w and V variables reversed from paper in Tensorflow implementation?

[gmeehan96]

Hi - really like the OC-NN idea and looking forward to trying it myself. Just one thing I've noticed which, although I don't think it will affect how the code runs, caused me quite a bit of confusion when reviewing the tensorflow implementation of the OC-NN here (https://github.com/raghavchalapathy/oc-nn/blob/master/models/tf_OneClass_NN_model.py). The variable names for w and V in the nnScore function(s) seem to be swapped from their role in the paper. For example, in line 109, the score is defined as tf.matmul(g((tf.matmul(X, w))), V), where X is being multiplied by w, whereas in equation 3 in your paper it is the weight matrix V which is multiplying X. I'm pretty sure though that when you initialise these weights later on in lines 285-6 (with w1 as w and w2 as V), you then have them as the correct dimensions, so that w is playing the role of V and V becomes the vector w.

If I'm right, I thought this was worth flagging as it has the potential to confuse people trying to understand the code; if I'm wrong I apologise, and wouldn't mind some help in explaining what was going on!

Thanks again!

[raghavchalapathy]

Please find the updated code and experiments in the files mentioned below, The experiments can be found here <https://github.com/raghavchalapathy/oc-nn> from src.models.OneClass_SVDD import OneClass_SVDDfrom src.models.config import Configuration as Cfg

…

On Thu, Jun 6, 2019 at 3:24 AM GMU10 ***@***.***> wrote: Hi - really like the OC-NN idea and looking forward to trying it myself. Just one thing I've noticed which, although I don't think it will affect how the code runs, caused me quite a bit of confusion when reviewing the tensorflow implementation of the OC-NN here ( https://github.com/raghavchalapathy/oc-nn/blob/master/models/tf_OneClass_NN_model.py). The variable names for w and V in the nnScore function(s) seem to be swapped from their role in the paper. For example, in line 109, the score is defined as tf.matmul(g((tf.matmul(X, w))), V), where X is being multiplied by w, whereas in equation 3 in your paper it is the weight matrix V which is multiplying X. I'm pretty sure though that when you initialise these weights later on in lines 285-6 (with w1 as w and w2 as V), you then have them as the correct dimensions, so that w is playing the role of V and V becomes the vector w. If I'm right, I thought this was worth flagging as it has the potential to confuse people trying to understand the code; if I'm wrong I apologise, and wouldn't mind some help in explaining what was going on! Thanks again! — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#5?email_source=notifications&email_token=ABQESTFQYF2ITXKCM3BIESLPY7ZGPA5CNFSM4HUAIVU2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4GX2PXVA>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABQESTGTJR4ZKKWDXYFY2L3PY7ZGPANCNFSM4HUAIVUQ> .