Implementing Word Embedding Using Keras- NLP | Deep Learning

[@krishnaik06]

Steps To Follow 1. Sentences 2. One hot Representation-- index from the dic 3. Onhot Repre---> Embeddind Layer Keras To form Embedding matrix 4. Embedding matrix

[@vinitkanse5703]

How to decide the vocab_size ??

[@BhupinderSingh-rg9be]

sir please include this part in which u write steps in notepad in every video, it feels easy to understand the code.

[@mathavraj9378]

is the embedding layer values here formed by word2vec ?

[@tararawat2955]

how dimensions are decided for textual data...in your case what all features we have considered for each word..i mean on what basis we can decide dimension....

[@DineshBabu-gn8cm]

​@@vinitkanse5703 please explain me what is vocabulary.

[@ijeffking]

I cannot thank you enough for this particular video. The length to which you have gone to explain Word Embeddings is highly appreciated. A world of Thanks.

[@BrandoMalqui]

Thank you so much. I was having trouble understanding embedding which I need to implement for a model in one of my classes but you have made it very clear and easy to understand.

[@arpitbaranwal7236]

Thanks Krish for wonderful playlist.

[@Gamezone-kq5sx]

The best than applied ai... Really best video ...

[@googlecolab9141]

better explanation than Stanford CS224N: NLP with Deep Learning | Winter 2019 course. thank you sir

[@spartacuspolok6624]

Your video helped me a lot to understand it and to start working as a beginner.

[@debashisghosh3133]

Awesome explanation Krish hats off...thnx a ton

[@shaiksuleman3191]

You and Codebasics are 2 eyes in teaching.There are some many doctors ,only few will inject injection with out pain.

[@gauravsahani2499]

Thankyou so much Krish Sir, for this wonderful Playlist!, Learned a Lot!

[@vivekbhat720]

Thankssss bro for putting such great effort in teaching.

[@vibivij1948]

Nice Video Krish...

[@sir-lordwiafe9928]

Thank you Sir. Very very helpful.

[@affandibenardi548]

it make sense and simple to understand thx bro

[@enricowijaya8668]

AMAZING EXPLANATION, thank youu!!

[@bhushanchaudhari378]

Mind blowing explanation 🤙

[@louerleseigneur4532]

Thanks Krish

[@radhikapatil8003]

Hi sir please suggest the face recognition CNN model..which is comparable with mobile face recognition

[@maralazizi]

Thank you, it was a great explanation!

[@praneethaluru2601]

Literally my doubt got clarified...

[@arjitdabral8449]

Sir plz explain 1.) what were the features here according to which the vectors were created 2.) Where does the features come from can we use our own features

[@nobalg3482]

@Krish Naik, would love to have an answer to this..... Please can you explain this as well.

[@tararawat2955]

even I have the same question..Kindly answer @Krish

[@hang1445]

@Krish Naik I would like to ask the same question

[@RitikSingh-ub7kc]

Krish, can you explain some applications of nlp using lstm like next word prediction, translation and Image captioning ?

[@kmnm9463]

Hi Krish, Great video on Word Embedding. At 03:10 I think it is not one_hot which passes the values [2000, 4000, 5500] to Embedding layer. This is done by Tokenizer class from the same Keras. Tokenizer will create a dictionary of Words and their integer value - the length of the dictionary will be equal to total unique words in the corpus. Also one_hot is not efficient and after Tokenizer has come into the scene, one_hot is seldom used. from user - Krish ( my name shortened too is Krish :)

[@sampaths4932]

Good

[@shamussim137]

This is so goodd!!!Thanks Krish

[@rishikeshthakur6497]

Your videos are really great sir. Hat's off you. Please, also make video on sentence embedding technique like infersent

[@ramonolivier57]

You speak a little too fast in some section, but you explain everything very well. I still am missing understanding on "global average pooling". Thank you!

[@EliRahimkhani]

Great video! a couple of questions: how we can see these 10 features for each word? are the features the same? if not how the features are selected?

[@sowmyakavali2670]

Krish , How the word embedding layer convert one hot encoding vector to fixed dimensional embedding matrix ? How the 10 k values converted into 300 values? Here we assume that one hot vector means its a sparse vector only , and is the embedding matrix also s sparse matrix or else dense matrix ? @KrishNaik

[@User-nq9ee]

Hi very nice explanation, but we did not mention any feature but only feature size then how all those words got assigned dimension values. on what basis.

[@infinitum90]

Krish, can u explain how index 6654 get converted to vector of 10 dimension vector, exactly what algorithm keras used to convert an index into vector .

