WEBVTT 00:01.200 --> 00:08.280 In this particular session, we will discuss about the implementation of decision to use this court 00:08.280 --> 00:14.820 file contains the implementation for random forest also, but that is something which we will be covering 00:15.060 --> 00:19.360 once we have already discussed victory for random forest. 00:20.490 --> 00:25.290 So the first thing what we do is we will import the required libraries. 00:25.500 --> 00:32.040 That is, find us as number as in the Escalon as. 00:33.630 --> 00:36.310 Three, we will import trees from Escala. 00:36.660 --> 00:39.970 We will import numbers, and B, this is not required. 00:39.990 --> 00:43.880 It's again then we have Escalon metric. 00:44.310 --> 00:53.940 Eskil on that metric contains a lot of metrics which we can use for analyzing and for evaluating our 00:53.940 --> 00:54.470 models. 00:55.580 --> 01:04.580 So I would suggest you to go to Eskil on Dot Matrix and check the documentation from Escalon and see 01:04.580 --> 01:13.820 what metric is used for what purpose, I will do a separate video for different types of matrix, so 01:13.820 --> 01:15.030 don't worry about that. 01:15.500 --> 01:24.110 So for now, we will import a U.S. score and we will be making in mudflat clip in light, which means 01:24.110 --> 01:25.010 that we want to. 01:27.110 --> 01:31.400 Pretty much a plot in this before. 01:32.490 --> 01:41.460 So once we run this, we will import the CIA tree so the CIA train data will be coming from this particular 01:41.480 --> 01:41.820 fight. 01:44.270 --> 01:46.070 So we will import the five. 01:47.280 --> 01:52.350 Census info, and we will read this file using. 01:53.470 --> 01:55.300 They did not read CSFI. 01:56.370 --> 02:03.960 Now to explain the fine can be placed in any folder, but currently this file is placed in the same 02:03.960 --> 02:12.180 folder as this Jupiter notebook, which is the reason why I have simply given the file name in case 02:12.180 --> 02:15.000 the file was placed in any other directory or folder. 02:15.180 --> 02:18.570 Then I would have said slash data. 02:19.170 --> 02:21.600 Slash some folder name. 02:23.430 --> 02:34.200 And so on for the topic, and this roistering reader will actually help in understanding the characters, 02:34.200 --> 02:42.340 these data in the raw form so that these values are not rendered some other character. 02:43.200 --> 02:48.800 So we will import this and we will import data from us three. 02:50.480 --> 02:58.490 Now, in case you have a test data, then you can combine the best data as earlier in the earlier model, 02:58.490 --> 02:59.670 as we have shown. 02:59.840 --> 03:10.120 So what you can simply do is you can combine the data in the same data frame and from the data frame 03:10.130 --> 03:12.790 itself, you can give it a name. 03:12.950 --> 03:15.080 So we'll see then how we can do that. 03:17.030 --> 03:20.300 So far now we have this data c.i.t.. 03:21.180 --> 03:28.680 And we have got hit, so we are checking the details which are present in this particular data frame, 03:29.040 --> 03:33.550 so we have age, which is numeric in nature. 03:33.930 --> 03:36.360 Then we have the will class. 03:37.740 --> 03:43.540 Class contains values like Steve Goman, self-employed, so what is the class? 03:44.100 --> 03:50.670 It belongs to if someone is doing a government job or a private job or if he's self-employed. 03:50.680 --> 03:55.970 So all these details are present in the middle class, then we have final vote. 03:56.460 --> 04:02.280 We don't really know what this value is the big thing for, but we have this numeric value here. 04:03.120 --> 04:04.660 Next is education. 04:05.010 --> 04:11.990 Now, this education value has bachelor's bachelors, high school graduate, 11 bachelors. 04:12.150 --> 04:15.000 So these are different values which are present in the education. 04:15.300 --> 04:20.610 And in decide this, we have another column that is education. 04:20.610 --> 04:21.210 No. 04:22.550 --> 04:28.760 This education number, again, has values 13, 13, nine, seven, 30, so. 04:30.340 --> 04:36.490 It seems to be that these are numeric codes for this one, but we will evaluate that little. 04:37.650 --> 04:46.110 Next, we have marital status, there are multiple statuses like never married marriage, civic spouse, 04:46.440 --> 04:50.670 then divorced, then we have occupations. 04:50.670 --> 04:52.350 These occupations are. 04:53.650 --> 05:05.470 And local executive managerial handlers, cleaners, profession speciality, then we have a relationship 05:05.470 --> 05:13.160 that is someone is not in the family or someone has a husband and if someone has a wife. 05:13.450 --> 05:16.530 So these are different relationship status of the people. 