Encoder And Decoder- Neural Machine Learning Language Translation Tutorial With Keras- Deep Learning

Поделиться
HTML-код
  • Опубликовано: 5 янв 2025

Комментарии • 114

  • @thetensordude
    @thetensordude 4 года назад +17

    YOLO, BERT, TRANSFORMERS!! Please bring explanations on these

  • @MASadat-lz9yz
    @MASadat-lz9yz Год назад +1

    Greetings from Austria, thanks for your knowledge sharing!

  • @anjanas6048
    @anjanas6048 4 года назад +3

    Hi Krish..Thanks for the video...waiting for more topics like attention mechanism, transformers etc

  • @asim-turivlogs
    @asim-turivlogs 9 месяцев назад

    I will suggest you watch the output context of LSTM. The output context of LSTM would be not a single vector it would be multiple vector generated at each time step. This is were attention would be implemented and differentiate from the RNN. I appriciate your tutorials

  • @tyylermike2830
    @tyylermike2830 4 года назад +10

    Greetings!! Can you please upload more stuff on Deep learning like attention models, Transformers, BERT and do cover unsupervised learning too if possible. It would be highly appreciated.

    • @darianjudah4380
      @darianjudah4380 3 года назад

      A trick : watch series on Flixzone. I've been using them for watching all kinds of movies lately.

    • @makaistanley6490
      @makaistanley6490 3 года назад

      @Darian Judah Yea, have been using flixzone} for months myself :D

  • @shashankpal376
    @shashankpal376 3 года назад +8

    What is the function of dense layer after decoder? Aren't we actually interested in decoder output? Why adding dense layer would not hamper the actual output by decoder? I would be very thankful if someone answers all my questions.

  • @8889705048
    @8889705048 3 года назад +4

    Can you please take a small sample text and perform encoding and decoding functioning briefly, so that we can understand it briefly as there are few doubts regarding t=timestamp.

  • @rajak7410
    @rajak7410 4 года назад +4

    Sir take a sample text and perform encoding and decoding functioning briefly, so that it makes sense

  • @mukulsharma9673
    @mukulsharma9673 4 года назад +1

    Thanks krish for this video... I was waiting for this .!!

    • @hasiburrahman96
      @hasiburrahman96 4 года назад

      Can i use this, for question answering instead of language translation??

  • @DeependraSingh-jh8xf
    @DeependraSingh-jh8xf Год назад +1

    i dont understand why my final testing loop is decoding every input to ' i want to go to room'
    i have made hindi to english translation and used the dataset of the blog which was just shown below english french database

  • @Kumar08
    @Kumar08 4 года назад +9

    Please provide Github link for this code.

  • @mukulbhardwaj5390
    @mukulbhardwaj5390 7 месяцев назад

    Hello any one please tell me this model does not accurate i tune with various hyper parameter but the accuracy is not good. Can someone tell me what to do exactly to achieve so?

  • @DeependraSingh-jh8xf
    @DeependraSingh-jh8xf Год назад +1

    i have followed every step. still my encoder_input_data is all same, for each sentence, please help

  • @naivedhshah2980
    @naivedhshah2980 4 года назад +5

    Can you share the notebook?

  • @Rajkumar-vc2pg
    @Rajkumar-vc2pg 3 года назад +3

    okay now ill try to train english-telugu and add it on my resume

  • @HARIS-q3n
    @HARIS-q3n 2 месяца назад

    Sir, I have a doubt,
    During the validation, what if the decoder outputs an output sequence shorter than the true output sequence. How is the loss calculated in such cases. will categorical cross entropy work in such cases.

  • @narendraparmar1631
    @narendraparmar1631 3 года назад

    Thanks Krish

  • @willdidplays
    @willdidplays 3 года назад +2

    excuse me sir, i wonder why use one hot encoder method instead word embedding layers ?

    • @naveenpoliasetty954
      @naveenpoliasetty954 11 месяцев назад

      we can use them but we should be precise about the dimensions

  • @lol-ki5pd
    @lol-ki5pd 8 месяцев назад +1

    why suddenly we are talking about characters when in previous vedios you mentioned of doing one-hot encoding of words?

