dhruvramani / C2AE-Multilabel-Classification

Tensorflow implementation for the paper 'Learning Deep Latent Spaces for Multi-Label Classfications' in AAAI 2017
101 stars 24 forks source link

why the Macro F score or Micro F Score are so low in validation.log? #3

Open shaomai00 opened 6 years ago

shaomai00 commented 6 years ago

I saw this in validation.log Test : Coverage = 719.66, Average Precision = 0.18053248555916795, Micro Precision = 0.06627056672003306, Micro Recall = 0.6256742172322824, Micro F Score = 0.11984691841392547 => Test : Macro Precision = 0.011802305630313965, Macro Recall = 0.14954214765979945, Macro F Score = 0.021877804367018278 => P@K = [ 0. 0.02093145 0.0188383 ]

I'm a little confused about why the F score is so low, while in your paper is C-F1 is 48.6 and O-F1 is 67.6. and why is P@K so strange _
How could I know this model is really worked on 'delicious' dataset?

please help~ thank you very much~

chihkuanyeh commented 6 years ago

Hi, I am the author of this paper. We implemented our method in Matlab and here is the GitHub link: https://github.com/yankeesrules/C2AE Feel free to try it out as it should reproduce results in our paper.

DanqingZ commented 6 years ago

@chihkuanyeh , if there is something wrong in this repo, could you give some suggestions how to improve this? I am trying to use your model, but I prefer deep learning framework, like tensor flow/Keras instead of using Matlab code. Is there any reason we cannot use deep learning framework for C2AE? Thank you.

lwzzzzzzzz commented 5 years ago

your guys problem is solved?I am so puzzled about it~

chihkuanyeh commented 4 years ago

maybe you can try this implementation? https://github.com/znhub/C2AE