lessonxmk / head_fusion

17 stars 1 forks source link

Sorry to say but your code is not at all clear. #1

Open rishikeshraj5 opened 3 years ago

rishikeshraj5 commented 3 years ago

I read the readme file, in three lines t=you have described it. You should start by explaining the original dataset files, as anybody could have saved right? Where did you use the renamed files in train.py? where did you save the renamed files?

Regards

lessonxmk commented 3 years ago

line 58 in in train.py: WAV_PATH = "E:/Test/IEMOCAP/"

Please move the renamed files the this path

rishikeshraj5 commented 3 years ago

Ok sir thank you so much for this reply, we are organizing an AI workshop here in our college , certainly going to cite your work also, But some how I am not able to implement it. So I hope if this is your original work and already published you can help our team.

Also the renamed files is json file right? Sir this PAIR dict has files name and label right? and this renamed file folder is moved here in train.py file right?

Uptill now, I have changed the original files name, the PAIR dict is saved in json file format.

apart from this whatelse is required to change please guide.

Regards

On Fri, May 21, 2021 at 3:36 PM lessonxmk @.***> wrote:

line 58 in in train.py: WAV_PATH = "E:/Test/IEMOCAP/"

Please move the renamed files the this path

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-845840521, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQG6MAPE5HTKGIIFXXEDTOYWEHANCNFSM45I2APGA .

lessonxmk commented 3 years ago

I have been busy lately and may not always reply in time.

The pair.json is not important. It just helps the work of renaming. The handleIEMOCAP.py is used to rename all wav files like this formant: "Ses01F-impro01-XX-01-F000.wav" (We used the RAVDESS data set at the beginning, this format is compatible with the file name of the RAVDESS data set.)

This format is used when generating train and test data set in line 111 in train.py.

rishikeshraj5 commented 3 years ago

Sir here in lineos.rename(_, dst='name') can I change the dst location actually I don't want to change the original files name. This code is changing the files names there in my original data, Plz guide

lessonxmk commented 3 years ago

Why not back up the original files?

rishikeshraj5 commented 3 years ago

Sir thats whole dataset, but if there is no other way around I hv to do it only

On Sat, May 22, 2021 at 3:16 PM lessonxmk @.***> wrote:

Why not back up the original files?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-846383534, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQGYWA2SGCVJPZWX7FYTTO54OTANCNFSM45I2APGA .

lessonxmk commented 3 years ago

Although I think the whole data set is not very large, if this is a bit difficult for you, you can try to backup only the wav files, they are only about 1GB.

rishikeshraj5 commented 3 years ago

Okay sir, whole data is about 24.5 gb .

On Sat, May 22, 2021 at 3:27 PM lessonxmk @.***> wrote:

Although I think the whole data set is not very large, if this is a bit difficult for you, you can try to backup only the wav files, they are only about 1GB.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-846384794, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQG3ZZGBCHIYJ3IWHCYTTO55Z3ANCNFSM45I2APGA .

rishikeshraj5 commented 3 years ago

So sir, I ran this handleIEMOCAP.py file , My data files names are changed now. So in line 58 or train.py I have changed the location as PATH_WAV in ''handleIEMOCAP.py file''.......right?

Then finally training is required. But I where is these features are calculated by features.py, I think using get_mfcc, But how this file named 'features_mfcc_impro.pkl' is created at line 56.

On Sat, May 22, 2021 at 3:41 PM rishi rajput @.***> wrote:

Okay sir, whole data is about 24.5 gb .

On Sat, May 22, 2021 at 3:27 PM lessonxmk @.***> wrote:

Although I think the whole data set is not very large, if this is a bit difficult for you, you can try to backup only the wav files, they are only about 1GB.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-846384794, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQG3ZZGBCHIYJ3IWHCYTTO55Z3ANCNFSM45I2APGA .

lessonxmk commented 3 years ago

Please set featuresExist on line 54 to False. See line 205, when featuresExist is false, the feature will be extracted and saved as .pkl file

rishikeshraj5 commented 3 years ago

Sir plz help I am getting this trackback of error

0%| | 6/8030 [00:00<02:35, 51.60it/s]constructing meta dictionary for E:/dataset/data/IEMOCAP_full_release//sentences/wav//S*.wav... 11%|█ | 848/8030 [00:08<01:07, 105.75it/s] Traceback (most recent call last):

