As long as the feature ordering has a meaningful spatial component (as is almost always the case when you are dealing with raw pixels as features) CNNs will almost always be better. CNNs actually have a lot fewer parameters than MLPs (depending on architecture of course) because of weight sharing among the parameters of the convolutional kernel within a feature map.
On Sat, 3 Dec 2016 at 23:00 Alekh Karkada Ashok <alek...@gmail.com> wrote: > No, I am not saying it is better than CNN, but my images aren't real-life > images but computer generated silhouettes. So CNN seemed to be overkill. > I'll revisit CNN. I resized the images and converted it to grayscale. Now I > am feeding [1,4800] now and I am getting good output with MLP. I looped > over all my images and used partial_fit to train each one. > I didn't get what you meant by MLPClassifier doesn't support multi-output. > Thanks for the help! > > On Sun, Dec 4, 2016 at 2:11 AM, Andy <t3k...@gmail.com> wrote: > > > > On 12/03/2016 03:10 PM, Alekh Karkada Ashok wrote: > > > Hey All, > > I chose MLP because they were images and I have heard MLPs perform better. > > Better than a convolutional neural net? Whoever told you that was wrong. I > usually don't make absolute statements like this, but this is something > that is pretty certain. > > > 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? > > MLPClassifier actually supports multi-label classification (which is not > documented correctly and I made an issue here: > https://github.com/scikit-learn/scikit-learn/issues/7972) > MLPClassifier does not support multi-output (multi-class multi-output), > which is probably what you want. > > > _______________________________________________ > scikit-learn mailing list > scikit-learn@python.org > https://mail.python.org/mailman/listinfo/scikit-learn > > > _______________________________________________ > scikit-learn mailing list > scikit-learn@python.org > https://mail.python.org/mailman/listinfo/scikit-learn >
_______________________________________________ scikit-learn mailing list scikit-learn@python.org https://mail.python.org/mailman/listinfo/scikit-learn