05:18.060 --> 05:25.140 Next, we have the breeze, which is either someone is white, black, Asian, whatever it is, then 05:25.140 --> 05:28.840 we have sex that is male, female, these values. 05:29.430 --> 05:37.560 Next, we have values like capital gain, capital loss, then hours per week, how many hours per week 05:37.560 --> 05:41.390 a person walks, then the native country for that person. 05:41.670 --> 05:45.060 And finally, we have value, which we want to predict. 05:45.300 --> 05:51.630 That is if the person is earning less than equal or greater than 50. 05:53.470 --> 05:58.840 So from this data set, you can see that why is the value which we want to predict? 05:59.770 --> 06:08.800 And the rest of the columns are actually the X values, which we want to use as the features or attributes 06:09.070 --> 06:12.750 or input values or independent values, you can see. 06:13.060 --> 06:20.170 And why is the label or target or you can also say dependent variable. 06:20.200 --> 06:22.870 So these are different films which we use for the white. 06:24.530 --> 06:29.930 Now we can look at the head value for the train dataset. 06:30.320 --> 06:36.560 So this is the train data center, which is, again, just to see now what we can do is. 06:39.360 --> 06:46.560 Here I am showing the values which have been created after the transformation. 06:46.860 --> 06:53.860 Now you can see the values have been converted into numeric values, which is the target here. 06:54.120 --> 07:01.830 So our main target is to convert all the data into all the data instead of it being the. 07:03.140 --> 07:05.210 Categorically or any data. 07:05.330 --> 07:11.090 So all of this data has to be converted into numeric data right now, what all can we do to convert 07:11.090 --> 07:12.370 this into unlimited data? 07:12.380 --> 07:16.510 Because the algorithm cannot really understand these words. 07:16.910 --> 07:22.500 The algorithm does not know what BATCHELLER means and what private means, what has been means, what 07:22.640 --> 07:23.130 it means. 07:23.300 --> 07:26.780 So all these things are nothing for the algorithm. 07:26.810 --> 07:31.130 It will only understand what thirty nine zettl one these numbers are. 07:31.160 --> 07:32.930 So it will only understand the numbers. 07:33.110 --> 07:38.840 So our main focus here is to convert all this data into numeric form. 07:38.890 --> 07:40.490 So now how will we do that? 07:41.690 --> 07:48.290 So let us have a look at each and every column, so the first column which we have is each now this 07:48.290 --> 07:50.660 column is already in a numeric form. 07:51.500 --> 07:53.690 Next, we have a class. 07:53.900 --> 07:57.950 So the world class is in a categorical form. 07:58.100 --> 08:02.380 So we will have to convert these categories into numbers. 08:02.480 --> 08:03.830 And how do we do that? 08:03.830 --> 08:08.600 We can avoid that by using one word, encoding or dumbing creation. 08:10.120 --> 08:16.630 Next, we have finally this again, looks like a numeric variable, but in case it is not, we will 08:16.630 --> 08:22.450 convert it into a numeric by error course. 08:23.110 --> 08:24.810 Next, we have education. 08:25.000 --> 08:28.680 Now, again, it looks like a categorical variable. 08:28.960 --> 08:33.850 So we will see what all categories are present here and here. 08:33.850 --> 08:35.530 We have education number. 08:35.740 --> 08:41.830 So we will try to compare the relationship between education and education number and find out if these 08:41.830 --> 08:42.930 are related or not. 08:44.630 --> 08:51.920 Next, we have marital status, so this, again, looks like a good column, so we will have to convert 08:51.920 --> 08:58.430 this into one two, including next we have occupation, which again would have to be converted into 08:58.430 --> 08:58.650 one. 08:58.830 --> 09:07.140 The same applies to columns like relationship, race, sex, then the column Native Country. 09:07.280 --> 09:13.100 So all of these columns and column have to be converted into a dummy form. 09:14.730 --> 09:21.730 So now let us do one thing, so let us apply Findus profiling to it. 09:21.800 --> 09:29.610 So I have just applied Fondas profiling by importing profiling report from Findus profiling, and I 09:29.610 --> 09:35.590 have generated the report using fear and putting Citrine in it. 09:35.880 --> 09:39.460 So this is the profile report which I have obtained from this. 09:39.750 --> 09:45.330 So here you can see that the number of variables on 15 numeric variables are six. 09:45.330 --> 09:47.070 Categorical variables are nine. 09:48.490 --> 09:58.570 Out of this, the data set has twenty four duplicate laws and capital gain has ninety one percent zeros 09:58.570 --> 10:01.510 on capital loss, has ninety five percent zeros. 