ajinkyat / cnn_intent_classification Goto Github PK

i will give you a shoutout next week june 1 great work

Thanks for sharing the code.
I was trying to do intent classification using phoneme sequence instead of the word-based approach.
Instead of word embeddings from glove I was using integer phoneme indexes. By somehow my model is performing overfitting. I have tried regularizers to prevent but it couldn't help.
Do you have any idea how to solve this problem?

Train set: (21265, 100) (21265, 7)
validation set: (2362, 100) (2362, 7)

Epoch 1/20
21265/21265 [==============================] - 16s 732us/step - loss: 0.0954 - acc: 0.9876 - val_loss: 4.4168 - val_acc: 0.5881
Epoch 2/20
21265/21265 [==============================] - 16s 733us/step - loss: 0.0854 - acc: 0.9903 - val_loss: 4.4084 - val_acc: 0.5881
Epoch 3/20
21265/21265 [==============================] - 16s 731us/step - loss: 0.0854 - acc: 0.9895 - val_loss: 4.6018 - val_acc: 0.5910
Epoch 4/20
21265/21265 [==============================] - 15s 718us/step - loss: 0.0867 - acc: 0.9896 - val_loss: 4.4348 - val_acc: 0.5893
Epoch 5/20
21265/21265 [==============================] - 15s 722us/step - loss: 0.0815 - acc: 0.9902 - val_loss: 4.6274 - val_acc: 0.5876
Epoch 6/20
21265/21265 [==============================] - 16s 734us/step - loss: 0.0797 - acc: 0.9905 - val_loss: 4.8017 - val_acc: 0.5876
Epoch 7/20
21265/21265 [==============================] - 16s 750us/step - loss: 0.0823 - acc: 0.9901 - val_loss: 4.7837 - val_acc: 0.5881
Epoch 8/20
21265/21265 [==============================] - 16s 729us/step - loss: 0.0769 - acc: 0.9904 - val_loss: 5.0681 - val_acc: 0.5876
Epoch 9/20
21265/21265 [==============================] - 15s 729us/step - loss: 0.0769 - acc: 0.9909 - val_loss: 4.7879 - val_acc: 0.5881
Epoch 10/20
21265/21265 [==============================] - 16s 760us/step - loss: 0.0763 - acc: 0.9905 - val_loss: 4.7429 - val_acc: 0.5881

Hi
thanks for sharing your code!
would u mind to upload "glove.6B.100d.txt" file pls?
I would like to check this code my self but I got stuck with this line:
f = open(os.path.join(GLOVE_DIR, 'glove.6B.100d.txt'), encoding='utf-8')
thank u!