WEBVTT 00:00.960 --> 00:07.470 In this session, we will be discussing about the implementation of Chian, so let's have a look at 00:07.470 --> 00:07.800 this. 00:08.500 --> 00:13.770 So for this particular implementation, I have faith feed my indians'. 00:13.780 --> 00:14.980 Diabetes does it. 00:15.720 --> 00:18.950 So let us start with importing the required libraries. 00:19.500 --> 00:24.460 So the importance libraries are no Fondas and my. 00:28.060 --> 00:36.280 Next, we will load the data set for loading the data set, we will again use the dock re CSP, which 00:36.280 --> 00:41.750 will allow us to read the CSC fight the CSP files there. 00:41.980 --> 00:47.140 These are CSFI and there is no specific different delimiter. 00:47.530 --> 00:51.400 So that is why we are not giving any delimiter here and here. 00:51.400 --> 00:53.800 We have the data from the. 00:54.660 --> 00:57.860 Now we will be printing the first five rows of the data. 00:58.620 --> 01:03.920 So here we have the first five rows of data from which we have printed using the dotted. 01:04.940 --> 01:16.370 Now, the columns I pregnancy's glucose, blood pressure, skin thickness, insulin, BMI, diabetes, 01:16.370 --> 01:20.450 videgaray function, age and outcome. 01:22.210 --> 01:30.670 Here outcome one person that the person has diabetes and outcome zero represents that the person does 01:30.670 --> 01:32.080 not have diabetes. 01:33.010 --> 01:35.950 So now we will check the shape of the data frame. 01:36.310 --> 01:39.370 So the shape of the data frame is seven. 01:39.370 --> 01:41.210 Sixty eight, Guama nine. 01:41.470 --> 01:45.970 This we have obtained for using the shape attribute. 01:46.330 --> 01:51.120 So the shape gives the shape of the data frame the seven sixty eight point nine. 01:52.000 --> 01:58.960 Now we will check that we have to do seven sixty eight rows, nine columns. 01:59.230 --> 02:08.020 And out of these nine columns, the first eight columns are the features that this these first columns 02:08.140 --> 02:15.400 are the features and the last column is the Thugged or the label value which we want to predict. 02:17.800 --> 02:27.530 So we will create No by Eddie for features and targets, so the fullest Eddie will be ex and the second 02:27.580 --> 02:30.040 will be via X will contain the. 02:31.470 --> 02:37.170 X values, that is from pregnancy's to each and Y will contain the outcomes. 02:38.190 --> 02:45.990 How we do that, we simply beg the veto threat and if we will drop the outcome from this, we will get 02:45.990 --> 02:48.010 all of these columns except the outcome. 02:48.600 --> 02:50.060 So that is what we do here. 02:50.550 --> 03:03.330 We do the don't drop the column name and the axis Rumbo axis Zettl stands for Raus and Axis one stands 03:03.330 --> 03:04.590 for columns. 03:04.860 --> 03:08.880 And then we find out the values using DOT values. 03:09.060 --> 03:13.140 So we are getting all the values from the data from. 03:15.010 --> 03:23.050 And after dropping the outcomes from this, now, Vivi will be having only the outcome. 03:23.710 --> 03:31.630 So we are doing the outcome and hopefully we could have taken the data from itself also by doing the 03:31.930 --> 03:35.700 proper outcome while my ex is one and the. 03:37.370 --> 03:40.830 With the outcome, but this is like another way. 03:40.880 --> 03:42.800 So there are two ways how we can do it. 03:43.040 --> 03:48.200 So it is completely up to you if you want to get a Bayati or you want to get the delphiniums. 03:53.980 --> 03:58.300 Now we will split the data into training and testing data set. 03:59.310 --> 04:07.020 So let us split it using the brain split so we will import the brain split from Ascalon lot more this 04:07.020 --> 04:07.500 election. 04:08.070 --> 04:14.760 So from this election we are importing festering split and we will get split the data. 04:15.210 --> 04:20.520 Now, here I am splitting the data with this size equal to zero point for. 04:21.710 --> 04:30.260 And here I am, having started this fight right now, we have given random state to be 42. 04:30.290 --> 04:33.740 You can give any random state based on your convenience. 04:34.380 --> 04:37.810 I these are the values which we will be getting from this. 04:38.330 --> 04:50.600 So the X will be divided into extreme and X test and the Y will be divided into vibrant and whitest. 