  • @ankurlimbashia3697
    @ankurlimbashia3697 4 года назад +1

    Hey krish, I am getting some cardinality in inference model input with this code
    Model fits perfectly but while predicting the inference model I am getting this error.

  • @fatimaezzahrakharmouch7700
    @fatimaezzahrakharmouch7700 Год назад +2

    Hello Sir, thank you for this video. I did all the steps but in the end I don't receive the correct output, can you help me please. thanks

  • @manjeet_manu
    @manjeet_manu 3 года назад +2

    In this video only predefine values are converting to their respective meaning. I don't understand how this helps us to covert user input into other language.

  • @nawazalilone701
    @nawazalilone701 2 года назад

    What is the encoding scheme used in this tutorial e.g one hot, word2vec, glove etc

  • @MuhammadAli-ie7ps
    @MuhammadAli-ie7ps 3 года назад +1

    hey man, kindly give the link of code. if you did not upload there. please upload code there. it's a request.

  • @abhijittdhavlle
    @abhijittdhavlle Год назад

    @krishnaik06 This is super helpful video. I have been following the NLP playlist. Do you mind sharing this code in the Git repo? The folder for seq2seq in your repo seems empty. thanks:)

  • @davidhenry3553
    @davidhenry3553 3 года назад

    Hey Krish, I think you forgot to upload or provide a link to your notebook.Can you please upload it.

  • @anirbanmukherjee3028
    @anirbanmukherjee3028 Год назад

    Sir, unable to locate github code...can you please share?

  • @murari-goswami
    @murari-goswami 4 года назад

    Hi @krish, Can you tell how you create this new format. I mean which application you use. This looks so cool. I also want to record videos, but not sure which software can give me these features. Any insight ?

  • @deepaklonare9497
    @deepaklonare9497 Год назад +1

    can you please share the github link for above code

  • @madtech1803
    @madtech1803 4 года назад +1

    Please make videos on semantic segmentation

  • @piyalikarmakar5873
    @piyalikarmakar5873 2 года назад +1

    Sir, thanks for this nice explanations. But I have one query, Instead of text I have numeric indexes of the text, but these are not vectors. How can I translate those indexes into its corresponding text?

  • @SelinTosun
    @SelinTosun 4 года назад

    Is the Attention model posted? I have been waiting since this video was posted. Looking forward to it...

  • @tanish1018
    @tanish1018 Год назад

    Sir can you plz add the Jupyter notebook respective to the model

  • @joguns8257
    @joguns8257 Год назад

    Where is the dataset? How can I get it?

  • @thelife5628
    @thelife5628 Год назад

    🔥🔥🔥🔥🔥🔥🔥🔥

  • @nishantsharma2022
    @nishantsharma2022 4 года назад +2

    First Like

  • @babritbehera4087
    @babritbehera4087 4 года назад +10

    U skipped the main part ... Decoder input n decoder output . That how we r dividing the target as input n output with 1 timestep ..

    • @arkadeepdas7397
      @arkadeepdas7397 4 года назад

      can you explain "encoder_input_data[i, t+1:, input_token_index[' ']] = 1." this line? why we should use?

  • @siddharthaborpuzari9000
    @siddharthaborpuzari9000 4 года назад

    Thank you

  • @kaustubh52q45
    @kaustubh52q45 Год назад

    please provide your github repo where you're code is present, please provide the code

  • @DEWAAN-ud7ss
    @DEWAAN-ud7ss 11 месяцев назад

    krish kindly upload the image captions projects ..

  • @athiragopalakrishnan4316
    @athiragopalakrishnan4316 3 года назад

    Dear sir, I would like to join your channel. I tried to contact you (FB/LinkedIn). But the site can't be reached. How can I contact you, sir? Please help me

  • @iffatmehrindisha5517
    @iffatmehrindisha5517 8 месяцев назад +1

    Can you please share the code?

  • @sauravjha5666
    @sauravjha5666 4 года назад +16

    bro this video doesn't make any sense, sorry to say but it's not at all intuitive.

