search

kushagra2101 / ChatCrazie

A chatbot implemented using RNN and GloVe embeddings whch answers your query crazily
12 stars 13 forks source link

There is a bug, please replace with this code. Thanks. #6

Open AdroitAnandAI opened 5 years ago

AdroitAnandAI commented 5 years ago

There is a bug in line 70 of test_seq2seq.py file. The weights you save after training and the weights you try to load for testing are different.

Please replace with this line: self.model.load_weights('support files/model-weights.h5')

Thanks a lot for the awesome project.

abhishek305 commented 5 years ago

@AdroitAnandAI We had some kind of dimensions issue while running the code .How were you able to run .A short description would be helpful .

AdroitAnandAI commented 5 years ago

While running the test function, shape mismatch error would be thrown, as the shape of weight matrix we are trying to load doesnt correspond to the shape of the model built in train_seq2seq.py.

In train_seq2seq.py just replace this line: self.model.load_weights('support files/word-glove-weights.h5') with: self.model.load_weights('support files/model-weights.h5')

Now run the GUI_Tkinter.py. It will work. Hope this helps.

abhishek305 commented 5 years ago

While running the test function, shape mismatch error would be thrown, as the shape of weight matrix we are trying to load doesnt correspond to the shape of the model built in train_seq2seq.py.

In train_seq2seq.py just replace this line: self.model.load_weights('support files/word-glove-weights.h5') with: self.model.load_weights('support files/model-weights.h5')

Now run the GUI_Tkinter.py. It will work. Hope this helps.

Thanks !! It really helped .

abhishek305 commented 5 years ago

While running the test function, shape mismatch error would be thrown, as the shape of weight matrix we are trying to load doesnt correspond to the shape of the model built in train_seq2seq.py.

In train_seq2seq.py just replace this line: self.model.load_weights('support files/word-glove-weights.h5') with: self.model.load_weights('support files/model-weights.h5')

Now run the GUI_Tkinter.py. It will work. Hope this helps.

@AdroitAnandAI Sir sorry to bother you but every thing is working fine and we even trained the bot at the given hyperparatmeters (as given in the description) we are not getting a desired output as shown in there video .we are struggling to get an optimal output.It would be really helpful sir if you guide as through. Is there any other implementation available on github or somewhere else for chatbots using Deep Learning. As we are new to this .Thank You!! Sir this is our output but the answers are not up to the mark while comparing there and our output. Do you need to change something or increase the training hyperparameters

whatsapp image 2018-12-27 at 5 11 46 pm

AdroitAnandAI commented 5 years ago

I would suggest you to use attention along with LSTMs or bidirectional RNNs. The theory and code of an optimal implementation of chatbot can be found in my blog below. If you have any questions, plz post in the blog comments coz i don't frequent here. Thanks a lot.

https://medium.com/datadriveninvestor/how-to-make-digital-version-of-you-2a29cf823e85 https://github.com/AdroitAnandAI/LSTM-Attention-based-Generative-Chat-bot

AdroitAnandAI commented 5 years ago

@abhishek305

abhishek305 commented 5 years ago

I would suggest you to use attention along with LSTMs or bidirectional RNNs. The theory and code of an optimal implementation of chatbot can be found in my blog below. If you have any questions, plz post in the blog comments coz i don't frequent here. Thanks a lot.

https://medium.com/datadriveninvestor/how-to-make-digital-version-of-you-2a29cf823e85 https://github.com/AdroitAnandAI/LSTM-Attention-based-Generative-Chat-bot

@AdroitAnandAI Thanks you again . It really helped and your blog is really well documented.