How many times more probable This is to belong to this place in another class One is that requirement So I conducted artist ical analysis using this probably varies which is not possible isn’t wrong numbers the next one page is when I was he can’t lead error How do I define ever tyrannous actual minus predictor If when you’re doing some Oscar let us It doesn’t matter Virginia for put first Next Okay If the actual class to which I expect I expect expected profit everyone right If I’m inputting a zero I expect the poverty of this to be one and rest of them Toby zeros How do we counter-terror.
I can’t read That really backfired I expected Probably minus predicted Probably this better conclusion I will not be able to do if I use that all of us Because in raw numbers and some of the numbers are native I’m not be beautiful Defined layers That’s what you can return the soft plaques and I’ll put the soft Max is used to find the others Axl s the problem Big decisions What to me honorably Are you wondering what is going on Sure I’m in person Yeah Yes sir.
What I understood is it Order the hour Come from all the player is pastoral function calls off Max And then you take that There’s actual or put Yes I get it So the softmax will be instruction with algorithm Whatever Outputs come here This is not a different layer of you know something It is It’s an instruction to the algorithm Whatever the output of this new network transform it into productivity’s Okay So soft Max Okay Uh I didn’t go Are you thinking Are you with me Sure Okay good Okay let’s move for the propagation We already discussed Forward propagation Is your entire data set going through the neural network from the initial layers Still doubt declare that is called forward propagation.
The result of that is an error okay And too lazy to do any indication So I just took a snapshot of a whiteboard which later on I realized it’s not so white enough I told you, in the beginning, I told you doing that after the break All neural networks are nothing but Macronix algebra We really don’t need to know more these things but I just trying to show you an example here So you’re a neural network then put values of these points on appointment Pointed off the cuff No reason why Porn 9.8 There’s some example.
This gets stored in metrics All the weights that you’re having here these rates are shown on the connections But as I told you these weights are inside the neurons 0.9 points three in 30.4 is inside the neuron Similar 0.2 point eight and point to these are inside the neurons They’re stored in a day inside the Newmans Okay so what we show here is this is the metrics 0.9 points 2.1 belonged to the first You know on the second you don’t own your own So this forms your layer one Okay When you multiply these two through the daughter operations the result of the daughter operation is some output which has shown here at one point.
I think 66.42 and 60 that again get stored in this thing metrics which then guilt gets through on which you performed the signboard operation All part of a signal depression for these three neurons and stored here in a Metrix this matrix again gets multiplied by the weight metrics over here 0.3 points 6.8 points three Ah point Oh hi Mr. Imagine this again Wait for Matrix This weight metrics get multiplied with this The result of that is an output This output undergoes a signal transmission again here And then you get all parts of what you see here is diagram It’s one leaf or explanation.
It’s only for into Shin Nothing of the sort Actually it happens And either what actually happens under the hardest Mavericks all it so entire neural matrix-matrix operations Have you heard of TensorFlow by agents Sensible We’re going to do tomorrow Tensorflow and care us pencil floor is engine computing engine designed by Google on the world Spencer comes from the fact that in mathematics one dimension Larry even one scaler value one dimension Larry Prudential every three-dimensionally All these accords sensors on Since the tenses transform the floor from here to here to give the outboard This whole thing is going TensorFlow Hence the name TensorFlow toe that engine.
We’re going to use that tomorrow and see what it is Yes ma’am You earlier said this is a black box So when at every point you ever toe um given name to the function So you know that you start from Image E And then you applied some function got the probability and then you have a subsequent explanation which is functions Why do you call it a black box No I want you to explain it in simple English Okay Why are you before God I didn’t find this offensive reporter so I can’t go there and the court of law and say you know what This new one gave a rate of this that you don’t get rid of this It’s inexplainable to Lehman Whereas your data centers my next memorable I wanted to basement Freeman.
If I use a decision tree in decision theory I can follow the part I can print off those parts is English rules And put that ass my reason for you I did find some days before Yeah So Banks because of the laws they have around it you cannot But otherwise mathematically you can say applying these functions We get to this point It’s just And when we say English white books it means ordinary implements I shall move Okay I’m not going to cover too many concepts Then this is against saying that every step is a lot Bye Stumpy Already discussed I had to jump This old too can have the liberty of jumping this night Help ever functions Very important Rule the play in the definitional.
Always your model works inter function, in this case, I’m sure you an example of square later Least square using the new tradition, Okay Now let me explain this to you I think you would have already seen this talk about two Then we’ll wind up but yeah Another five minutes with my many are better function or defined like This inter function mathematically has a very good probably the probable school convicts property Which means if I block okay I already discussed this morning This is competitiveness y minus a mix plus C raised apart too That means there is a function off am and see.
If a plot a graph between error on the different combinations off MNC that I can have then this plot will always look like a parabola There you will have some combination of MNC refuses the least error may not be zero but least any other combination of MNC will give a hider Okay The child’s off error function is very important given the problem statement a tag If I make yourself a simple quadratic function then I have an advantage Motor functions were going to look like this on this What the grading descent elder will do us You start with any random MNC What Do whatever you’re doing.
It’s we’ll find out for you It will gravitate the words that combination feminine See which uses the least er on the morning that you’ve been using these to that model is the best fit like oh live All of okay In logistic regression you have seen another Erica log loss If you’re doing bad reclassification using logistic regression there, two classes, They’re zero and one a class which belongs to one Okay in the train that you’re going down in the train say a data point which belongs to one the more it is identified as one Okay the more it is identified as one the more confident you are That is one they did This is it This is the inner call for glass one where this is the probably belonged to class one as a problem A data point which belongs to classman The more the property less of the error in your predictions.