2-class problem

Question

2-class problem

Chenpuh opened this issue 3 years ago · 12 comments

Is 2-class the following code in fairseq/data/raw_audio_text_video_dataset.py?
If so, why are the sentiment_score 0 and 1?

            self.emotion_dictionary = {   #modei senti 2 class
            
                '0': 0,
                '1':1
            }

Answer 1 · 2021-04-07T01:24:09.000Z

It is just assigning a class label in two class cases. In MOSEI and MOSI , same script used to assign the real score.

…

On Wed, Apr 7, 2021, 13:22 Chenpuh ***@***.***> wrote: Is 2-class the following code in *fairseq/data/raw_audio_text_video_dataset.py*? If so, why are the *sentiment_score* 0 and 1? self.emotion_dictionary = { #modei senti 2 class '0': 0, '1':1 } — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#2>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA4FGQD2N7YAFCY2XYLUZTTHOXV5ANCNFSM42PZJBVA> .

Answer 2 · 2021-04-07T01:32:49.000Z

Sorry, I don't understand what you mean.

in the paper MSAF Multimodal Split Attention Fusion:
For binary classification, we consider [-3, 0) labels as negative and (0, 3] as positive.

I want to know what 0 and 1 stand for in your repository?

Answer 3 · 2021-04-07T01:57:49.000Z

It is for the binary sentiment classification. 0 for negative and 1 for positive.

…

On Wed, Apr 7, 2021 at 1:33 PM Chenpuh ***@***.***> wrote: Sorry, I don't understand what you mean. in the paper *MSAF Multimodal Split Attention Fusion*: *For binary classification, we consider [-3, 0) labels as negative and (0, 3] as positive.* I want to know what *0* and *1* stand for in your repository? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA4FGRL3W3IOLW3NPF3FZTTHOY43ANCNFSM42PZJBVA> .

-- [image: Augmented Human Lab] <http://www.ahlab.org/> [image: uni] <https://www.auckland.ac.nz/en/abi.html> Gayal Shamane Ph.D. Candidate Augmented Human Lab Auckland Bioengineering Institute | The University of Auckland

Answer 4 · 2021-04-07T02:07:51.000Z

In *.csv, the format is as follows:

FileName,sentiment_score
03bSnISJMiM_1,2.400000095367432
03bSnISJMiM_3,-1.0
03bSnISJMiM_4,-1.75
03bSnISJMiM_5,0.0
03bSnISJMiM_6,0.0
03bSnISJMiM_7,0.8000000119209291
03bSnISJMiM_9,0.20000000298023224
03bSnISJMiM_11,-0.5
03bSnISJMiM_12,2.200000047683716
0h-zjBukYpk_1,-0.4000000059604645
0h-zjBukYpk_2,-2.799999952316284
0h-zjBukYpk_3,-0.20000000298023224
0h-zjBukYpk_4,-1.399999976158142
0h-zjBukYpk_5,0.8000000119209291
0h-zjBukYpk_6,-1.399999976158142
0h-zjBukYpk_8,0.8000000119209291

They are all floating point numbers. How can we make them 0 or 1?

Answer 5 · 2021-04-07T02:31:46.000Z

no no in MOSEI and MOSI you need to predict sentiment scores. So itis basically a regression task. In the paper, I have cited CMU-MOSI and MOSEI. I know it is bit different :).

…

On Wed, Apr 7, 2021 at 2:08 PM Chenpuh ***@***.***> wrote: In *.csv, the format is as follows: FileName,sentiment_score 03bSnISJMiM_1,2.400000095367432 03bSnISJMiM_3,-1.0 03bSnISJMiM_4,-1.75 03bSnISJMiM_5,0.0 03bSnISJMiM_6,0.0 03bSnISJMiM_7,0.8000000119209291 03bSnISJMiM_9,0.20000000298023224 03bSnISJMiM_11,-0.5 03bSnISJMiM_12,2.200000047683716 0h-zjBukYpk_1,-0.4000000059604645 0h-zjBukYpk_2,-2.799999952316284 0h-zjBukYpk_3,-0.20000000298023224 0h-zjBukYpk_4,-1.399999976158142 0h-zjBukYpk_5,0.8000000119209291 0h-zjBukYpk_6,-1.399999976158142 0h-zjBukYpk_8,0.8000000119209291 They are all floating point numbers. How can we make them *0* or *1*? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA4FGT4JQRTSEIP6IUKTHDTHO5AHANCNFSM42PZJBVA> .