10:02.520 --> 10:06.270 Next, you can see the type of values which are present in each. 10:07.140 --> 10:09.540 Here is the detail of the capital gain. 10:09.570 --> 10:14.130 Here is the videos of the capital loss for education. 10:16.610 --> 10:28.400 You can see that the top four categories contain most of the data, while these categories contain less 10:28.400 --> 10:30.060 than five percent of the data. 10:30.380 --> 10:37.270 So we can decide if we should keep the top five categories on the order of four categories only and 10:37.640 --> 10:39.590 we can remove the other categories. 10:39.740 --> 10:47.540 So for this, we can actually run different variants, one variant with the top categories and another 10:47.540 --> 10:54.500 variant with all the categories present and lifting, which we will be learning, is getting the feature 10:54.500 --> 10:55.230 importance. 10:55.430 --> 11:03.890 So we were able to get future importance from the logistic regression also so we can use that and obtain 11:03.890 --> 11:07.800 the of important features from that as well. 11:09.190 --> 11:13.190 So next, what we have is these days. 11:13.210 --> 11:15.790 So, again, we have education, No. 11:17.480 --> 11:20.960 Here we have the final eight hours per week. 11:23.050 --> 11:25.330 You can see there are four different values. 11:26.720 --> 11:36.080 So here are different values, which are married, spouse, never married, divorced, then separated, 11:36.290 --> 11:40.240 then widowed, battered spouse absent, married if spouse. 11:40.490 --> 11:48.770 So out of all of these, we can clearly select only the top three categories and keep other categories 11:48.770 --> 11:49.610 in others. 11:51.860 --> 12:00.090 Next, we have these values where the major people are actually from United States. 12:00.350 --> 12:06.080 And next, there are a lot of countries for other people. 12:06.290 --> 12:15.260 So we can keep the first category that is United States and make under the category as non United States. 12:17.820 --> 12:27.870 Next, we have occupation, so for occupation, you can see that we have one, two, three, four, 12:27.870 --> 12:36.240 five, six, seven, eight, eight categories which are having more than five percent of the data so 12:36.240 --> 12:42.240 we can decide upon if we want to keep the rest of the categories or if we can put them into other categories 12:42.480 --> 12:45.110 or should we keep those categories or not. 12:49.450 --> 12:57.850 Then we have this, so for this again, we have eighty five percent white, ninety nine point six percent 12:57.850 --> 13:00.080 black and three point two percent Asian. 13:00.310 --> 13:07.720 So what we can do is we can simply have three categories, one being white, another being black, and 13:07.720 --> 13:11.590 the rest of the values could be put into other categories. 13:13.700 --> 13:21.890 Then we have a relationship, so in relationship, again, as these four seem to be the major categories, 13:22.100 --> 13:28.580 we can create these four major categories and rest we can put in the other categories. 13:31.720 --> 13:36.380 Or as there is only one category which would be left, we can also keep all of these. 13:36.670 --> 13:39.190 So that is all we can mean the means for this. 13:41.040 --> 13:48.240 Then because sex has only two values, so we can easily create one variable for that. 13:49.350 --> 13:54.540 Next, we have world class, so for this world class, again, we have. 13:56.100 --> 14:01.920 Around six prominent categories, and apart from that, all could be put into others. 14:03.910 --> 14:12.640 And then here we can see that for the very values, less than Feki is almost twenty four thousand and 14:12.640 --> 14:15.410 the greater than fifty seven thousand values. 14:15.640 --> 14:20.830 So here you can see that the ratio is quite different. 14:20.860 --> 14:27.580 So that is almost seventy six, forcing all values into one class and twenty four percent values in 14:27.580 --> 14:28.450 the other class. 14:28.660 --> 14:35.770 So when we have such kind of data, the one class has more values on, the other has less number of 14:35.770 --> 14:36.320 values. 14:36.580 --> 14:39.940 This means that the classes are unbalanced. 14:40.720 --> 14:49.420 So when they have such unbalanced classes, then what can happen is that because there is more number 14:49.420 --> 14:58.030 of values having less than four Feki, then the training might get biased towards less than 50. 14:58.570 --> 15:07.450 And for all the values, it might start of predicting that the value is actually less than 50. 