WEBVTT 00:01.810 --> 00:08.590 In this session, we will discuss about decision trees, so now we have discussed about the Linnean 00:08.590 --> 00:17.110 models, we have seen that the linear models try to create an equation of a straight line and try to 00:17.110 --> 00:19.180 fit the data out with. 00:20.100 --> 00:28.200 In case of decision trees, we create a sequence of decision nodes or a tree like a structure where 00:28.350 --> 00:36.120 we will make decisions at different point and at the end of all the decisions, we will get a clear 00:36.120 --> 00:36.610 winner. 00:36.810 --> 00:38.990 What should be the target value? 00:41.030 --> 00:42.620 So let us have a look at that. 00:45.180 --> 00:50.100 So in this particular tree, we will be having different. 00:51.690 --> 01:02.820 Values now we have values like dependent variables and the independent variables in this kind of a tree, 01:03.030 --> 01:08.040 the dependent variable is the data which is present inside the three. 01:09.870 --> 01:20.100 That is our data, which will be pointing to either a child should play or not play and the independent 01:20.100 --> 01:26.250 variables or the features or the attribute become the decision points. 01:27.580 --> 01:36.670 So each decision point will be created from the feature columns, so in this particular data, the feature 01:36.670 --> 01:43.420 column contains the outlook of the day, the humidity of the day, and if the day's windy or not. 01:45.410 --> 01:52.820 So this kind of structure would be generated and based on this structure at the end of the tree, I 01:52.820 --> 01:58.190 believe we will be getting if a child should play or not play. 01:59.680 --> 02:02.840 So let us have a look at this thing. 02:05.210 --> 02:09.280 Types of bombs, which we have in case of decision. 02:10.100 --> 02:19.050 So the first bomb is to promote the rule of law, to represent the entire population or the sample. 02:19.610 --> 02:24.350 And this further gets divided into two or more homogeneous sect. 02:26.440 --> 02:35.080 So here we have this entire three and the place where the tree begins is called the root node. 02:36.110 --> 02:47.960 This room contains all the data that is all 14 rows of data which we have and for each decision, which 02:47.960 --> 02:53.340 we think divides these 14 rows of data into smaller chunks. 02:54.260 --> 03:03.440 So these 14 rows of data on the basis of the decision node that its outlook is subdivided into five 03:03.440 --> 03:09.830 rows of data here, five rows of data in this section and another five rows of data in this section. 03:12.120 --> 03:19.530 Now, that child structure, the story structure where we have one child at the top and then a. at the 03:19.530 --> 03:26.760 bottom is called a parent child relationship, where the node above is called the parent node. 03:26.850 --> 03:30.840 And the node which is created from it, is called the child. 03:31.560 --> 03:39.060 Similarly, this is a parent node I the node which is created from it is called the child. 03:40.600 --> 03:47.050 And in the field of parent child would be this as the parent and this as the child. 03:48.130 --> 03:55.840 So let us look for the now what a splitting splitting is a process of dividing a node into two one more 03:55.840 --> 03:56.530 sub not. 03:58.260 --> 04:01.960 So here we have the first node where we have 14 rows of data. 04:02.340 --> 04:05.830 This note is subdivided into three nodes. 04:06.630 --> 04:12.540 So this node is divided on bases of one rule that there's one condition. 04:12.720 --> 04:13.980 So what is the condition? 04:14.160 --> 04:16.350 The condition is what is the outlook? 04:16.650 --> 04:20.820 Is the outlook sunny or is it overcast or is it rainy? 04:22.670 --> 04:32.960 So based on this decision, the note on the door has been subdivided, this subdivision of Naude on 04:32.960 --> 04:37.340 basis of a coalition is called splitting of the node. 04:41.140 --> 04:44.020 Now, the next is decision not. 04:44.830 --> 04:51.790 When a sub node split into four those sub nodes, then it is called the decision node. 04:52.970 --> 04:55.080 So this is the road north. 