[@bhavinmoriya9216]

THanks for the awesome video. Don't we need to fit before predict? If not, why?

[@fidaullah3775]

Thanks for sharing video, but please also make video on LSTM

[@lemoniall6553]

if we have a sentence "vishy eat bread". then we vectorize the word "eaat"(misspelled word), why does fasttext see that the word "eaat" is more similar to the word "eat"?. How is the architecture?, is it possible for fasttext without using skipgram to be able to classify words?. Thanks

[@snagseeker]

sir,how it is dividing row words and columns word why both rows and columns does't have same words ?

[@sandipansarkar9211]

Thanks Krish once again

[@theniyal]

Can't you do the one hot representation with Tensorflow 'tokenizer and sequence' fucntion?

[@ManelTIC]

What is t'he relstion between embedding words and context window?

[@cool_videos6016]

Hi krish, It was great video I am a beginner and I started seeing some projects on kaggle related to lstm I always had one doubt that is when to use a specific layer like in some projects they use two lstms they use dropout with certain value and these things are different in different projects and I get confused how did they choose these layers . I would request u to make a video on how can we know when to use a certain layer and why

[@DineshBabu-gn8cm]

I don't understand what vocabulary and vocabulary size are please some one explain. then what if our word is not in our vocabulary of 10 k. Please someone explain.

[@Ankitkumar-qh6tx]

very helpful

[@ahmedosama4973]

Thnx sir I have one doubt for this ..what will be the benefit for the word representative is that we can predict sentience or what the advantages

[@kalppanwala6439]

Me waiting daily for Krish's videos be like me waiting for Money Heist Season 5 (aage kya hoga iss video ke baad) :)

[@barax9462]

Hello, im doing a project in which im not allowd to use AI librarries so, Can you pleaes explain this: - if i have an initial weight-matrix consisting of embeding matrix of size (300)x(|vocab_size| = 4k) and its filled with random values. And then we have an one-hotted sentence input of size say 3k sen = [0,0,0,0,157, 8, 900,100] etc.. my mian question is how to multply/dot-prodct the embedding-matrix with the sentence vector??? im really confused about this. should i convert the sentence vector into a matrix of size |sentence_vector| x embedding. or should just multply the indeices with the embedding matrix?????

[@chandanbp]

How are we deciding no of features, what are those features exactly in the given problem?.

[@krishnaik06]

It is difficult to interpret..

[@chandanbp]

@@krishnaik06 Or can you just give an analogy what those features could be

[@patrickadjei9676]

Please Explain Why your dimension is 10 when your set of features are 8. And why are you using a vocal of 10000 when your actual vocabulary is far less than 10000. Please explain.

[@pulunghendroprastyo5868]

Up

[@cCcs6]

Hello Krish, I have one question. I followed your tutorial and created those word embeddings. However, how can I fit them to a model (for example SVM) since they have a 3D-shape the model does not accept it. Thank you in advance, Christ

[@muntazirmehdi503]

how do we set the vocab_size to 10k or any other number

[@varunpusarla]

How to decide the value for vocab size ?

[@abdelhomi836]

At the very last part of your video can you do an inverse transform to your input for Semantical predictions instead of matrix prediction output?

[@infinitum90]

Krish,thank you for the clear visualization but 300 dimension is the parameter but how keras embedding layer calculated the 300 values of vector.Pl. can u explain.

[@suvarnadeore8810]

Thank you sir

[@tanmoybhowmick8230]

Hey krish can you show some deployment of ml model ??

[@mohamednajiaboo9817]

Thanks for the video. When we add the embedding we need to set the feature size. Here we set as 10. So how we the Keras know which of the 10 features need to be selected?

[@Trouble.drouble]

Krish is dictionary " is bag of words "

[@krishnaik06]

Dictionary are all the possible words

[@Trouble.drouble]

Then what is this Bag of words ji

[@krishnaik06]

Check my NLP playlist there I have explained

[@Trouble.drouble]

Sure sir 👍 I'll see thank you sir

[@siddharthpandit2117]

why not padding added at the end. What changes if padding is at the end?

[@SB-bu3xt]

Sir make a video on step by step guide for beginners who wants to learn ML

[@RAZZKIRAN]

actually one hot encoding is vector represation, again embedding layer converts to vector representaion?

[@RAZZKIRAN]

confused ? do we need to converted one hot repesentation embedding layer?

[@kmnm9463]

Hi - I think No. Just use Tokenizer from Keras and pass it into the Embedding Layer. One Hot is not required. This is from a user ( my name too is Krish - so don't confuse with Krish(presenter - the great teacher). :)

[@khangamminh512]

Is the output of this with word2vec different in nature, guys?