15:08.940 --> 15:16.770 So what we need to do here is instead of keeping these values like this, we will be using a great idea 15:16.770 --> 15:26.190 that this glass beads the glass, which will actually help us in a way that they will allow us to give 15:26.190 --> 15:29.190 equal weight age for this particular class. 15:30.460 --> 15:35.880 So that it is not getting overpowered by the other guys. 15:36.040 --> 15:41.530 It will give equal weight to the values of this class also in comparison to this class. 15:43.010 --> 15:52.500 Next is the correlation values, so as there are no specific columns which are very dark in Cologne, 15:52.970 --> 16:00.230 neither in the right side nor in blue side, so we can easily see that there is no specific correlation. 16:00.590 --> 16:04.590 And the same thing has been depicted in the above. 16:04.610 --> 16:10.930 There was no specific correlation depicted here and there were no columns which were rejected. 16:11.180 --> 16:15.680 So we are going to go with these columns and these details. 16:17.630 --> 16:24.710 So the next thing which we will be doing is we will be comparing the education problem with the education, 16:24.710 --> 16:25.550 no problem. 16:26.660 --> 16:37.570 So here you can see that the values for education and education number are actually correlated. 16:37.760 --> 16:48.950 That is that first and fourth actually refers to the education level one, level two five to six level 16:49.610 --> 16:51.520 is basically education. 16:51.530 --> 16:54.980 Number three, seven to eight is education number four. 16:55.220 --> 16:56.600 Ninth is education. 16:56.600 --> 16:59.150 Number five, the intent is education. 16:59.150 --> 17:01.460 Number six 11 is education. 17:01.460 --> 17:05.660 Number seven, 12 is education, number eight and so on. 17:07.370 --> 17:15.860 So what we can do is we can simply remove the column education and keep only education number in our 17:15.860 --> 17:16.580 dataset. 17:18.220 --> 17:26.200 So we will simply Skytrain ball drop and remove the education column from this particular dataset. 17:27.070 --> 17:30.410 Next, we will get the count of the value. 17:30.470 --> 17:36.120 So we already know that the count of the values is actually unbalanced in nature. 17:36.340 --> 17:41.870 So we will have to take that into consideration by training our model. 17:42.160 --> 17:48.260 But for now, what we will be doing is we will convert the CIA train by column into a dummy column. 17:48.520 --> 17:56.590 So for that we will simply use ITIN VI is equal to greater than 50. 17:56.600 --> 18:01.560 So whenever the value is greater than 50, it will put that as one. 18:01.780 --> 18:06.380 And whenever the condition is not satisfied, if that does zero. 18:08.420 --> 18:09.770 Because this is a condition. 18:10.960 --> 18:17.500 This condition will be evaluated, and when this condition is evaluated, it will give one false. 18:18.070 --> 18:20.350 So when this condition is true, it will give. 18:21.480 --> 18:28.800 True and true, Wengen voted to impeach and gives one and false is converted to invasion, it will give 18:29.160 --> 18:29.510 them. 18:32.200 --> 18:39.660 Next, we are getting all the columns, which I've got the vertical columns by using Katrien to select 18:39.670 --> 18:40.310 the types. 18:40.590 --> 18:44.020 So this will give us all the data, such as object data. 18:44.560 --> 18:47.020 And then we are selecting the columns out perfect. 18:47.230 --> 18:49.530 So we get all the categorical columns. 18:49.690 --> 18:54.280 So these are the different categories of columns which are present in this particular dataset. 18:56.170 --> 19:04.270 Next, we will we have created this particular law, which will actually allow us to create a cutoff 19:04.660 --> 19:14.470 for the column creation so that we can actually obtain only some amount of categorical columns, only 19:14.470 --> 19:19.690 some categories converted into the dummy columns and not all of them. 19:21.610 --> 19:27.820 So it is only new thing on top of the different categorical columns and. 19:28.970 --> 19:36.800 Getting the values of columns and then it is running on the citrine, I'm getting the different value 19:36.800 --> 19:41.090 counts and these value columns, I get best frequency's. 19:41.780 --> 19:49.970 Then from the frequency data, it is checking if the index of the it is getting the indexes of the frequency 19:49.970 --> 19:57.750 is the frequency value is greater than five hundred and it is checking the category in. 19:57.780 --> 20:00.800 And so it is just creating a name. 20:00.800 --> 20:04.430 That is the column plus the category name. 20:05.650 --> 20:12.830 And it is creating the column name using this, so what it does is it just creates the got the got a 20:12.830 --> 20:15.010 good column out of the columns, which we have. 20:15.010 --> 20:22.990 So we get we have one plus column marcosi, this occupation, relationship, race, sex and native country. 