-- [image: Augmented Human Lab] <http://www.ahlab.org/> [image: uni] <https://www.auckland.ac.nz/en/abi.html> Gayal Shamane Ph.D. Candidate Augmented Human Lab Auckland Bioengineering Institute | The University of Auckland

Answer 6 · 2021-04-07T02:50:30.000Z

In mosei/BA senti/label_file_*.csv, the format is as follows:

FileName,SEVEN,TWO
5Fs_7A_V2kA_0,1,1
5Fs_7A_V2kA_1,1,1
5Fs_7A_V2kA_2,1,1
5Fs_7A_V2kA_3,1,1
5Fs_7A_V2kA_4,1,1
Wfn7XvSEwUA_0,-2,0
Wfn7XvSEwUA_1,0,non
Wfn7XvSEwUA_2,2,1
Wfn7XvSEwUA_3,1,1
Wfn7XvSEwUA_4,0,non
Wfn7XvSEwUA_5,1,1
Wfn7XvSEwUA_6,1,1
Wfn7XvSEwUA_7,1,1
47472_0,-2,0
47472_1,0,non
47472_2,-1,0
47472_3,-2,0
47472_4,-2,0

How are they generated? Ceiling function?

Answer 7 · 2021-04-07T02:53:02.000Z

Yes, it is.

…

On Wed, Apr 7, 2021, 14:50 Chenpuh ***@***.***> wrote: In *mosei/BA senti/label_file_*.csv*, the format is as follows: FileName,SEVEN,TWO 5Fs_7A_V2kA_0,1,1 5Fs_7A_V2kA_1,1,1 5Fs_7A_V2kA_2,1,1 5Fs_7A_V2kA_3,1,1 5Fs_7A_V2kA_4,1,1 Wfn7XvSEwUA_0,-2,0 Wfn7XvSEwUA_1,0,non Wfn7XvSEwUA_2,2,1 Wfn7XvSEwUA_3,1,1 Wfn7XvSEwUA_4,0,non Wfn7XvSEwUA_5,1,1 Wfn7XvSEwUA_6,1,1 Wfn7XvSEwUA_7,1,1 47472_0,-2,0 47472_1,0,non 47472_2,-1,0 47472_3,-2,0 47472_4,-2,0 How are they generated? Ceiling function? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA4FGUWX7ACFBEUUF4LZ4LTHPCADANCNFSM42PZJBVA> .

Answer 8 · 2021-04-07T02:59:30.000Z

Both sentiment_score or BA senti can be used in this repository, right?
Or need to modify some code to use it?

Answer 9 · 2021-04-07T03:02:18.000Z

Just the score. Later u need to ceil the scores inorder to calculate the BA. During the training we do not use any classes.

…

On Wed, Apr 7, 2021, 14:59 Chenpuh ***@***.***> wrote: Both *sentiment_score* or *BA senti* can be used in this repository, right? Or need to modify some code to use it? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA4FGVELQPJAJAXF67HY4TTHPDB3ANCNFSM42PZJBVA> .

Answer 10 · 2021-04-07T03:10:24.000Z

In other words, 7-classes and 2-classes are the same, but the evaluation criteria are different. Is 7-classes more detailed?

Answer 11 · 2021-04-07T03:13:37.000Z

Exactly. I also faced the same problem when I just started this. So I really understand how you feel :). But for some reason this is how they have evaluated. But I have seen recent work which just devices the score in to clases prior to the training. You can use that way. So think this as a regression task. Then after the training predict the score for test set. Finally use the predicted score to assign them to classes -3 to + 3.

…

On Wed, Apr 7, 2021, 15:10 Chenpuh ***@***.***> wrote: In other words, 7-classes and 2-classes are the same, but the evaluation criteria are different. Is 7-classes more detailed? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#2 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA4FGW4OP7M7IEAHINB7ITTHPEKZANCNFSM42PZJBVA> .

Answer 12 · 2021-04-07T03:15:59.000Z

Thank you very much for your help.