train result question

[tianyexing]

when i trian the model ,i use 200000W word ,batchsize = 128 ,adam = 0.00001, epoch = 100 , the result as nonlister
nonlaterzzzzzzzz neep miepzzzzzzzz i don't know why ? the result is not good ?

[Belval]

Hi,

You parameters seem good.

1. Did you alter the code in anyway?
2. What did you use for training and how long did it take?
3. Please post an example of your training data.
4. If you can also provide the trained model so I can test it on my side. That would be great.

I have never seen the zzzzzz thing so it's puzzling to me that you got that.

[tianyexing]

I am so happy to see you reply.

1. I change the config ,i use lower case letters. and i ues some depth_wise convolution instead of convolution in cnn part
2. I use 1060 training and use about 2 day.
3. The traing data same as yours but i removed the words with numbers.
4. I am willing to give you the code, I don't know how to operate it.

[Belval]

I'd be very interested to see your modifications.

In the meantime, I must confess that I am a bit tight on free time. I will try to replicate your bug this weekend, but until then I'm afraid I won't be of much help.

Regards

[tianyexing]

thank you for your reply. I am also trying to improve this model. I found that too much deep convolution will cause this problem. I now have an acceptable result.