IEMOCAP datasets question
Opened this issue · 0 comments
Thank you for sharing your work !
I'm confused about the data read from dataset IEMOCAP.pkl
` iemocap_data = pickle.load(open(data_path + "iemocap.pkl", 'rb'), encoding='bytes')
print(np.sum(iemocap_data[b'happy'][TRAIN][TEXT]))
print("------------------------------------")
print(np.sum(iemocap_data[b'sad'][TRAIN][TEXT]))
print("------------------------------------")
print(np.sum(iemocap_data[b'angry'][TRAIN][TEXT]))
print("------------------------------------")
print(np.sum(iemocap_data[b'neutral'][TRAIN][TEXT]))
`
I use the above "print" to show the vector value in four classes, but I get the same result like the following:
`
Temp location for models: models/model__b'angry'.pt
Grid search results are in: results/results__b'angry'.csv
-7770.0356465404475
-7770.0356465404475
-7770.0356465404475
-7770.0356465404475
Audio feature dimension is: 74
Visual feature dimension is: 35
Text feature dimension is: 300
`
Meanwhile, I try another keys "VIDEO" and "AUDIO", the four classes value are still same.
No matter how I choose the emotion value, the dataset is the same.
The reason I find this is that I want to know how many data in each emotion type,
but I get the same length of "train_set" when I switch the parameter "emotion".
And may I ask why don't you train with all emotions together
Thanks