[scikit-learn] Fwd: Scikit-learn MLPRegressor Help

[Alekh Karkada Ashok]

Hey All,

I chose MLP because they were images and I have heard MLPs perform better.
My application is detecting body parts from these images and therefore, the
mapping would be pretty non-linear and this was my idea behind selecting
MLP. Otherwise, I would have to engineer high dimension features by hand. I
have 2030400 pixels and making higher dimensional features would require a
lot more memory.

Where do you want to me to open the issue? GitHub? I don't think the error
is only in documentation. Because when Y is [2030400,1] there is no
MemoryError (treated as 2030400 samples with a single feature) and when I
try to fit [1,2030400] it throws MemoryError. If the case was memory, both
should have thrown the error right?

I am still a novice but I am fairly good with Python. I am taken aback by
scikit's sheer beauty and simplicity. I would love to contribute code to
it. Can you please tell me how I can get started?

Thanks a lot!

On Sat, Dec 3, 2016 at 11:40 PM, Andy <t3kcit at gmail.com> wrote:

>
>
> On 12/03/2016 05:29 AM, Gael Varoquaux wrote:
>
>> On Sat, Dec 03, 2016 at 03:08:00PM +0530, Alekh Karkada Ashok wrote:
>>
>>> I want use the Scikit-learn's MLPRegressor to map image to image. That
>>> is I
>>> have a numpy array of size [1000,2030400] (1000 samples, 76800x3 (RGB)
>>> pixels).
>>> Corresponding labelled images I have. Therefore Y is also [1000,230400].
>>> But
>>> according to documentation:
>>>
>> 1 thousands samples and 2030 thousands features: you are using the wrong
>> tool, I multi-layer perceptron model will be too complex and overfit in
>> these settings. I would suggest a ridge.
>>
>>
>> These are images! Don't use ridge, use a convolutional neural network.
> Our MLP is not convolutional, it will not be useful.
> There is a lot of material out there on how to use covolutional neural
> networks
> for image labeling (it looks like you have one label per pixel, not per
> image)
>
> _______________________________________________
> scikit-learn mailing list
> scikit-learn at python.org
> https://mail.python.org/mailman/listinfo/scikit-learn
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-learn/attachments/20161204/2a938473/attachment-0001.html>