167 - Text prediction using LSTM (English text)

LSTMs are great for timeseries forecasting and sequence predictions. This makes them appropriate for natural language prediction. This tutorial explains the process of training an LSTM network on English text and predicting letters based on the training.

Code generated in the video can be downloaded from here:

[pranavkushare]

Thank you very much !!
The most Underrated channel on YT

[mr.n.v.subbareddy]

Wow, I have been seeing the videos on language modeling and LSTM, yours is the best explanation so far. Thank you Sir.

[pmiller]

Great video!! I have seen other tutorials with the char_to_int and int_to_char code but yours was the best explanation by far. Keep up the wonderful work. Very helpful stuff.

[WahranRai]

23:32 You dont need to recompute the softmax in your prediction, the probabilities of prediction are in softmax form ( sum (proba) =1, your output activation is softmax) just take argmax of your prediction giving you corresponding character from your dictionary

[leamon]

Thank you. Looking forward to the videos about transformer!

[varshikjain]

can you demonstrate how this model can be implement in a normal front-end text box made by flask...
like we write something and it renders some autocompleted text.

[OsscarBones]

Thank you very much! Thanks to you, I was able to train a network with discord logs to create a bot that talks like my friend.

[yhi]

Thank you a lot sir !
Your explanation is clear.
I tried the process and have a question : for the same model, why for a same seed, i have different prediction (when I run again)?

[sonalichakrabarty]

Hello sir, can you pls help in solving the error at prediction...not able to convert int to char giving errors: next_char = int_to_char[next_index] is not working

[akshathamanjunath]

Thank you,
could you please suggest which is the best approach for handwritten charcater recognition

[waleligntewabe]

Thank you very much it gives me a hint for my project. I am trying to do project work on "Automatic Question Generation using LSTM". Do you have any project work related to this?

[rvsivabalan]

Can you please advice on this..Getting error here-->
model.add(LSTM(128, input_shape=(seq_length, n_vocab)))

NotImplementedError: Cannot convert a symbolic Tensor (SampleLSTM/strided_slice:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported

[sameerasalih]

This was a great video. Can anyone guide me to a full implementation regarding data to text generation using NLG

[akshathamanjunath]

can i train the charcaters using CNN make a database and keep and in the testing phase can i input a word and segment the charcters based on the trained input charcters

[guohanzhao]

do we have tranformer network turtorial by far?

[thelastone]

Thanks a lot... You are amazing... Could you please add evaluation metrics (perplexity and BLUE score)for text generation )

[efremyohannes]

Could you pls make video on text line images recognition using lstm like recognition of line of text or words from images, and how we can prepare the dataset like ground truth and vectoring the images.

[rathnakumarv]

Sir can I try transformers for multivariate time series analyses??

[peshawriankhan]

Hi sir
Sir we are stucked in a Text generation. we train our model on 70 epoch and with batch size 150.
But when we give pre define sentence its creating just a symbols like that
";;$?ha j +#?
Gan j@($!")#?
While our coding is same as u done abmbd write theb What's wrong in it ?

[aakashverma]

plzzz make a video on UNet++ and attention UNet know no one better than u can make it so simple to understand