20:24.370 --> 20:33.850 So for all of these columns, these have been converted into a categorical two or three to one, including 20:33.850 --> 20:34.960 two Dubnyk lists. 20:37.130 --> 20:42.780 Next, we are just checking the shape of the data now, so now we have thirty nine columns in hand. 20:43.040 --> 20:51.200 So the next task, after converting all the columns from can go to the column to the columns or whatnot, 20:51.200 --> 20:54.630 including yours, to check if there is any value or not. 20:54.800 --> 21:01.560 So we will check the data by using DOT is not a dot dorsum, so we get all the column values. 21:01.580 --> 21:05.220 So here we can see that all the columns are actually Ziegel. 21:05.450 --> 21:09.020 And here you can see that these are the columns which have been created. 21:09.020 --> 21:16.100 So the relationship value hundred husband relationship, not in a family relationship or child relationship. 21:16.760 --> 21:20.200 Unmetered Relationship five, this white is black. 21:20.510 --> 21:24.680 So these are different columns which has been created using the dummy creation. 21:26.130 --> 21:28.260 Now we are simply getting the data. 21:29.420 --> 21:38.960 Into extreme, so extreme contains all the problems from CIA except for the vehicle, so that is why 21:38.960 --> 21:42.710 we drop the big column from the axis equal to one. 21:43.880 --> 21:45.410 We can ride this lake. 21:46.570 --> 21:50.710 Axis equal to one or two, it'll be just the same thing. 21:51.100 --> 21:52.840 Next, we have vitrine. 21:54.100 --> 22:02.140 So we are putting the V column into right now, there are different type of barometers for Decision 22:02.140 --> 22:05.390 B, which we have already discussed about. 22:05.620 --> 22:14.100 So we will be training for these different hyper barometers and these on these different hypovolemic 22:14.140 --> 22:20.290 as we will be getting the values of the hyper barometers which give us the best model. 22:22.690 --> 22:32.440 So we are implementing the randomized so TV, so we have already discussed about the grid search now 22:32.440 --> 22:41.860 in a grid search c.v, what happens is that the grid search will give us the combinations. 22:41.860 --> 22:45.130 It will run all the combinations which are pressing everything. 22:45.380 --> 22:50.380 And then from all the combinations, it will select the best one. 22:50.560 --> 22:53.480 It will run all the combinations from the bottom. 22:54.040 --> 22:59.980 So if I'm giving these barometers, it will run all the combinations out of it and then give me the 22:59.980 --> 23:08.530 result, like what random solar TV does is that because we don't want to put in so much time into running 23:08.530 --> 23:08.990 the code. 23:09.190 --> 23:15.000 So what we will do is we instead of using the digital TV, we will use the random thoughts. 23:15.430 --> 23:22.300 So what it will do is let's say we have one hundred five orders which are being created using grid TV. 23:22.450 --> 23:30.250 So instead of running those 100 models, it will randomly select 10 or 20 models, whatever number we 23:30.250 --> 23:32.440 define and run those. 23:33.760 --> 23:40.510 So in this way, we will actually run a random selection from those markets. 23:40.810 --> 23:44.950 Now, the benefit is that it will save us a lot of. 23:46.020 --> 23:53.850 But there is another drawback that is we will be losing on a particular one. 23:54.390 --> 24:02.100 So there might be some model which would have been better than the one which we will receive by running 24:02.120 --> 24:03.000 the random lines. 24:03.000 --> 24:11.370 So see why we would have used the TV, but because we want to see if that is why we are using randomizer 24:11.370 --> 24:11.800 TV. 24:12.030 --> 24:18.630 This is good when you are running for sartin proof of concept or you want to get a model very quickly, 24:18.840 --> 24:21.100 then you can use my TV. 24:21.420 --> 24:30.080 But otherwise if you want to do a quick thing but actually run extensively and get the best model. 24:30.300 --> 24:32.930 So in that case you would run the grid. 24:32.940 --> 24:42.360 So it would be and there are different methods which will actually ease the running of TV, but that 24:42.360 --> 24:45.750 is something which we will learn in something later. 24:45.900 --> 24:52.370 Maybe in another topic because of the I don't want to make it really fast so that you get confused. 