[@ameerhussain5405]

Finally I understood the embedding layer!! I had gone through many tutorials but failed when I started implementing but with this one I came through. Can't thank you enough Krish!! I would like to add 1 point to your code if it helps any 1, if we could add a Flatten() layer after the embedding layer then we could add Dense() layers and make predictions say if we are doing text classification.This model wouldn't do any good in terms of accuracy. But this helped me build an intuition on shapes of the tensors and what happens to text when fed into a deep networks which is much difficult to visualise than what happens to images when fed into a CNN.(at least for me 😅)

[@SameerSk]

instead of tf.keras.layers.Flatten() use tf.keras.layers.GlobalAveragePooling1D().

[@suryanarayanan5158]

amazing

[@mohdzaid6533]

Sir how can I install tensor flow and keras package R ?

[@rishabhvarshney2234]

can you tell me how vocab = 10000, vocab means unique word in whole corpus write?

[@wilman9206]

hi sir i wanted to know how can i search for the sentences using word embedding thanks in advance.

[@maYYidtS]

can anyone suggest to me....what if the text size more than 8 words we lose information right....is there any other way to overcome this problem?

[@kaziasifahmed2443]

plz can anyone tell me how did 100000 parameters are formed?? meaning what works behind 100000 parameters.10 dim *10000 voc size ? but why?

[@sutopa8377]

Hi Krish, Can you explain the concept of attention mechanism and please explain it in general, not related to encoder-decoder machine translation application.

[@vinitkanse5703]

How to decide the value of vocab_size ?

[@krishnaik06]

How many important words may be there in a dictionary?

[@vinitkanse5703]

@@krishnaik06 okay, but how do we differentiate between important words and other words? And I have one more doubt, In the video of word embeddings you said the "Embeddings are nothing but a feature representation " . So while implementing it (tf.keras.layers.Embedding) which features and what type of features and on what basis they are generated?? Like in the video there were gender , age ,etc (300 features) please look into my doubt, Thanks and Reagrds

[@krishnaik06]

Features things we cannot see as such...but based on the research paper it is how it is represented

[@nobalg3482]

@@krishnaik06 can you please refer to some research papers or make some video. I think you meant by 'we can;t see as such' is that one should thing this as an abstract idea happening behind the scenes; the system is creating and mapping features on its own?

[@aartiahlawat8228]

@@krishnaik06 yes sir same doubt i am also facing

[@rog0079]

so whats the difference between word2vec, and this embedding layer provided by keras?, do they perform the same job>?

[@noureddineghoggali9995]

Can you please make some tutorials on how to integrate the constraints in machine learning or deep learning? Thank you

[@Maryamkhan-lo1hq]

Great explanation. one question what if we pass our dataframe which consist of 500 posts with label text instead of sentences, does it works?

[@ADESHKUMAR-yz2el]

how can dictionary have index? it is traversed by keys here we don't care about the position of elements , we call elements through keys regardless of its position. plz correct me if i am wrong , index of dictionary sounds confusing 😩

[@anagham2413]

How can I convert more than 32 words?

[@BharathKResu]

Hey krish, can u guide us on NER, like resume parsing for example..

[@MRaufTabassam]

Did it work same for urdu?

[@krishnaprasad-un3hy]

Krish this is a wonderful explanation. I just wanted to know that, I have watched your previous three videos on NLP and i want to learn this technique from scratch. So, is that enough or we have other topics to cover?

[@fineescape1257]

Ifi watched this video and I don't understand, anyone has a suggestion of what I should watch first?

[@chaitanyakulkarni6012]

please do a video of installation of tensorflow for gpu i have same laptop as yours facing issues from a long time

[@aaryannakhat1842]

which gpu do you have?

[@biswajitroy-zp6lk]

how to judge how many features to take

[@akhilyeduresi8145]

Hi Krish, Where Can I find Word2Vec and GLove implementation the same as above?

[@pranayghosh1584]

why need to pad each sentence before embedding?

[@VijayKumar-bk5be]

Hey krish can u help us more videos by using system as well (practical)?

[@mohsinimam2048]

Thanks!

[@bilalhameed248]

voc_size=10000 dim=10 You did not explain these two variable clearly, why we consider dim=10 and voc_size=10000 please explain with some logic

[@priyanshramnani1751]

Thank you! :)

[@ravishankar2180]

your vocab size is hardly 50 and you have taken as 10000?? strange!

[@tagoreji2143]

thank you so much Sir