1 00:00:00,960 --> 00:00:01,370 Hello. 2 00:00:01,410 --> 00:00:01,900 Welcome back. 3 00:00:02,190 --> 00:00:09,800 So let's see how to build to the hidden layer neuro network. 4 00:00:10,090 --> 00:00:12,960 I'm gonna make a copy of the last project over here. 5 00:00:12,960 --> 00:00:18,180 Copy paste I'm going to call this hidden layer and then 6 00:00:30,360 --> 00:00:30,830 open it. 7 00:00:42,180 --> 00:00:42,830 Right. 8 00:00:42,870 --> 00:00:45,420 So let's take a look at the architecture again. 9 00:00:47,610 --> 00:00:51,680 So that's what the hidden layer neural network looks like. 10 00:00:51,720 --> 00:01:02,180 So we've got our input here and our outputs pass through another layer before we have. 11 00:01:02,180 --> 00:01:08,600 And before we get our output to this computation to run at a layer before we have our output layer. 12 00:01:08,720 --> 00:01:09,570 Right. 13 00:01:09,620 --> 00:01:13,050 And we're already described how we could do this. 14 00:01:13,100 --> 00:01:14,680 So let's implemented in code. 15 00:01:14,690 --> 00:01:19,730 Now this is the pseudo functions we wrote and we're going to write the actual functions now. 16 00:01:20,030 --> 00:01:20,470 Right. 17 00:01:20,480 --> 00:01:22,300 Or minimize this. 18 00:01:22,860 --> 00:01:30,600 And I'm going to go to our neural network to see file and I'm going to create the function is down here. 19 00:01:31,310 --> 00:01:32,110 Right. 20 00:01:32,150 --> 00:01:38,360 So the function we're going to create is going to be called the hidden layer and n and this function 21 00:01:38,360 --> 00:01:41,150 shall take seven argument. 22 00:01:41,150 --> 00:01:44,510 The first argument is going to be the input vector. 23 00:01:44,600 --> 00:01:50,300 Then the next argument is going to be the length of the input and then after the the third argument 24 00:01:50,300 --> 00:01:58,340 will be the the length of the hidden layer which means the number of nodes or the number of units in 25 00:01:58,340 --> 00:01:59,870 the hidden layer. 26 00:02:00,020 --> 00:02:04,390 Whenever I say node or unit just think of this. 27 00:02:05,290 --> 00:02:05,710 Right. 28 00:02:05,720 --> 00:02:08,790 The Psychos are the nodes or the units. 29 00:02:08,990 --> 00:02:09,350 Right. 30 00:02:09,560 --> 00:02:14,060 So I'll just start right in the function here. 31 00:02:14,160 --> 00:02:17,560 Call a void hidden and 32 00:02:25,440 --> 00:02:32,390 well like I said and like I said it takes a number of argument come over here. 33 00:02:32,390 --> 00:02:42,780 The first argument is a pointer to the input vector. 34 00:02:43,650 --> 00:02:48,540 The next argument is the input land you in 232 on the score t 35 00:02:51,070 --> 00:02:59,900 input line like this. 36 00:03:00,530 --> 00:03:21,400 The next argument after this is the length of the hidden layer. 37 00:03:21,660 --> 00:03:25,560 Then the next argument is the input to hidden layer weight. 38 00:03:26,790 --> 00:03:31,490 And that's what I mean by input to hit only a weight the weight for these snaps. 39 00:03:31,560 --> 00:03:33,080 The weight for these are roots. 40 00:03:33,090 --> 00:03:33,850 Yeah. 41 00:03:34,040 --> 00:03:42,350 Inputs to hidden layer weight these weight we're going to pass them as I comment number four. 42 00:03:42,600 --> 00:03:47,360 So it's going to be a two dimensional array or it matrix if you want to call it that 43 00:03:53,690 --> 00:03:59,030 so I'll just call this in to hit weight 44 00:04:01,880 --> 00:04:03,320 and the size is going to be 45 00:04:06,110 --> 00:04:16,060 hit in length the number of rows of course hit in length and the number of columns of course input land. 46 00:04:16,910 --> 00:04:21,680 Once that is done the next argument is going to be the length of the output 47 00:04:40,870 --> 00:04:41,430 after this. 48 00:04:41,440 --> 00:04:49,900 The next argument is going to be hit into output weight the weight for for this side hidden layer 2 49 00:04:49,960 --> 00:04:52,580 outputs layer the weight we need for them. 50 00:04:53,680 --> 00:05:00,300 Oh come over here and this is going to be a two dimensional array as well. 51 00:05:00,300 --> 00:05:05,920 Double I'll she hit out weight 52 00:05:12,090 --> 00:05:15,810 and the first size is output Lent. 53 00:05:15,980 --> 00:05:27,110 Second of size is it in length like this and the next and last argument is a pointer to a vector to 54 00:05:27,110 --> 00:05:28,850 store the 0 attributes 55 00:05:34,190 --> 00:05:38,980 like this right. 56 00:05:39,260 --> 00:05:46,310 So like I mentioned we simply need to perform we simply need to perform a matrix vector multiplication 57 00:05:47,210 --> 00:05:53,230 of the input vector and DMM to head in to inputs weight. 