    • @ridj41
      @ridj41 Год назад

      if basics aren't clear then yes it isn't

    • @sauravjha5666
      @sauravjha5666 Год назад

      It’s a three year old comment 😂😂

  • @cinebuffhometheatres
    @cinebuffhometheatres 4 года назад +1

    i have a simple doubt is this character encoding or word encoding

  • @rabiaaiqbalmalik5208
    @rabiaaiqbalmalik5208 3 года назад

    Can you help for english-urdu machine translation model?

  • @gurdeepsinghbhatia2875
    @gurdeepsinghbhatia2875 4 года назад

    Sir i dont understand the input dimension , that how input dimension is like that

  • @ponrajs5396
    @ponrajs5396 4 года назад +1

    I've been waiting long time for this video,
    I have one doubt, in case if need to update the neural network with some extra stacked LSTM layer how could I do that, because this is not like other sequential model.

    • @babritbehera4087
      @babritbehera4087 4 года назад +1

      U need to use functional api for model creation

    • @ponrajs5396
      @ponrajs5396 4 года назад

      @@babritbehera4087 sorry what is functional api. I don't know.

    • @babritbehera4087
      @babritbehera4087 4 года назад

      Juzz Google ... Model building with functional api

    • @hasiburrahman96
      @hasiburrahman96 4 года назад

      @@babritbehera4087 Can i use this, for question answering instead of language translation??

  • @mohsala5498
    @mohsala5498 Год назад

    honestly, I watched many great videos for you, but this one is very poor and lacks explanation and clarity... I suggest in future you remake this session again. Any way thanks a lot for your efforts.

  • @drashyabansal7749
    @drashyabansal7749 4 года назад

    Hello Sir! I tried computing the code but its crashing when working on google colab and in case of initialising the zero matrices system is crashing

    • @alihaiderabdi9939
      @alihaiderabdi9939 3 года назад

      probably because of more data and less ram, try fixing with inputs take less samples of dataset around 10k i took in my 8gb ram.

  • @apica1234
    @apica1234 3 года назад

    does this apply to time series as well?

  • @gurdeepsinghbhatia2875
    @gurdeepsinghbhatia2875 4 года назад

    Sir why we not given the inputs like this : encoder_inputs=Input(shape=(max_encoder_sequence_length,num_encoder_tokens))

  • @soumyagupta9301
    @soumyagupta9301 3 года назад

    Are encoder_outputs and h_t not the same thing?

  • @ravinderbadishagandu2647
    @ravinderbadishagandu2647 Год назад

    can you please provide notebook link

  • @sairajdas6692
    @sairajdas6692 Год назад +1

    Where is the code ?

  • @Manojrohtela
    @Manojrohtela 3 года назад

    Where i can get this code ?

  • @sagarnarula660
    @sagarnarula660 4 года назад +2

    Please provide the github link for the code.

  • @gurdeepsinghbhatia2875
    @gurdeepsinghbhatia2875 4 года назад

    sir its compulsary to give input as one hot encoder ,, may we use the word embedding

    • @hasiburrahman96
      @hasiburrahman96 4 года назад

      Can i use this, for question answering instead of language translation??

    • @AvinashSingh-bk8kg
      @AvinashSingh-bk8kg 3 года назад +1

      Yes we can use embedding too.

  • @alphonseinbaraj7602
    @alphonseinbaraj7602 4 года назад

    decoder_input-data and decoder_target_data will be same ?

    • @rajeevu3051
      @rajeevu3051 4 года назад

      Yes, both will be same vector representation.

  • @akshayabusa007
    @akshayabusa007 4 года назад

    for line in lines[: min(num_samples,len(lines)-1] why are we doing -1 can anybody explain and why are we taking min ?

    • @milindsai5254
      @milindsai5254 3 года назад

      Even I couldn't understand the -1 part, logically it makes no difference, coming to the min() part, it is just used to choose the min value between num_samples,i.e,10k as mentioned in the code and len(lines),i.e, approx 185k. So, min(10k,185k) returns 10k and the first 10k records are considered for training the model. The underlying purpose is faster computation as computing 10k samples would take less time than in comparison to 185k samples.