04:55.790 --> 05:04.240 Now we have another road, which is a sub node, this sub node is again divided into different Jagels. 05:04.610 --> 05:09.050 So this is subdivided into different Miloje based on one condition. 05:09.320 --> 05:17.330 Now here the condition is humidity, which is either the humidity is less than equal to 70 or the humidity 05:17.330 --> 05:18.710 is greater than 70. 05:20.080 --> 05:28.560 In this particular case, this sub node is subdivided into two sub nodes where the condition is Vendy, 05:28.800 --> 05:31.400 the decision which is being made is windy. 05:31.660 --> 05:36.490 So these two modes are known as the decision nodes. 05:38.730 --> 05:48.690 Next is the leaf node or to the node that Doolan's plate is called the need for determining node, that 05:48.690 --> 05:56.320 is the node at the very end of the tree are known as the leaf node or the dominant. 05:58.830 --> 06:05.980 Now, what is a branch or a subtree, a subsection of the entire tree is called a branch or a subtree. 06:06.780 --> 06:08.530 So this is a subsection. 06:08.790 --> 06:11.190 Again, this one is another subsection. 06:11.340 --> 06:16.020 So both of these are called branches or subtree. 06:19.050 --> 06:20.770 Next is parent and child. 06:21.030 --> 06:27.210 So, again, the node which is above the sub node, which is above, is called the betting node and 06:27.210 --> 06:31.590 the node which is created from the better node is called the child. 06:36.370 --> 06:36.820 Now. 06:37.850 --> 06:45.320 Let us have a look at the process of the decision tree generation, they now what we have seen is how 06:45.320 --> 06:46.850 the decision tree looks like. 06:47.480 --> 06:55.310 Now we need to learn about what is the process of creating a decision tree now for creating a decision 06:55.310 --> 06:55.660 tree. 06:55.850 --> 06:57.840 We need to have the rule of law. 06:58.070 --> 07:02.400 We need to have an initial node and all the rules of detainment. 07:03.230 --> 07:10.730 Now, after we have all these rules of data, we need to define one particular rule on the basis of 07:10.730 --> 07:14.060 which we will be splitting the nodes. 07:14.840 --> 07:22.030 And once we have selected this particular rule, then we will be doing this split. 07:22.340 --> 07:28.700 And after this particular split has been made, then we will again evaluate all the rules which are 07:28.700 --> 07:36.260 present and we will compare all the rules and find out the next best rule on basis of which we will 07:36.260 --> 07:37.430 be making the split. 07:39.260 --> 07:46.850 Now, how we will generate these rules and how we will compare these rules is a secondary thing and 07:46.850 --> 07:50.640 we will discuss that in some in some time. 07:51.170 --> 07:55.780 Now, let us have a look at how we will make the decision. 07:56.700 --> 08:01.970 So we will have certain rules on the on the basis of which we will be subdividing. 08:02.760 --> 08:07.170 So I'm the final stage when we have this leave node. 08:07.350 --> 08:13.950 So based on the majority values of the leaf node, we will decide if the child should play or not play. 08:14.280 --> 08:18.540 So here we have two values for play and zero values for born to play. 08:19.020 --> 08:26.580 So what we will do is we will decide if that the child should play because the majority is with play. 08:28.550 --> 08:35.180 So let us look at the process so the trees are generated by selecting the best route possible at every 08:35.180 --> 08:35.490 split. 08:35.750 --> 08:42.450 So we will have a lot of rules, we will create a lot of rules, and we will have to select the best 08:42.470 --> 08:44.750 rule to make the flick at every point. 08:45.620 --> 08:51.950 Now, there are two types of problems, which is classification, and another one is regression. 08:53.230 --> 09:00.280 Classification is where we want to find out a particular class, for example, if a loan should be approved 09:00.280 --> 09:09.040 or not, if someone will be defaulting on the loan or not, if a person should play or not, if I should 09:09.040 --> 09:10.420 be cooking today or not. 09:11.980 --> 09:14.950 So these would be classification problems. 