It actually says that you have calculated the probability of the next word whatever you are predicting But actually that whole sentence doesn't exist in your car purse But you are saying there is some probability that this can happen It's greatest So I'm saying like books W is booked on it doesn't occur in the whole corpus Okay so what I'm saying I'm calculating probably of next four It can be booked What is the probability that the next world can be booked Okay given whatever the previous statement is actually whole statement So here if that particular, not books What actually poverty off w students opened a book.

Here don't student open they're to book the counterfeit So there is no count actually zero cones on Then you're riding some small value to it And then after that just waiting to count students open their if it exists in that corpus on Do you have some value here But what if those students open they also don't exist in your car pass And that's a problem because you are the wedding play zero That's something that nor the probability So that's the other problem Okay so ho you do that.

This is some kind of example Okay so this same example But you are like making sure that students open there is like a statement And after that, there should exist something But it can be that we just end three words off a statement And after that, there is gets nothing but yeah but you think that with traditional ml techniques you can do that kind of stuff Probability is not zero going t zero, Yeah I mean if you want Yeah Yeah probably the zero But you're talking about the concentrate If it doesn't exist that so and so open their books doesn't exist Then it's zero counts Is it One of So how you prepare your traditional animal models for text analytics.

You do pre-process You do feature extraction And then after that in future exception what you're doing you are preparing a numeric form off your converting text to numeric form and then you are importing that numeric to your moral But if you want to do this I'll do that You have to take care at valuer converting it to numeric form Yeah that's the reason traditional ML doesn't work for a second Straight up, Yeah it's on you Actually this is just example, Okay So if you feel like if I just buy grounds and it will give me better than you you can go for my ground actually But sometimes we will have to hear.

If we are choosing like we want to go with 4 g like we want to know the fourth word OK then we have to have a try, Graham, Also because that's what will be like used for as like to predict the Fort Worth So we want to have they come and 4 g also But that's two vectors only That's don't only input betting that we can use And that's during feature extraction Only they're only you're manipulating all these things Yeah you can't give something like two inputs to your model Uganda You have to do it in future ex section And after that whatever you probably you're getting you can give it to your model as input.

So the okay We didn't talk about the second problem right Yeah So But if students open their it doesn't exist in your data then what Then it's a bigger problem right It's infinity Probably going to be infinite So there just condition and open their instrument This is a cold Becker So instead of going with like syringe open there we can use open there just open But here we have to make over like this feature extraction this way How we can add that how we can or we have to write logic, Okay If we are getting the probably zero then go for like 2 g but with like just last words that kind of logic You can also write the first one also But how many games If you are dealing with a foreground of 5 g then Yeah it's like you're using machine learning but still you're using a rule-based thing.

They're So you are not doing something that is very just useful So that's two problems And then there is one more so What's that need to store Come for all possible And gum So mortal sizes, Ah you know the complexity Thank complexity of modern whatever model you have study or feature extraction technique PC if you know you have principal company Delaney Yeah you're lived in their correct So you know the complexity level for peace here that kind of stuff right there So similar way It's also a feature extraction techniques So here the complexities' explanation off an on his number off words that do you have in you become so to store that kind of tank I took a lot of sense.

If your car passes you and for actually for like prediction you need that a large amount of corpus So that's 30 problems that you face in the traditional family You can have practically you can have particularly you can have But if you want to implement it on like tactical basis it's impossible It will take to train your model It'll take like years Yeah so recalled on Large's mortal task What Skins Awards for the distribution of next word That's what we have done well now right whole about a window-based neural model So the window We are already where a window is Nothing but what're the seconds What's lost the forwards on If we say like forwards we have to consider than Windows.

We want to have an even know based neural model So how we can achieve that So you already started neural networks right So you know the basics Anybody can tell me that Uh okay I will not talk about window-based neural model first but we'll talk about neural networks Okay so here you can see the fixed window The students open there So that's a window here So we are going to use a similar concept in neutral networks But it's coiled fixed window, not N-grams are kind of stuff So it's slaked We have time You will take a break on 11 11 30 11 30 years We have time No problem So anybody I can explain it to me This thing have you seen this kind of thing And your neural net approx sessions Hair is for nerds actually ever four nodes.

This is just one note on Then we are getting output I'm not explaining a lot of things here on this leg because it just that and to give you information about what a neural network is actually and how we can use that way have just had it here But we'll talk about the noodle network first the basics We want to recap all those basics because it will make it possible for you to understand it more clearly Ordinance on Alston's.

We talk about all those things So here the thing that I want to talk about this again you can see that one Heart victors Honda we have concocted later on word on Then we have a day earlier and then we have assault Max What We go talk Max winds And what's this actually off this layer This softmax function that we have used What It is actually called in journalists Yeah its activation on what is this he didn't live on And General Lee's name for its Aggregation function We use aggregation function right They lose them, Yeah but most of the time we'll talk about all they'll recap All those things for you actually because it will give you a heads-up What we are going to cover and for an organ and on how we can relate all those things OK but what the word soft makes I want to ask a word Soft makes uh anybody veered off soft makes function.