Check our Deep Learning Course (in PyTorch) with Latest, Industry Relevant Content: tinyurl.com/4p9vmmds Check out our premium machine learning course with 2 Industry projects: codebasics.io/courses/machine-learning-for-data-science-beginners-to-advanced
The good example for converting word to numbers is the coding of color in RGB : black = (0,0,0) red =(255,0,0) yellow=(255,255,0) green = (0,128,0) blue = (0,0,255) etc....
how do we define the features for every word in the language , like in this example in the video it works well for a country or a person or a cricketer what is a word red is used here here is no feature called color , if we increase the features then there will more number of features again if we want to compute for every word in the language by describing the features
Whether it is an colored pixel or a word, machine learning model doesn't get any of these things. The only thing in the entire universe which a machine learning model or a computer can understand is *_Binary Digits_* or *_Numbers_* , nothing else. So, whatever information we are feeding into the model (whether a colored pixel in case of images or words or any other object or information), we must grind that information to numbers, so the model can digest it and give you result.
Hi Dhaval, The analogies with which you start every DL concept in your videos are pure gems. Here the examples of players and Australia - with the float value for each dimension ( feature ) is pure streak of genius. One just goes spellbound when the whole concept become so clear. It is said that great teachers are the ones who make complex things look simple. And when one explains in simple terns it means that the person has understood it completely. It fits you perfectly. Before watching this video, though I could understand what word vectors were in terms of dimensions but the real meaning(conceptual understanding) become immensely clear when you presented them with values like - Person, Fitness, Location, Has Government - was mind blowing. Just one question - let us say that we decided the dimensions to be 100 for each word in an embedding scenario. Should all the values of the dimensions add up to 1 ( like probabilities adding up to one)? Just want to get that clarified. Thanks again for excellent session, Krish
I'm not the teacher and I legitimately have no idea but judging from his example and from the first vid he mentions alone I would say no. Also logically it's just a matrix representing like qualifiers about the object that can be represented in quantitative form in a matrix to determine it's relationship and similarities to other words.
Nice video. "I am working in Apple and I will eat Apple later" Apple giving different meaning in two sentences. How to handle this in word embedding ? This is very famous tricky interview question. Could you please try to make one separate video on this.
Let me make a video on this but this is a named entity recognition problem where bases on the context (surround words) one can find out if it is fruit or a company. For example if you look at a huge text corpus most likely apple (fruit) will be surrounded by words such as eat, vitamin, healthy etc. Where as Apple (the company) will be surrounded by words such as working, iPhone, revenue etc. One can train a model that analyzes these surrounding and can tell you which word it is. I will make a video with code etc
Good explanation. Some things not clear like if we are training the model with data from wikipedia data, how are we coming up with the feature data for each words. Like as you explained for the words - Dhoni, cummins and Australlia these were the feature data (person, location, fit etc), how to comeup with feature data for each words? .
We are nearing the end of this series. May be 10 more videos. After that I will start deep learning projects that uses the concepts taught in this series
how do we define the features for every word in the language , like in this example in the video it works well for a country or a person or a cricketer what is a word red is used here here is no feature called color , if we increase the features then there will more number of features again if we want to compute for every word in the language by describing the features
can you give details on what exactly is the problem statement? Is it comparing one image and one block of text for similarity? IF that is the case then one way you can solve this is for the image, use some type of vision api to come up with image description from an image and now that you have generated text from image, you can compare it with another block of text using vector embeddings and cosine similarity
@@codebasics the problem statememt is exactly what you wrote in the reply. I was trying to extract image features with models like VGG16/Image Net and converting those features to vectors. I am sucessfully able to do that but the dimensions of Image vector array is way higher than text vector array therefore compatibility issues while checking the cosine similarity. Alright i will try to do what you said. Thanks for the quick response.
Is it possible with C instead of Python? I use a 14 gigabyte corpus of English on a Raspberry Pi 4B and I am afraid Python is not fast enough for that.
For numeric datasets, you can use simple artificial neural network. Even before that you should explore using non DL methods such as random Forest, svm etc. CNNs are useful in capturing spatial features and for numeric datasets often that is not that case
Check our Deep Learning Course (in PyTorch) with Latest, Industry Relevant Content: tinyurl.com/4p9vmmds
Check out our premium machine learning course with 2 Industry projects: codebasics.io/courses/machine-learning-for-data-science-beginners-to-advanced
The hallmark of a great teacher is to explain complex topics in a simple manner. This is one such example. 👏
This is so far the best video i saw on understanding word embedding concept.
The good example for converting word to numbers is the coding of color in RGB :
black = (0,0,0) red =(255,0,0) yellow=(255,255,0) green = (0,128,0) blue = (0,0,255) etc....
Genius
how do we define the features for every word in the language , like in this example in the video it works well for a country or a person or a cricketer what is a word red is used here here is no feature called color , if we increase the features then there will more number of features again if we want to compute for every word in the language by describing the features
Whether it is an colored pixel or a word, machine learning model doesn't get any of these things. The only thing in the entire universe which a machine learning model or a computer can understand is *_Binary Digits_* or *_Numbers_* , nothing else. So, whatever information we are feeding into the model (whether a colored pixel in case of images or words or any other object or information), we must grind that information to numbers, so the model can digest it and give you result.
you are my savior, my friend. God bless you
this video has been much more helpful than any other videos I've found thank you
Really Dhaval Sir, No One Can Explain Better than This.Your way of Teaching is Unique and Intresting. Thank you So Much.