24:52.560 --> 24:58.050 So for now, we will have a look at this randomizer TV. 24:58.200 --> 25:04.950 So if you want to or don't really list any time, I just want to have a one week run. 25:05.220 --> 25:13.020 Then you can use a screen split in case you have a moderate amount of time. 25:13.020 --> 25:14.670 You can use randomized. 25:14.970 --> 25:15.840 So TV. 25:16.140 --> 25:23.850 But if you want to run the phone extensively and try all the possible combinations, in that case you 25:23.850 --> 25:25.280 will use the grid. 25:25.300 --> 25:25.730 So it's. 25:27.330 --> 25:28.690 So let us go for the. 25:29.830 --> 25:30.660 So here. 25:32.170 --> 25:39.370 We have different parameters, so these are different barometers, which we have that this class meet 25:39.990 --> 25:46.330 criterion, maximum depth minimum, some believe minimum sample split. 25:46.590 --> 25:48.990 So these are different criteria, which we have. 25:49.180 --> 25:51.670 So we will be trying all the combinations. 25:53.820 --> 26:01.350 Next is another thing is like instead of having just one minimum belief, you could have had a larger 26:01.680 --> 26:02.960 size of criteria's. 26:03.210 --> 26:06.480 So this is another criteria which would work better. 26:06.720 --> 26:08.990 So it is a completely dependent on you. 26:09.000 --> 26:13.170 How many samples, how many parameters you want to put in? 26:13.890 --> 26:22.350 Usually what I would do is I will try of three values or four or five values, and out of those values 26:22.500 --> 26:24.180 I will select the best ones. 26:24.330 --> 26:25.520 So how would I do it? 26:26.040 --> 26:27.780 Minimum's, I believe values. 26:27.780 --> 26:30.870 So I'll try five then. 26:32.340 --> 26:41.740 15, 20 and 25 now out of these values, let us see, I get no value for theme. 26:42.990 --> 26:52.020 So in that case, I will eliminate five and twenty five and then I will add another value. 26:52.020 --> 26:54.120 Here I will add value 12. 26:55.140 --> 27:00.780 And Values 17 here, and I will eliminate the five. 27:02.110 --> 27:03.280 I'm twenty five. 27:05.650 --> 27:10.870 This way, I will actually get to know and I will then also. 27:12.590 --> 27:15.960 Because if it was actually then, then it would have chosen then. 27:16.670 --> 27:22.860 So now I have these three values now based on which in which direction this election goes. 27:23.060 --> 27:29.280 So if the best model comes out to be 15, then we are good in is the best model comes out. 27:29.390 --> 27:29.990 We do it. 27:30.620 --> 27:33.560 Then what we can do is then we can eliminate 70. 27:34.640 --> 27:41.240 And we can run something which is between 12 and 15, 13 and 14. 27:42.710 --> 27:44.510 Then we can run something like this. 27:45.500 --> 27:48.920 This way, we will be able to narrow down the selections, which we are. 27:49.640 --> 27:51.530 So this is how you can run this. 27:52.000 --> 28:00.380 So here what we are getting is we will be running through of the Glaspie to criterions. 28:01.380 --> 28:01.890 For. 28:02.980 --> 28:04.570 Maximum bet values. 28:06.470 --> 28:13.760 For three minimum, some beliefs and three minimum sample split, so the total number of runs which 28:13.760 --> 28:20.540 we will be having a number of models, which we will be having, will be to cross the cross for cross 28:20.540 --> 28:21.020 three. 28:22.180 --> 28:29.440 So in total, I have 144 murders and I'm a third thing is that this is a decision three and I will be 28:29.440 --> 28:31.980 running then cross-validation. 28:32.290 --> 28:33.960 So it will be in Dotan. 28:34.300 --> 28:34.750 So. 28:36.640 --> 28:43.870 They will be in total one thousand four hundred and forty more dollars, which will be created if I 28:43.870 --> 28:45.250 was running a grid. 28:45.250 --> 28:46.110 So a TV. 28:47.520 --> 28:55.170 But the good thing is I'm using randomise so TV, so now what will happen is first thing first I will 28:55.170 --> 28:59.670 import the Escalon tree that this decision to declassify. 28:59.700 --> 29:05.100 From this, I will create an object of decision to be classified as sealife. 29:05.370 --> 29:06.900 And in my randomise. 29:06.900 --> 29:14.520 So it's KVI, I will give the details that I want to run this cliff model that are in cross-validation, 29:14.580 --> 29:15.930 which I want to have. 29:16.200 --> 29:22.800 These are the parameters which I want to run this clip for, and this is the scoring method which I 29:22.800 --> 29:26.010 want to use a bathroom that I will give. 29:26.460 --> 29:38.420 And I do this and either means how many more those do I want to select from these 144 words? 