File "E:\Python_On_All_Dataset\IEMO\long_codes\multiscale_att_impro_only\head_fusion-master\train.py", line 215, in train_X, train_y, val_dict = process_data(WAV_PATH, t=T_stride, train_overlap=T_overlop)

File "E:\Python_On_All_Dataset\IEMO\long_codes\multiscale_att_impro_only\head_fusion-master\train.py", line 113, in process_data label = str(os.path.basename(wav_file).split('-')[2])

IndexError: list index out of range

Sir this is working then y it stopped in between? plz guide

On Sat, May 22, 2021 at 11:10 PM rishi rajput @.***> wrote:

Sir its working when I checked line by line for single wave file then why this error any idea sir?

On Sat, May 22, 2021 at 10:45 PM rishi rajput < @.***> wrote:

Yes sir did that already, Actually I did read the code carefully, But I don't know why its saying

IndexError: list index out of range at line

label = str(os.path.basename(wav_file).split('-')[2])

does it mean wav file name doesn't have index [2] . file file path is

'E:/dataset/data/IEMOCAP_full_release\Session1\sentences\wav\Ses01F_impro01\Ses01F-impro01-01-01-F001.wav'

then basefile name is Ses01F-impro01-01-01-F001.wav right sir? Then why such error. Is it related to next lines in code coz when I ran it with same error line by line its showing label ='01'

STUCK UP I have tomorrow in my hand only to train this model sir.

On Sat, May 22, 2021 at 5:44 PM lessonxmk @.***> wrote:

Please set featuresExist on line 54 to False. See line 205, when featuresExist is false, the feature will be extracted and saved as .pkl file

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-846399725, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQG2CCKMUDUNVMAWXURLTO6NZRANCNFSM45I2APGA .

lessonxmk commented 3 years ago

Please check wav file names. Whether there are some files renamed incorrectly and the file name can not be split by '-' ?

rishikeshraj5 commented 3 years ago

No sir all files names are absolutely correct with '-' and every third segment label like 01,02,,09. Sir plz brief me about your data arrangement I think there is some problem. In my original dataset the path for wav files is : "E:/dataset/data/IEMOCAP_full_release//sentences/wav//S.wav" using handleIEMOCAP.py i have changed all files names in this folder only. Then I replaced the path in line 58 as WAV_PATH = "E:/dataset/data/IEMOCAP_full_release//sentences/wav//S.wav"

is there any mistake sir?

In your readme file here in second step you are saying move renamed files to dict, what does it mean sir?

1.run handleIEMOCAP to rename files in IEMOCAP corpus.

2.move the renamed files to a dictionary (e.g. '/data/.wav').*

3.run train.py to train a model and evaluate.

On Tue, May 25, 2021 at 6:31 PM lessonxmk @.***> wrote:

Please check wav file names. Whether there are some files renamed incorrectly and the file name can not be split by '-' ?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-847849234, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQGZFJALYUEFLCP4C7HLTPONT3ANCNFSM45I2APGA .

lessonxmk commented 3 years ago

The second step means moving all the renamed files to an independent folder. I am not sure if it is correct to use regular expressions in WAV_PATH.

And WAV_PATH should be folder name instead of file name. For example, there are some files (.wav) in 'E:/dataset/' , WAV_PATH should be 'E:/dataset/' instead of 'E:/dataset/.wav'

Finally, you can use debug mode in your IDE to find out which file it is when this error occurred.

rishikeshraj5 commented 3 years ago

I think the error is in the folder path itself, as you are saying all renamed files in a single folder (all 10000 renamed files in a single folder). so there should not be any subfolders?? right? But sir that's really tough as, there are almost 30 files inside each session file and then inside each such files 50 to 100 wav files, so unfolding each of these almost 10000 files that's really time-consuming.

and yes that's folder name only as I have changed the third line in process_data definition to

wav_files = glob.glob(path) instead of file extension.

