jatinmandav/neural-networks

All about Neural Networks!

neural-networks's Issues

Is it a typo to set tweets to empty array?

In word2vec and fastText, tweets is set to "[]" just before "lemmatizing". It seems the effect of "tokenizing" will be lost. Is it a typo? Thanks.

Files for the training of the USE

Hello,

I hope that you are doing well.

Where Can I find the pos_1.2M.txt and neg_1.2M.txt files?

Thanks in advance

test accuracy

Nice job man, I read all your code and I realized that the test accuracy for word2vec and fasttext is higher than universal sentence encoder so they are much better than universal sentence encoder with this architecture of the model. so I think you should report the test accuracy for each model.
best regards

The U.S.E transformer architecture gives a kernel died error!

Hi I wouldn't say this is exactly an issue.
I tried your method of using the USE for a CNN model for my business data and instead of using DAN, I used the Transformer architecture,
I use the GCP - GPU Tesla K 80 to embed data and train the model.

I keep on getting the 'Kernel died and will start automatically error' for all my combination of runs.

I have reduced the data to the minimum, the batch size, filter size everything, it still gives the same error.

Thought the problem with my model, so I replaced it with yours, and still I get the same error.

If you could help me through this, that would be great!

Thanks
Arav