09:15.190 --> 09:16.930 Now, what is a regression problem? 09:17.110 --> 09:20.810 A regression problem will be what is the height of the person? 09:20.980 --> 09:22.700 What is the income of the person? 09:22.900 --> 09:26.410 What should be the interest rate for a particular loan? 09:26.590 --> 09:31.510 So these are all values which are continuous in nature would be a regression problem. 09:31.900 --> 09:37.770 And Decision Tree will help us to find out solutions to the classification. 09:37.780 --> 09:39.890 Also, I do regression also. 09:40.210 --> 09:43.780 We can solve both type of problems using a decision to. 09:45.450 --> 09:47.270 Now, how will we do that? 09:47.730 --> 09:55.620 So in case of classification, we can find out either the probabilities or the hype, plus if we want 09:55.620 --> 10:01.830 to find out the probabilities, then we can find out the probability by opening the proportion of the 10:01.830 --> 10:03.150 value that the methanol. 10:03.450 --> 10:07.460 So we have these leads not so we can find out the probability of. 10:08.070 --> 10:11.670 So here the probability of playing would be one here. 10:11.670 --> 10:14.850 The probability of playing would be zero here. 10:14.850 --> 10:18.600 The probability of not playing would be one here. 10:18.600 --> 10:21.300 The probability of not playing will be zero. 10:21.600 --> 10:25.320 So this is all we can find out, the probability of the classes. 10:25.710 --> 10:28.710 Now, how can we find out the hard class? 10:28.830 --> 10:34.820 If you want to find out the actual high gas, what we can do is either we can select on the basis of 10:34.820 --> 10:35.430 the majority. 10:36.000 --> 10:39.780 So here we have two words for playing, zero words for not playing. 10:39.780 --> 10:40.770 So we would play. 10:41.280 --> 10:44.790 Here we have zero words for playing and three words we're not playing. 10:44.790 --> 10:47.070 So majority is for don't play. 10:48.810 --> 10:56.180 Next is applying the cutoff on the probability score, so many will be finding out the probability value. 10:56.310 --> 10:58.080 So we will apply a cutoff. 10:58.230 --> 11:02.410 You see, we can have a cutoff of, say, 60 percent. 11:02.580 --> 11:07.050 So if the probability is zero point six, then go out to play. 11:07.050 --> 11:10.880 If the probability is less than zero point six, then don't play. 11:11.010 --> 11:12.690 So we can do something like that. 11:14.310 --> 11:21.540 Now, in case of regression, so in case of regression, what we do is we make the decision and it is 11:21.780 --> 11:26.550 dependent on the average of the target value, either Dominello or leave. 11:27.120 --> 11:32.100 Now, in case of classification problem, we have values as play. 11:32.100 --> 11:35.130 I don't play present in the leave. 11:37.670 --> 11:46.400 But in case of a problem of regression in the lymph node, we will have continuous values. 11:47.510 --> 11:54.710 We will have values of, let's say, 10 rows of data or 20 rows of data, which will have different 11:54.710 --> 11:59.680 interest rate values or different salani values or different height values. 11:59.960 --> 12:03.720 Now, based on all of those different values which are present here. 12:04.010 --> 12:11.720 We will be taking an average of all of those values and that average will actually be the result of 12:11.720 --> 12:12.890 this decision that. 12:17.480 --> 12:25.790 Now we need to create a decision tree and we will be creating this decision tree using a lot of splitting, 12:25.790 --> 12:28.020 which will be doing now. 12:28.250 --> 12:36.650 We cannot keep doing this splitting endlessly, because if we keep on splitting the decision tree, 12:36.860 --> 12:40.920 then at the end we will have a very huge decision day. 12:41.210 --> 12:51.320 With all the rules of data in a different block of liefooghe, each data rule will convert into one 12:51.320 --> 12:59.030 leaf node and the tree will be very large in structure and also it will become very complex in nature. 12:59.540 --> 13:06.110 And the forest guideline, which we have for machine learning, is that the simpler, the better. 