04:53.180 --> 04:56.970 Now, let's create a classified using dangerous neighborhood algorithm. 04:57.440 --> 05:04.320 So first, let us observe the accuracy for different values of now what case? 05:04.340 --> 05:07.070 The number of nearest neighbors. 05:07.760 --> 05:16.700 OK, so we what we will do, we will import the nearest neighbors, classify it from the Escalon neighbors. 05:18.030 --> 05:24.360 Now they will set up the area to store the training and the testing accuracy, so we have created this 05:24.360 --> 05:25.350 at a neighbour's. 05:26.560 --> 05:30.010 Which has values from one to nine. 05:31.100 --> 05:32.600 Now, we were creating. 05:34.330 --> 05:39.220 I've won it as drin accuracy and another area as best accuracy. 05:40.890 --> 05:42.870 Now they are running value. 05:44.110 --> 05:49.060 On the values of think, that is on the values of the. 05:50.630 --> 05:56.180 Neighbors which will be enumerated so basically on all the neighbors, the. 05:57.390 --> 05:58.780 Running this particular. 06:00.040 --> 06:03.940 Now we are creating an object of the cannon classify it. 06:04.830 --> 06:13.110 And in this object, we are putting the number of brzeski, so each time it will keep running the loop 06:13.440 --> 06:20.380 and it will change the number of neighbors from one do it, all the values will be taken in this case. 06:20.730 --> 06:27.120 Now, the other thing, the model using, again, is not fit and giving in the extreme, and rightly 06:27.510 --> 06:27.900 so. 06:27.900 --> 06:31.230 It belongs on the extreme and vibrant values. 06:32.770 --> 06:41.320 And after the food is done, we can get the cocaine in school, so for in school, we're running the 06:41.320 --> 06:50.090 values from ecstasy and victory and we are getting the training accuracy from this eye for testing accuracy. 06:50.410 --> 06:53.980 We are running it on its best and brightest. 06:55.170 --> 07:02.760 Now, from this, we have obtained these two areas which have been accuracy and testing accuracy. 07:04.690 --> 07:13.170 So now we will create the lot to actually visualize what we have obtained from this train accuracy and 07:13.180 --> 07:14.830 the best deputies. 07:16.690 --> 07:22.660 Now we can see that this blog title we have given game, nearest neighbor, varying number of neighbors, 07:23.140 --> 07:25.620 the plot has two lines. 07:25.930 --> 07:32.820 So the first line is from the accuracy and the second line is from the three inaccuracy. 07:34.820 --> 07:42.320 And we have created a legend, provided the exlibris and vibe when we are printing the Lord, we are 07:42.320 --> 07:45.400 able to see these two lines generated. 07:47.530 --> 07:56.020 Now, when you see the training, accuracy is decreasing as the number of members increase. 07:57.630 --> 08:02.160 And the testing accuracy is actually increasing slowly. 08:03.500 --> 08:06.050 And if we compare the values. 08:07.000 --> 08:16.210 These values increased in the number seven, number seven is the point where these accuracy's are closest 08:16.420 --> 08:19.100 and the best accuracy is pleasing. 08:19.390 --> 08:24.280 And after seven, the best accuracy actually starts to go down. 08:24.730 --> 08:30.010 So this is the reason why we will select seven as the number of neighbors. 08:31.930 --> 08:39.730 So we can observe that we get maximum testing accuracy for people to seven, so we will create the dangerous 08:39.730 --> 08:42.710 classifier with a number of neighbors, that's seven. 08:42.940 --> 08:44.340 So how will we do that? 08:44.350 --> 08:47.290 We will meet again another object of neighbors. 08:48.700 --> 08:54.770 And this time, we will provide the number of neighbors s7 in this object. 08:54.790 --> 08:59.540 We will provide the extreme and vitrine and then fit the model. 09:00.130 --> 09:04.170 Now, when we put the models, the model will learn from this data. 09:06.030 --> 09:10.980 And after learning from this data, we get the Cannon School. 09:12.180 --> 09:17.220 This Ganin score comes out to be zero point seven three zero. 09:18.850 --> 09:26.980 Now, next is creating a confusion matrix now of confusion matrix, is it a move that is often used 09:26.980 --> 09:34.000 to describe the performance of a classification model on a set of test data for which the true values 09:34.000 --> 09:35.170 are already known? 