58 00:05:54,200 --> 00:06:02,950 And once we get a result we perform another matrix vector multiplication with a result with the hit 59 00:06:02,960 --> 00:06:05,200 into output layer weight. 60 00:06:05,660 --> 00:06:11,920 So I'm gonna start off by creating a buffer here which is going to be like a crutch array. 61 00:06:12,140 --> 00:06:17,030 Now we're going to use to store the hidden layer predicted values for the hidden layer. 62 00:06:17,900 --> 00:06:26,930 So we need to perform a matrix to vector multiplication using these inputs and these weights these weights 63 00:06:27,500 --> 00:06:28,900 are represented here. 64 00:06:28,970 --> 00:06:31,360 We need to take these inputs this a vector. 65 00:06:31,650 --> 00:06:38,420 It's gonna be a one dimensional array and this two dimensional array the result that we get is going 66 00:06:38,420 --> 00:06:45,290 to be hit in 0 hit in one hit into these are going to be the result and we take this resort and then 67 00:06:45,290 --> 00:06:49,570 we perform another matrix vector multiplication. 68 00:06:49,610 --> 00:06:56,050 But this time with this weight to hit into output weight and after that we get the results for side 69 00:06:56,110 --> 00:06:57,610 sick and active. 70 00:06:57,620 --> 00:07:02,760 So basically this is how a neural network with hidden layer computes the resort. 71 00:07:02,930 --> 00:07:04,870 Right. 72 00:07:04,880 --> 00:07:12,160 So come over here and see double it in predicted value 73 00:07:19,200 --> 00:07:21,050 it is predicted vector. 74 00:07:24,960 --> 00:07:28,650 And this is going to be the size of the hidden layer to hit it. 75 00:07:28,650 --> 00:07:28,920 Then 76 00:07:32,190 --> 00:07:37,240 and then I'll see matrix vector multiplication. 77 00:07:37,340 --> 00:07:40,830 The first argument is going to be the input vector. 78 00:07:40,910 --> 00:07:47,190 The second argument is going to be the length of the input vector which is input land. 79 00:07:47,200 --> 00:07:54,620 The third argument is going to be the output vector and we're going to store the output gonna story 80 00:07:54,660 --> 00:08:00,480 output in this buffer that we've created it in predicted vector values. 81 00:08:01,740 --> 00:08:06,580 So I'm gonna see hidden predicted vector here. 82 00:08:06,660 --> 00:08:08,390 That's where we store the output. 83 00:08:08,700 --> 00:08:18,880 And after this the next argument is what is shown here with the color blue is output land. 84 00:08:19,010 --> 00:08:21,630 Now after this the next argument is 85 00:08:24,670 --> 00:08:26,790 the next argument is the weight matrix 86 00:08:30,360 --> 00:08:31,150 and the weight. 87 00:08:31,150 --> 00:08:38,710 We're going to use is input to hit in weight so pass in to hit and weight right. 88 00:08:39,600 --> 00:08:47,130 So after this computation we're going to have the result stored over here in this buffer here. 89 00:08:47,130 --> 00:08:53,760 So we're going to take this buffer and then hit it to output weight and perform matrix victim with application 90 00:08:53,760 --> 00:08:54,330 again. 91 00:08:54,830 --> 00:08:56,100 So I'm gonna perform another one. 92 00:08:58,530 --> 00:09:05,010 And if we had multiple layers multiple hidden layers we would have to keep computing like this for each 93 00:09:05,010 --> 00:09:06,480 hidden layer. 94 00:09:06,480 --> 00:09:13,020 So it would be nice for you to try adding another hidden layer to the neural net work example and try 95 00:09:13,020 --> 00:09:15,210 to write the function to compute that one. 96 00:09:16,020 --> 00:09:26,130 So I see a matrix vector multiplication and the input vector this time is hit in predicted vector and 97 00:09:26,130 --> 00:09:34,500 the length of this is Headland and we see in Hidden Land it just so happens that in our example I'll 98 00:09:34,500 --> 00:09:43,590 show you it just so happens that in our example the hidden length or number of nodes and the the input 99 00:09:43,590 --> 00:09:48,570 length or number of nodes and the heat in length and the outputs are the same. 100 00:09:48,600 --> 00:09:52,970 It just happens that all have three nodes but you can have two here three here. 101 00:09:52,970 --> 00:09:54,830 Did they come in different forms. 102 00:09:54,990 --> 00:09:59,910 So that is why our function is not just saying give the length. 103 00:09:59,910 --> 00:10:07,050 If we wrote the function to accept just one length value then it's going to be problematic cause our 104 00:10:07,050 --> 00:10:08,520 lives have different length. 105 00:10:08,520 --> 00:10:13,430 That is why we're dealing with hidden length output lint etc.. 106 00:10:14,970 --> 00:10:23,510 So the next argument is the output for buffer and we want to store this in the output buffer. 