29:39.120 --> 29:45.960 So if I say I want to select 10 models out of these 144 models, then it will select. 29:47.880 --> 29:54.240 Then more days and it will run then cross-validation from this, so it will give. 29:55.240 --> 30:01.900 Into then that is one hundred more, those will be built instead of one thousand four hundred and forty. 30:03.320 --> 30:13.720 So now you can understand how it is of easing the run, so instead of having such a huge run, then 30:13.760 --> 30:15.330 we will have a lesson one day. 30:15.590 --> 30:21.230 That is, we will have to run only a hundred more days instead of one thousand four hundred more. 30:21.950 --> 30:27.400 So running the randomizer, TV has its own pros and cons. 30:27.410 --> 30:34.190 So that is something that you can evaluate based on what kind of work you are doing and what exactly 30:34.190 --> 30:35.620 you want, what you promised. 30:37.050 --> 30:42.570 So the next thing which we will be doing is we have created the object of randomizer TV. 30:42.780 --> 30:46.560 So this is the object of random TV and we will find the. 30:47.520 --> 30:54.450 Extranet, vitrine, when we find the extreme and white rain, it will take some time to run because 30:54.450 --> 30:59.790 we have so many parameters ReGive running and there are around hundred more of those which are being 30:59.790 --> 31:01.030 created internally. 31:01.350 --> 31:03.120 So it will take some time. 31:03.120 --> 31:08.910 It will learn from this extreme NYG and after learning from this data. 31:10.550 --> 31:16.340 It will give the best estimate using random search talk, best estimate. 31:17.340 --> 31:26.490 Now this is the best model which we have received from this and this best model is having the class 31:26.510 --> 31:27.770 with as balanced. 31:27.980 --> 31:30.020 The criteria is Guiney. 31:30.230 --> 31:32.450 The maximum depth is five. 31:32.690 --> 31:34.970 The maximum number of features is none. 31:34.970 --> 31:36.660 The maximum Leaford is none. 31:36.830 --> 31:43.670 So these are different parameters which it has chosen from these group of barometers, which we have. 31:45.550 --> 31:54.490 Now, you can also go through the documentation of Decision three and see what all other hippopotami 31:54.490 --> 31:54.880 does. 31:54.910 --> 32:03.130 You can select and view this model for these are all called hyper barometers, which we are tuning. 32:03.640 --> 32:06.320 And this is exactly what you need to do. 32:06.340 --> 32:11.020 You need to have patience and give it the time to clean. 32:11.320 --> 32:15.290 And once you have the model in hand, you can make the predictions. 32:16.060 --> 32:20.890 So this is everything here is actually a task of patience. 32:20.900 --> 32:28.330 If you have enough patience and you know which type of barometer to choose and how to deal with these 32:28.330 --> 32:34.600 hyper barometers, how will check which value you need to use is exactly what you need to know here. 32:36.420 --> 32:46.350 Now, let's see if I would have had values, let us see five and then and 15 and somehow I would have 32:46.350 --> 32:52.750 got values 15, that means that the value is marginal value. 32:53.010 --> 32:57.110 Now, the value could have been less than 15 dollars, one greater than 15. 32:57.780 --> 33:05.780 So in that case, in the next run, in the next training, what I will do is I will give some value. 33:05.820 --> 33:15.750 See twenty five here and let's say 20 or so, so that I will get to know if actually the value was 15 33:15.960 --> 33:22.650 or it was something less than 15 or greater than 15, because I don't want to lose on something just 33:22.650 --> 33:27.940 because I didn't run the random search again or grid search again. 33:28.140 --> 33:34.830 So once you run this is randomized, so you will get some parameters, then you can modify the parameters 33:34.830 --> 33:41.690 and run these models individually or by changing these hypovolemic and running a grid search on top 33:41.700 --> 33:45.860 of these so that you can get a better result and improve your fine yield. 33:45.870 --> 33:46.890 More value using that. 33:48.220 --> 33:53.890 And the Mysore TV will also help you to get the benchmark for your. 33:55.210 --> 34:01.480 So you can decide, OK, for now, my model is performing with seventy five percent accuracy. 34:01.630 --> 34:06.600 So now after the morning the hippopotami goes, how much is it improving? 34:06.820 --> 34:08.820 Is it actually improving or not? 34:08.980 --> 34:15.910 And what is the accuracy value for this in comparison to other models, let's say logistic regression 34:15.910 --> 34:22.330 or random sort random forest or examples which we will be learning in coming times. 