrote:
> hi! I am very naive here, thanks a bunch for the help
>
> what I am trying to do is to build a convolutional net, but with regression
> output (not softmax classification); the network structure might look like :
> the input is a square image, but output is a continuous value
>
> the input space is specified as:
>
> input_space: !obj:pylearn2.space.Conv2DSpace {
>
> shape: [8, 8],
>
> num_channels: 1
>
> },
>
> but the last layer is a !obj:pylearn2.models.mlp.LinearGaussian
>
> but I got the errors: "ValueError: The value of a VectorSpace batch must be
> 2D, got 3 dimensions"
>
> I guess the last layer requires "vector space"?
>
> How to resolve this problem? using CompositeSpace? what yaml will look
> like?
>
> mega thanks
>
>
> --
>
> ---
> You received this message because you are subscribed to the Google Groups
> "pylearn-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to pylearn-dev...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.