    • @Manojrohtela
      @Manojrohtela 3 года назад

      I think we iterating for loop for lines so if num_samples are minimum then we don't need -1 but if len(lines) are minimum then we have to iterate the for loop till -1 because it starts from 0 and without -1 it will give out of bound error

  • @mohan5867
    @mohan5867 4 года назад

    Sirni have written the code exactly the same... but my accuracy is just 0.0020.... what could be the reason.?

    • @hasiburrahman96
      @hasiburrahman96 4 года назад

      Can you please share this written code which you Did ?

    • @AvinashSingh-bk8kg
      @AvinashSingh-bk8kg 3 года назад

      @Chaitanya Kaul
      Were you able to solve the issue?

  • @hasiburrahman96
    @hasiburrahman96 4 года назад

    Can i use this, for question answering instead of language translation??

    • @beingraja1212
      @beingraja1212 4 года назад

      No as questions and answers are different in every sense

    • @sayantanmazumdar9371
      @sayantanmazumdar9371 3 года назад

      Most likely a transformer network will satisfy your needs

  • @user-or7ji5hv8y
    @user-or7ji5hv8y 4 года назад

    may be better audio would be really helpful.

  • @TheShaan22
    @TheShaan22 4 года назад

    Can you explain the inference code please, thanks.

  • @arkadeepdas7397
    @arkadeepdas7397 4 года назад

    can any one explain "encoder_input_data[i, t+1:, input_token_index[' ']] = 1." this line? why we should use?

    • @ShodaiThox
      @ShodaiThox 3 года назад

      probably too late but,

    • @prathameshjadhav780
      @prathameshjadhav780 3 года назад

      @@ShodaiThox hey did any of guys understand why he did that

  • @sunnybhojwani3199
    @sunnybhojwani3199 4 года назад

    Kindly upload more videos

    • @hasiburrahman96
      @hasiburrahman96 4 года назад

      Is it wotk for question answering also??

  • @no_num_no_gum
    @no_num_no_gum Год назад

    great video but u are mistaking the term character by word which creates confusion

  • @MurariMahaseth
    @MurariMahaseth Год назад

    17:36

  • @madhu1987ful
    @madhu1987ful 3 года назад

    Hi Krish,
    2 questions:
    1. why are we not using word2vec here and instead using OH encoding?
    2. when u say A,B,C as inputs to encoders -- are these characters or words? I am confused

  • @piyushmajgawali1611
    @piyushmajgawali1611 4 года назад +1

    You are using *word* for characters

  • @koustavdutta5317
    @koustavdutta5317 4 года назад

    sir please provide your notebook ... it will be really helpful, the notebook which you have prepared

    • @hasiburrahman96
      @hasiburrahman96 4 года назад

      Can i use this, for question answering instead of language translation??

    • @funkupgamer5364
      @funkupgamer5364 3 года назад

      @@hasiburrahman96 yes

    • @sayantanmazumdar9371
      @sayantanmazumdar9371 3 года назад

      @@hasiburrahman96 Most people use transformers however encoder decoder can also do the job

  • @rahulrajpandey8232
    @rahulrajpandey8232 3 года назад

    Github link??

  • @pratikshaunale4885
    @pratikshaunale4885 3 года назад

    in latent dimensionality how you define 256 plz explain

    • @naveenpoliasetty954
      @naveenpoliasetty954 11 месяцев назад

      he took 256 randomly its upon you try taking different as they just reflect as time steps in an LSTM

  • @soumyagupta9301
    @soumyagupta9301 3 года назад

    Thanks for the video but you are not going deep into the code or the architecture. Please try to go a bit deeper.

  • @pvbsuresh4156
    @pvbsuresh4156 3 года назад

    Loss is very high

  • @Mohankumardash
    @Mohankumardash 4 года назад

    jeery

  • @Mohankumardash
    @Mohankumardash 4 года назад

    jerry

  • @N-fx9uz
    @N-fx9uz Год назад

    Seemed like you yourself were not understanding the code you have written. Zero logic building and only reading out the code. Very upsetting 😑.

  • @bluemonkify
    @bluemonkify 3 года назад

    Az yavaş anlat yiğido anlayamıyok!