09:35.920 --> 09:43.100 So Escalon provides the facility to calculate the confusion matrix using the confusion matrix method. 09:43.120 --> 09:45.160 So that is what we will be using here. 09:45.940 --> 09:48.240 So what is the confusion matrix? 09:48.250 --> 09:52.210 We are importing the confusion matrix using Escalon, dot matrix. 09:52.450 --> 10:01.390 Remember Escalon, Dot Matrix holds all the different types of matrix which we can use for evaluating 10:01.390 --> 10:02.310 our models. 10:02.560 --> 10:07.990 So always keep exploring these metrics and comparing different matrix. 10:08.230 --> 10:15.910 Use a lot of matrix and compare the models using these matrix and see which metric actually works well 10:15.910 --> 10:16.270 for you. 10:19.140 --> 10:21.180 Then let us give the prediction. 10:21.230 --> 10:29.220 So far, the predictions we are simply using can not predict on X this data and then we have the predictions 10:29.220 --> 10:30.310 from the best data. 10:30.700 --> 10:39.610 Now, we already have the very values and now we have generated divided values by using can and predict. 10:39.960 --> 10:43.800 So we are predicting the values on the X test values. 10:44.400 --> 10:51.870 We already have X values, so we predict the values for these X best values using the model which we 10:51.870 --> 10:55.560 have generated and we get the predicted values. 10:55.980 --> 11:05.400 Now we compare these Y predicted values with the widest value which we already have and in the confusion 11:05.400 --> 11:06.060 matrix. 11:07.990 --> 11:12.280 So this confusion matrix give these, Eddie. 11:13.560 --> 11:17.100 Which is basically true, negative as one sixty five. 11:18.460 --> 11:20.920 Proof positive as 60. 11:22.120 --> 11:28.120 False positive as 46 and false negative as forty seven. 11:30.550 --> 11:40.090 We can also obtain the confusion matrix using the crosstab method, so we can simply say we don't crosstab 11:40.870 --> 11:51.100 and provide Vytas here, then why predict in the role we will give the true value and in the column 11:51.250 --> 11:53.200 we will provide the predicted value. 11:53.980 --> 11:56.500 So this way we will get predicted. 11:57.690 --> 12:08.760 Zero one zero zero one, so Ventoux is zero, it is false, Ventoux is one, it is true I. 12:10.290 --> 12:17.090 This is true, this is false, and so we can decide it accordingly. 12:20.150 --> 12:28.640 So we have classification report, which is another matter, which is a textual summary of the precision 12:28.640 --> 12:36.890 recall, if one school for each and every class so we can use it on imported from Eskil on dogmatics. 12:37.980 --> 12:45.900 Now we will generate declassification report on Vytenis and predict now here you can see that the decision 12:46.350 --> 12:52.950 is zero point seven eight four nonminority and zero point six to four day. 12:52.950 --> 12:57.360 But the goal value is zero point eight to one zero point five six. 12:57.690 --> 13:04.380 If one score is zero point zero zero point five, me, I'm supposed value is two hundred and one and 13:04.380 --> 13:05.370 one hundred and seven. 13:07.460 --> 13:16.190 The accuracy at micro level on four weighted average of seventy and seventy three, so these are different 13:16.190 --> 13:16.760 values. 13:17.770 --> 13:27.640 Now we can find the orosco also for this, so what is Orosco again, the sequel is the plot, which 13:27.640 --> 13:32.800 is, if true, positive rate with respect to the false positive rate. 13:33.010 --> 13:38.180 We have discussed this during the time when we discussed the logistic regression. 13:38.890 --> 13:45.820 So it is a plot of the true positive rate against the false positive rate for the different possible 13:46.090 --> 13:51.860 points for a diagnostic test and auto cycle demonstrates several things. 13:52.180 --> 13:53.260 What are those things? 13:53.500 --> 14:01.390 Number one, the rate of between sensitivity and specificity, that is any increase in sensitivity will 14:01.390 --> 14:05.380 be accompanied by a decrease in the specificity. 14:07.530 --> 14:17.790 And the closer the Gulf follows the left hand border and then the border of the SP's, the more accurate 14:17.790 --> 14:19.950 the best is. 14:21.320 --> 14:28.250 Now, the closer the Gulf comes to the 45 degree diagonal, the less accurate the values. 14:29.270 --> 14:33.170 So we want to move it towards the top left corner. 