Plz reply so that I can go with the unfolding of these 10k files in a single folder sir.

rishikeshraj5 commented 3 years ago

sir could you plz tell me if that can work in this situation, what should be the modification in this code then. plz

On Tue, May 25, 2021 at 8:08 PM rishi rajput @.***> wrote:

I think the error is in the folder path itself, as you are saying all renamed files in a single folder. so there should not be any subfolders right?

and yes that's folder name only as I have changed the third line in process_data definition to

wav_files = glob.glob(path) instead of file extension.

On Tue, May 25, 2021 at 7:22 PM lessonxmk @.***> wrote:

The second step means moving all the renamed files to an independent folder. I am not sure if it is correct to use regular expressions in WAV_PATH.

And WAV_PATH should be folder name instead of file name. For example, there are some files (.wav) in 'E:/dataset/' , WAV_PATH should be 'E:/dataset/' instead of 'E:/dataset/.wav'

Finally, you can use debug mode in your IDE to find out which file it is when this error occurred.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-847888719, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQG77PC434Z3Q2BVXKJLTPOTRLANCNFSM45I2APGA .

rishikeshraj5 commented 3 years ago

Not able to make it sir! well could you plz tell me what is the acc you are getting from this model? are you actually getting 76% from this model? or there are some other hyperparameters adjustments?

lessonxmk commented 3 years ago

Since the data set is randomly divided, the results also have some randomness. The accuracy may be a bit lower or higher, just like this figure. 76% is the average accuracy of several experiments. QQ截图20210526200601

rishikeshraj5 commented 3 years ago

Thank you for the clarification. So here acc is the weighted acc at 50th epoch not average of all 50 epochs right? and you are considering acc after 50th epoch, I think it would be better to see this in an loss accuracy graph for train and val data. and sir this is improvised data only for four emotional classes right ie 2943 utterances as in paper you merged happy and exc or some other combination?

On Wed, May 26, 2021 at 5:40 PM lessonxmk @.***> wrote:

Since the data set is randomly divided, the results also have some randomness. The accuracy may be a bit lower or higher, just like this figure. 76% is the average accuracy of several experiments. [image: QQ截图20210526200601] https://user-images.githubusercontent.com/44159619/119657203-32000f00-be5e-11eb-87ff-7e61a124adcc.png

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-848716120, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQG7W5EY35DMWI3VZ56DTPTQLLANCNFSM45I2APGA .

rishikeshraj5 commented 3 years ago

Sir in this screen shot , The maxUA and maxWA should be the final result I think, so in this case 78% NO? Plz correct me If m wrong? Also for three versions of datasets impro , script and all do we need to change anything else in code, because while enumerating the train and val files there are some conditions in dark fonts :

for i, wav_file in enumerate((valid_files)): label = str(os.path.basename(wav_file).split('-')[2]) if (label not in LABEL_DICT1): continue

On Wed, May 26, 2021 at 6:53 PM rishi rajput @.***> wrote:

Thank you for the clarification. So here acc is the weighted acc at 50th epoch not average of all 50 epochs right? and you are considering acc after 50th epoch, I think it would be better to see this in an loss accuracy graph for train and val data. and sir this is improvised data only for four emotional classes right ie 2943 utterances as in paper you merged happy and exc or some other combination?

On Wed, May 26, 2021 at 5:40 PM lessonxmk @.***> wrote:

Since the data set is randomly divided, the results also have some randomness. The accuracy may be a bit lower or higher, just like this figure. 76% is the average accuracy of several experiments. [image: QQ截图20210526200601] https://user-images.githubusercontent.com/44159619/119657203-32000f00-be5e-11eb-87ff-7e61a124adcc.png

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/lessonxmk/head_fusion/issues/1#issuecomment-848716120, or unsubscribe https://github.com/notifications/unsubscribe-auth/AUCSQG7W5EY35DMWI3VZ56DTPTQLLANCNFSM45I2APGA .

XYQ159371 commented 2 years ago

Hello, I am very sorry to disturb you. Could you please explain the source of the noise set? I would like to use it. If it is convenient for you, plz send my email. (12shq12@163.com) I would appreciate it.

Coding511 commented 2 years ago

@XYQ159371 you can download these noise files from here dierctly.