I am happy this was helpful to you.
loves your videos...thanks sir...you are the savior.....
Best teacher ever!
You are a SUPER Teacher...
Fire Explanation 🔥
love that avengers painting on the wall😂
Dhoni == 7 ..... Thala For a reason .... BOLE JO KOYAL >>>>>> 🔥
Thank you for your excellent tutorials. Merry Christmas and Happy New Year!
Awesome explanation about word embeddings.
Superb Explanation 👏
Very clear explanation for beginners 👍
Your videos are great !
Glad you like them!
Very informative, makes perfect sense
Thank you so much for the good work you are doing here...
Hi Dhaval,
The analogies with which you start every DL concept in your videos are pure gems. Here the examples of players and Australia - with the float value for each dimension ( feature ) is pure streak of genius. One just goes spellbound when the whole concept become so clear. It is said that great teachers are the ones who make complex things look simple. And when one explains in simple terns it means that the person has understood it completely. It fits you perfectly. Before watching this video, though I could understand what word vectors were in terms of dimensions but the real meaning(conceptual understanding) become immensely clear when you presented them with values like - Person, Fitness, Location, Has Government - was mind blowing.
Just one question - let us say that we decided the dimensions to be 100 for each word in an embedding scenario. Should all the values of the dimensions add up to 1 ( like probabilities adding up to one)? Just want to get that clarified. Thanks again for excellent session, Krish
I'm not the teacher and I legitimately have no idea but judging from his example and from the first vid he mentions alone I would say no. Also logically it's just a matrix representing like qualifiers about the object that can be represented in quantitative form in a matrix to determine it's relationship and similarities to other words.
Could be but not necessary.
TOO!! good explanation!
Nice video.
"I am working in Apple and I will eat Apple later"
Apple giving different meaning in two sentences.
How to handle this in word embedding ?
This is very famous tricky interview question.
Could you please try to make one separate video on this.
Let me make a video on this but this is a named entity recognition problem where bases on the context (surround words) one can find out if it is fruit or a company. For example if you look at a huge text corpus most likely apple (fruit) will be surrounded by words such as eat, vitamin, healthy etc. Where as Apple (the company) will be surrounded by words such as working, iPhone, revenue etc. One can train a model that analyzes these surrounding and can tell you which word it is. I will make a video with code etc
@@codebasics Thanks a lot
@@codebasics where can I find this video
Great explanation
waiting for next part👍👍👍
can we apply similar concepts to pictures or videos? if no then what can we use so we can do RNN processing on vdos and images?
Good explanation. Some things not clear like if we are training the model with data from wikipedia data, how are we coming up with the feature data for each words. Like as you explained for the words - Dhoni, cummins and Australlia these were the feature data (person, location, fit etc), how to comeup with feature data for each words?
.
Awesome
Sir, please make End to End NLP tutorials!
Excellent
Thank you sir!
You are welcome!
thank you so much
Approximately how many more tutorial videos are there gonna be in this entire DL series?
Bdw , the tutorials are too intuitive, keep it up 👍👍
We are nearing the end of this series. May be 10 more videos. After that I will start deep learning projects that uses the concepts taught in this series
@@codebasics Hi sir, I am really waiting for your playlist on Deep learning projects which will teach us to make real-world AI projects... Thanks
you can put height, weight, runs tked etc for explanation
how do we define the features for every word in the language , like in this example in the video it works well for a country or a person or a cricketer what is a word red is used here here is no feature called color , if we increase the features then there will more number of features again if we want to compute for every word in the language by describing the features
Pls sir,will this ur tutorial be good for data science or is different
Yes good for data science when you are learning NLP
Dhoni =7...Thala For A Reason 😂😂😂
Hi, I have a specific Question. Is it possible to Convert Image to vectors and Text to vectors and finally compare the Cosine similarity between them.
can you give details on what exactly is the problem statement? Is it comparing one image and one block of text for similarity? IF that is the case then one way you can solve this is for the image, use some type of vision api to come up with image description from an image and now that you have generated text from image, you can compare it with another block of text using vector embeddings and cosine similarity
@@codebasics the problem statememt is exactly what you wrote in the reply. I was trying to extract image features with models like VGG16/Image Net and converting those features to vectors. I am sucessfully able to do that but the dimensions of Image vector array is way higher than text vector array therefore compatibility issues while checking the cosine similarity. Alright i will try to do what you said. Thanks for the quick response.
2:10 bole jo koyel
Is it possible with C instead of Python? I use a 14 gigabyte corpus of English on a Raspberry Pi 4B and I am afraid Python is not fast enough for that.
what corpus do you have? Is it freely available?
useful
Sir I am not asking question with this topic. My question is can we use CNN for numerical dataset classification.
Like using CNN for people taking insurance or not.
For numeric datasets, you can use simple artificial neural network. Even before that you should explore using non DL methods such as random Forest, svm etc. CNNs are useful in capturing spatial features and for numeric datasets often that is not that case
Thank you sir
2011 cricket example :)))
Yes 👍
Hi Sir,
Thanks for awesome videos.
Could you please share any courses to learn end to end nlp.
Unfortunately I dont have any recommendation but I am myself going to make end to end NLP tutorials soon.
@@codebasics thanks alot ..
how can bathroom be 2.5?
Dhoni == 7 Thala for a reason.
😀😀😀😀
Hi bro
Aap roti bhi khatty ho ya bus code khatty ho