13:06.740 --> 13:16.190 We will always decide on going towards a simpler model in comparison to a complex model because a complex 13:16.190 --> 13:18.310 model tends to overfit. 13:18.950 --> 13:26.030 This is the reason why we will choose a simpler model and hence we will have to stop splitting at one 13:26.030 --> 13:26.800 point of time. 13:28.260 --> 13:34.650 Now, hi, when will we stop this bleeding and how we will decide when should we stop? 13:34.650 --> 13:42.180 The thing is based on a few great ideas and a few set of conditions, which we have already decided. 13:42.810 --> 13:44.430 So what are those conditions? 13:44.620 --> 13:51.080 We will have a look at them once we actually know that, how we will start creating the decision to. 13:54.310 --> 13:58.510 Now, let us have a look at the Roy Creation's. 14:00.690 --> 14:08.220 So for creating rules for the new music video, we will split up then the ranges. 14:10.640 --> 14:17.490 So we will have to split up the random positions, so let's say we have a numeric variable each. 14:18.170 --> 14:21.720 Now this numeric variable, each can have any value. 14:21.920 --> 14:24.320 See five fifteen twenty four. 14:25.100 --> 14:27.480 Twenty nine fifty six hundred. 14:27.500 --> 14:29.300 Anything would be of value of each. 14:29.810 --> 14:35.900 Now there is no specific range or any specific distance on basis of which we will see. 14:36.950 --> 14:43.150 Now here you can see the rules are like if the humidity is less than equal to 70 or greater than 70. 14:43.430 --> 14:50.780 So there is any value can be created, converted into a rule, but the rule will be created randomly 14:51.500 --> 14:52.850 so it is not fixed. 14:53.270 --> 14:59.630 We will have a rule at zero, then under the rule at five, then next rule as then, then 15, then 14:59.630 --> 15:00.700 20, then 25. 15:01.100 --> 15:03.050 No, that is not a criteria. 15:03.200 --> 15:11.730 We can have one rule at age greater than 10 and under the rule at age less than 15 and next at 16. 15:12.410 --> 15:20.210 So there is no specific criteria or no specific range or interval which is present in creating a rule 15:20.360 --> 15:22.700 out of the numeric variables. 15:23.300 --> 15:26.840 Any value can be used for generating this rule. 15:28.780 --> 15:35.770 Now, let us create a rule for categorical video we have seen for numeric video, but the next rule 15:35.770 --> 15:38.650 will be generated on the categorical variable. 15:38.680 --> 15:39.850 How do we create the rule? 15:39.850 --> 15:45.890 Uncategorically, video will be created by splitting on the value created then zero point five. 15:46.210 --> 15:54.970 So let's say we have one categorical column as gender, another categorical column as Sipi Wichai values, 15:54.970 --> 15:58.250 let's say New York, Paris and Moscow. 15:58.720 --> 16:07.560 Now we can avoid the categorical column into numerical column by the creation or by one Hawtin pooling. 16:07.930 --> 16:11.680 So the gender column has been converted in the male column. 16:11.680 --> 16:18.250 Or we could have created a female column and if the value is zero, then we are saying it is a female 16:18.400 --> 16:21.510 and if the value is one, we are saying that it is Amedi. 16:22.920 --> 16:29.580 So the condition which we can evaluate here is if the value of milk is greater than zero point five, 16:30.450 --> 16:33.590 then it would mean that we are pointing towards me. 16:34.170 --> 16:37.320 Otherwise it would mean that we are pointing towards Fehmi. 16:39.410 --> 16:48.050 Similarly, we can create less city and New York, greater than 50 divide is greater than zero point 16:48.050 --> 16:48.330 five. 16:48.560 --> 16:54.440 So now, again, the value will be compared with the city value because it is one. 16:54.440 --> 16:56.980 So it will mean that we are pointing towards New York. 16:57.290 --> 17:00.740 If it is Paris, then we are pointing towards one. 17:00.990 --> 17:03.470 So this is accordingly evaluated. 17:05.670 --> 17:13.440 Now, the next thing which we will be discussing about is the splitting, so we will discuss about splitting 17:13.440 --> 17:19.400 and how we will do the splitting and different rule generation for the split in the next session.