34:22.510 --> 34:25.600 So you can be you will be able to compare these mortgage. 34:26.780 --> 34:27.150 Right. 34:27.260 --> 34:33.770 So this is what we will be doing, this is how we will be training on borders and working with this. 34:35.380 --> 34:43.390 So once we get the best estimate, then you can use this particular function to generate a report from 34:43.390 --> 34:43.750 this. 34:43.930 --> 34:50.110 So I have generated one report and you can see the score, which I have, of being this zero point eight 34:50.110 --> 34:52.570 nine one, which is the best score. 34:52.570 --> 34:56.740 And the next I have zero point eight nine zero zero point eight eight nine. 34:57.010 --> 34:59.860 So the best one is zero point eight nine one. 34:59.860 --> 35:01.930 And these are the barometer values. 35:04.110 --> 35:10.890 Next, what I will be doing is still I have the best estimate, but I have not received it anywhere 35:10.890 --> 35:12.840 on my model is still not prepared. 35:13.200 --> 35:18.420 So what I will be doing is I will take the random surge, Lord, best estimate. 35:18.840 --> 35:22.430 This is the model which has the best performance. 35:22.680 --> 35:26.550 I will take this message to me and put it in the big. 35:27.600 --> 35:34.590 This is my new video which will hold the new model, the best model out of all the models which random 35:34.590 --> 35:36.150 searches bring. 35:37.860 --> 35:47.430 After I get this dream, I will again fit my model on the screen and by train using the new parameters, 35:47.430 --> 35:54.360 the and that is these best barometers, these best barometers, again, I will fit the model and then 35:54.360 --> 35:59.760 the model, which I get that does this deeply will be the model which has the best performance. 36:01.580 --> 36:09.290 Now, I can actually visualize a decision tree, so for visualizing a decision tree, I can simply say 36:09.320 --> 36:13.130 open my tree dot w. 36:15.180 --> 36:20.390 Coma W, which means that I want to write this particular fight. 36:21.370 --> 36:29.800 And this is the trial has been created, then this is the new fire which has been created and I'm just 36:29.800 --> 36:33.520 exporting the model to. 36:35.370 --> 36:42.110 This particular fire took to this particular deadly fire, to this dog fight. 36:43.630 --> 36:52.130 So after that, I will simply close the file and I can visualize this particular tree using Web graph 36:52.150 --> 36:53.380 with dot com. 36:54.580 --> 36:58.870 In this, you can actually see how the decision tree looks like. 36:59.880 --> 37:02.160 So let us visualize that. 37:03.480 --> 37:11.100 So this is the fight which has been generated, which has all the details of my decision tree, so you 37:11.100 --> 37:18.670 can see that it has details like different rules which are used for this particular decision, tree 37:18.740 --> 37:19.260 creation. 37:19.470 --> 37:21.880 So it has all the rules associated. 37:22.200 --> 37:25.430 So I'll just show you what exactly will do. 37:25.430 --> 37:30.080 We just copy this entire text into this file, which we have. 37:30.330 --> 37:36.560 So I just copied this and I just click generate graph and it will generate a graph for me. 37:46.540 --> 37:52.780 And below it, you can see the graph which has been generated, so it is a very huge graph which has 37:52.780 --> 37:53.630 been generated. 37:54.040 --> 37:56.080 So this is what it looks like. 37:57.020 --> 38:05.180 So you can see it is a very huge graph, so this is the roof node, which is marital status that is 38:05.180 --> 38:08.870 married to a spouse, it is less than equal to zero point five. 38:08.870 --> 38:10.570 Guiney value is zero point five. 38:10.820 --> 38:12.560 The samples are one hundred percent. 38:12.590 --> 38:13.700 These are the values. 38:13.700 --> 38:14.750 These are the classes. 38:15.080 --> 38:23.080 So very now this is the condition where the spouse is true and accordingly, the capital gain. 38:23.090 --> 38:24.210 This is the next rule. 38:24.230 --> 38:25.690 This is the value of it. 38:26.090 --> 38:28.970 So here you can see the value was zero point five. 38:29.240 --> 38:36.530 Here the value has reduced to zero point two nine for the number of samples are now fifty four percent. 38:36.770 --> 38:41.480 And you can see that the values are zero point eight one nine and zero point one. 38:41.480 --> 38:45.050 And one has predicted the glass to be zero, so, so on. 38:45.050 --> 38:51.370 It has just generated different class values and thus it keeps on reducing the value accordingly. 38:51.680 --> 38:57.080 So you can generate your own decision using this. 38:58.260 --> 38:59.760 And analyze the scene. 39:00.390 --> 39:00.930 Thank you.