14:33.830 --> 14:37.970 Now, the area under the Gulf is a measure of the accuracy. 14:41.310 --> 14:48.800 What is now we will be making the predictions, we will be predicting the probability using the excess 14:48.840 --> 14:52.680 data and we get the probability to vibrate from. 14:53.820 --> 15:01.130 Now, let us just read the sequel, so again, we will import the sequel from the Eskil or Not Matrix. 15:02.210 --> 15:03.800 Now, here we will have the. 15:05.150 --> 15:10.630 Fear that this false positive debate, the true positive rate and the threshold value. 15:11.810 --> 15:20.480 So we will generate the sequel for Vytenis with the with companies in Dubai predicted probabilities 15:21.290 --> 15:24.080 and we will create the plot for the scene. 15:24.980 --> 15:27.820 So this is the plot which we have created. 15:28.010 --> 15:34.210 And here you can see that the these values are what are good enough. 15:34.220 --> 15:42.410 So this is one value, which is which is a nice value where we have it close to the left side also and 15:42.420 --> 15:43.510 towards the upside. 15:46.600 --> 15:52.460 Now, if you see the area of the Gulf, we can find it using you for. 15:53.470 --> 15:56.790 But just coming out with zero point seventy four five. 15:58.740 --> 16:07.050 Now, this is one method which is using this brain split, but we have used another method, which is 16:07.050 --> 16:09.730 a very good method that is cross-validation. 16:09.960 --> 16:13.260 So let us implement this using cross-validation as well. 16:14.010 --> 16:15.400 So how would we do that? 16:15.840 --> 16:20.730 So we will get the hyper barometer's and we will use the cross-validation. 16:21.000 --> 16:23.400 Now, again, what is false validation? 16:23.400 --> 16:27.690 The PRIMORDIA performance is dependent on how the device splitted. 16:27.990 --> 16:35.430 So instead of using the hold out method, we will use the cross-validation what is called validation. 16:35.640 --> 16:42.150 Cross-validation is a technique to evaluate predictive models by partitioning the original sample into 16:42.150 --> 16:47.400 a training set to train the model and test set to evaluate it. 16:47.760 --> 16:55.320 So we will create key features and then add every time each one will be selected as a distinct field 16:55.320 --> 16:58.800 and other fields will be selected as the training fields. 16:58.950 --> 17:01.230 And that is how we will get those values. 17:01.230 --> 17:05.730 And then we can average them out, then find the final validation. 17:07.060 --> 17:15.670 So we will be doing hypovolemic for this now, we have already selected the value of, but now we want 17:15.670 --> 17:20.320 to find out the optimal value of using hypovolemic. 17:23.130 --> 17:30.030 So for this, we will try some different hypovolemic, those values and then sort them all separately 17:30.030 --> 17:34.440 to the model, and then we will choose the best one out of this. 17:34.560 --> 17:39.660 So we will use TV for this, which we have already built on great TV. 17:40.080 --> 17:41.450 So how does it work? 17:42.450 --> 17:49.380 We will create the bottom three in this forum, great, we are giving the neighbors as the barometer 17:49.620 --> 17:53.370 and the values are ranging from one to 50. 17:55.300 --> 18:04.150 Now we are creating an object of canibus classifier and we are creating an object of great TV which 18:04.150 --> 18:12.550 has the model, which is Ganin and the Bottom Great and the C.V, that is the cross validation for number. 18:12.730 --> 18:14.200 That is five foot. 18:16.050 --> 18:20.330 Now, the feeding the data into it, so we have it X and Y it. 18:22.270 --> 18:27.190 After four X and Y, we have found out the best score. 18:28.410 --> 18:36.630 Is zero point seven five seven eight and the best barometer is anybody's guess 40. 18:37.410 --> 18:45.300 So this is what we have or been that is we have found almost seventy six percent accuracy using the 18:45.300 --> 18:47.130 neighbors numbers 14. 18:47.340 --> 18:50.470 And what is really what we have found out was seven. 18:50.730 --> 18:59.280 So here we have improved the accuracy by almost three percent using the grid to a TV and by selecting 18:59.280 --> 19:03.210 the higher number of neighbors. 19:04.340 --> 19:08.710 So next, we will learn about the next algorithm in the next session.