WEBVTT 00:02.290 --> 00:05.330 So let us have a look at the solution on this particular problem. 00:05.740 --> 00:07.030 So here we are. 00:07.060 --> 00:15.400 First of all, important, the important libraries and we have also got the data set into this data 00:15.400 --> 00:15.760 frame. 00:16.000 --> 00:21.340 And this is the data that we have, the satisfaction level, last evaluation number of project abridgment, 00:21.340 --> 00:26.950 the time spent in the company, Wilcockson, live promotion and all of the goals which we just discussed 00:26.950 --> 00:28.600 in the last review. 00:30.430 --> 00:35.620 So let us suppose to have a look at the value column so that we can see what is the distribution of 00:35.620 --> 00:36.070 our target. 00:37.180 --> 00:43.990 So our target value is having value zero and one zero means that person has not left the company, and 00:43.990 --> 00:46.040 one means that the person has left the company. 00:46.420 --> 00:54.250 And here you can clearly see that there are on seven thousand doors, which has that person does not 00:54.250 --> 00:57.550 leave the company, but three thousand rooms where the person needs the company. 00:59.500 --> 01:06.910 So when we look at the issue, it is two point four, which implies that there is a huge difference 01:06.910 --> 01:07.640 between these. 01:07.640 --> 01:12.640 So probably we will have to have a weighted classification. 01:13.240 --> 01:18.610 So next we will have a look at different columns. 01:18.640 --> 01:23.950 So there are only two columns which have objected to all of the non numerical columns. 01:24.190 --> 01:29.540 So the objective, they are the sales and salary. 01:30.910 --> 01:35.680 So next thing which we will be doing is on what is in the military. 01:36.880 --> 01:40.600 So this is the same code which you have seen several times. 01:40.780 --> 01:45.000 So we are converting this into a dummy variable. 01:45.010 --> 01:48.850 So we have converted into inside the column into dummy variable. 01:50.620 --> 01:58.360 Now, when we have a look at the shape of this particular data, we there are in Saudi in columns and 01:58.450 --> 02:00.790 around 10000 rows of it. 02:02.890 --> 02:15.250 Next, we are getting the data in two extreme and widely by dropping the left column from D.C.I and 02:15.820 --> 02:20.190 keeping only the left column from here in the extreme right. 02:23.250 --> 02:29.340 Next is just simply a code which will give you a report of the model, which we have just run. 02:29.760 --> 02:37.050 And next, what we are doing here is we are learning different parameters, which we have, for example. 02:37.080 --> 02:40.080 So here I am simply implementing X Eustachy. 02:40.440 --> 02:44.620 You will be implementing different algorithms and comparing amongst them. 02:45.030 --> 02:47.130 So this is the process which we will be following. 02:47.140 --> 02:54.930 You will be implementing different algorithms of logistic regression decision trees, random 460 most 02:55.620 --> 02:57.300 of my ways as. 02:57.720 --> 03:03.650 You will try all of those and then find out which works best on this particular dataset. 03:04.050 --> 03:07.920 And after that, you will be fine tuning that particular model. 03:07.920 --> 03:12.640 If it still does not perform, then you will be stacking different models. 03:14.490 --> 03:19.270 So these are different type of parameters which are present in extreme boost. 03:20.340 --> 03:28.500 What I have done here is I have first of all, implemented a randomized so Steube you'll find out how 03:28.500 --> 03:36.870 my model would actually perform and I have completed my to within so that out of so many combinations 03:36.870 --> 03:46.410 or nithin combinations will be created and I will be getting the model performance out of only this 03:46.780 --> 03:47.640 Benwood. 03:49.080 --> 03:53.760 So I'm simplifying the model and got the report out of the report. 03:53.760 --> 03:58.290 States that I have eighty three percent accuracy here. 03:59.220 --> 04:02.630 Next is eighty three point seven three point six. 04:02.910 --> 04:05.700 So the best one comes out to be three point seven. 04:05.700 --> 04:12.690 And these are the barometers which I have obtained from this now to finding this particular model, 04:12.690 --> 04:16.600 I will be applying these sequential parameter. 04:17.940 --> 04:24.410 So what I have done here is I have selected one parameter at a time or two or three parameters, ultrafine, 04:24.770 --> 04:28.680 and then I am simply fine-tuning each and every parameters like that. 04:29.160 --> 04:32.460 So first, I have been the number of investigators. 04:32.850 --> 04:36.100 I have seen the model and Gwatney Disvalue regarding that. 04:36.120 --> 04:40.470 So once they get the model, I get then the end estimate of nine hundred looks. 04:40.470 --> 04:43.960 But even then an estimated one hundred also looks very big. 04:45.060 --> 04:51.240 So what we do is we select something which we find is useful enough. 04:51.240 --> 04:55.350 So that is not much difference between the an estimate estimated. 04:55.350 --> 04:56.600 Nine hundred and a hundred. 04:56.910 --> 05:02.580 So we are thinking 500 because that is a good number for investigators. 05:02.580 --> 05:02.790 Right. 05:02.840 --> 05:03.630 We will be fine. 05:04.620 --> 05:08.980 So what we do for that is we will make the learning rate again. 05:09.000 --> 05:17.850 And here I have just kept morning zero point one and I will be picking of the gamein max depth value 05:17.850 --> 05:20.490 and fine tuning what these two parameters. 05:20.940 --> 05:29.280 So yeah, I have fixed the learning rate and we did and subsamples and I am fine tuning on top of it. 05:29.670 --> 05:33.630 And then I get these values regarding the parameters. 05:33.630 --> 05:35.520 I get the max to be 12. 05:35.520 --> 05:37.950 I'm gonna do it next again. 05:37.950 --> 05:39.660 I will fix these two values. 05:39.660 --> 05:42.480 I'm fine doing some other parameters and so on. 05:42.480 --> 05:44.050 This cycle would keep on going. 05:44.340 --> 05:46.500 So this is something that you will have to work own. 05:47.880 --> 05:55.950 And finally, when I keep on doing this entire process again and again, I end up with various values 05:55.950 --> 05:59.220 for these hyper parameters, which I keep fixing. 06:00.090 --> 06:04.420 And at the end I am left with an optimized model. 06:04.440 --> 06:09.090 So here you can see the mean validation is eighty four point one. 06:09.930 --> 06:14.910 When I look at the next one here, again, the mean validation is eighty four point one with standard 06:14.910 --> 06:17.520 deviation zero point zero two one. 06:20.040 --> 06:27.150 If we compare this from the previous one here, we have zero point equal one with the meanwhile standard 06:27.150 --> 06:29.750 deviation of zero point zero one seven. 06:31.290 --> 06:38.170 Here we have zero point eight four zero with standard deviation of zero point zero zero. 06:38.370 --> 06:44.580 So if you compare between these different models, you can pick out any model out of these where we 06:44.580 --> 06:52.740 have this eighty four point one, which seems to be a better one and select that particular model next. 06:52.750 --> 06:57.540 What I'm doing is I am simply getting the best model out of it. 06:58.080 --> 07:03.400 And from the best model, I will simply find out the cross validation score. 07:03.750 --> 07:09.960 So these are different cross-validation scores for this particular model and I am getting the mean and 07:09.960 --> 07:16.110 standard deviation out of to the mean that a school comes out with three point sixty six and the standard 07:16.110 --> 07:18.590 deviation comes out to be zero point one one. 07:19.020 --> 07:21.480 So this is my model. 07:21.690 --> 07:29.490 I have for you did you can try a different strategy, different importance of them together and create 07:29.490 --> 07:32.880 your own model of what this looks like, a good base for defense.