107 00:10:24,110 --> 00:10:27,020 And after that the next argument is the output lend 108 00:10:39,680 --> 00:10:41,580 and then the weight the weights. 109 00:10:41,570 --> 00:10:49,390 Here is the hit into output weight hit two out weights like this. 110 00:10:49,410 --> 00:10:51,500 So this is it. 111 00:10:52,680 --> 00:10:54,150 We can expose this function 112 00:11:19,020 --> 00:11:23,440 and we can go to our main file to test it out. 113 00:11:23,800 --> 00:11:24,710 Gonna come over here 114 00:11:29,210 --> 00:11:29,380 and 115 00:11:32,450 --> 00:11:39,810 I'm going to add a new defined statement here for hidden layer lent or say define. 116 00:11:40,010 --> 00:11:48,610 I did Len the length of the hidden layer is 3 as well right. 117 00:11:48,920 --> 00:11:55,910 And actually I'm going to keep these weights and this time these weights I'm going to make sure these 118 00:11:55,910 --> 00:11:57,570 are the weight for the hidden layer. 119 00:11:57,770 --> 00:11:59,090 So this will be hit zero 120 00:12:08,480 --> 00:12:09,280 hit 1 121 00:12:13,070 --> 00:12:13,790 head to 122 00:12:17,420 --> 00:12:23,150 like we have over here right. 123 00:12:23,630 --> 00:12:25,500 And we need a new set of weight. 124 00:12:25,570 --> 00:12:28,280 We've got to change the name to now. 125 00:12:28,400 --> 00:12:29,270 This is called 126 00:12:33,800 --> 00:12:35,270 input to hit in weight 127 00:12:45,490 --> 00:12:49,450 and it's hidden lent by input land 128 00:13:01,740 --> 00:13:02,290 right. 129 00:13:03,470 --> 00:13:09,480 Okay so we need to create a different set of weights. 130 00:13:09,510 --> 00:13:10,080 Come here 131 00:13:21,680 --> 00:13:22,240 turbo the 132 00:13:36,080 --> 00:13:38,590 this is going to be hit into output layer weight 133 00:13:48,890 --> 00:13:53,930 and output line and then hidden Len 134 00:14:06,520 --> 00:14:10,650 and we can just assign certain weights here let me see. 135 00:14:10,650 --> 00:14:15,270 I'm gonna say minus one point zero 136 00:14:18,640 --> 00:14:32,310 and one point one five zero point one one. 137 00:14:32,660 --> 00:14:45,770 This next one can be minus zero point twenty eight zero point one five minus two points you're one. 138 00:14:46,940 --> 00:14:57,020 And this last one can be zero point two five minus zero point to five minus zero point one. 139 00:14:57,230 --> 00:14:59,690 These are just like I said random numbers 140 00:15:08,280 --> 00:15:10,800 to put a bit of comment here. 141 00:15:13,400 --> 00:15:16,670 So this would give the sad result. 142 00:15:21,790 --> 00:15:25,630 This over here for sick or not sick 143 00:15:38,640 --> 00:15:42,300 and this is for active or not. 144 00:15:42,710 --> 00:15:51,970 And we know that this one is from if want to put this here. 145 00:15:52,670 --> 00:15:57,700 This is input from it and let it hit in here. 146 00:15:58,050 --> 00:15:58,550 Um no. 147 00:15:58,550 --> 00:15:59,750 Zero. 148 00:15:59,750 --> 00:16:09,780 And this is from head 1 and this is form hit 2 right. 149 00:16:09,950 --> 00:16:15,950 Once we have this we can um let's see put a semicolon here 150 00:16:20,390 --> 00:16:28,220 we can use the same input vector we have input vector here so we can come down here and call our function 151 00:16:36,870 --> 00:16:41,370 the function is called hidden layer and then you can simply see it in the 152 00:16:44,170 --> 00:16:49,900 or hidden n in does the name of the function the first argument is the input vector 153 00:16:55,220 --> 00:16:57,230 the second argument is the input lent 154 00:17:00,410 --> 00:17:07,790 the next argument is the hidden layer lent then the next argument is the input to hidden wait 155 00:17:12,880 --> 00:17:21,650 then the next argument is the output lent then the next argument is hidden to output wait 156 00:17:28,160 --> 00:17:29,530 do we have more argument 157 00:17:33,160 --> 00:17:42,030 the next argument is the output vector we don't have which not created an output vector here I'm gonna 158 00:17:42,040 --> 00:17:42,980 come over here. 159 00:17:42,990 --> 00:17:47,050 Okay this one hit predicted output I'm gonna copy this 160 00:17:50,210 --> 00:18:00,430 put it in here put it some call on here like this and um yeah everything else remains the same sad prediction 161 00:18:00,430 --> 00:18:07,100 is going to go here and then seek prediction here and then active prediction here. 162 00:18:07,180 --> 00:18:10,730 Right let's run out code and see what we have. 163 00:18:10,730 --> 00:18:18,660 Click here to build it to be successfully download onto the board let's see Terra to 164 00:18:24,860 --> 00:18:33,130 or reset my board and this is where we have sad prediction sick prediction and active prediction right. 165 00:18:33,230 --> 00:18:36,260 So this order is for this lesson and I'll see you later.