Speaker 1: Let me briefly explain you Subjective Segmentation or Cluster Analysis. In case of Cluster Analysis, your task is to figure out which are the objects which are similar. This is the case where you have been given some toy, their shape, their color, their make, shape is like triangle, circle, all those things, color you have red, blue, all those things, and make you have these are made of metal or plastic. And the task that has been given to you that find objects, observations which are similar. You can call these are observations or these are objects. So in Cluster Analysis, your task is to find observations which are similar. Now what is the whole idea? The idea is that when you group them together, they should be homogeneous within group. What does it mean? That each object in the group should be similar to each other and the group should be quite distinct. So each group should appear distinct from other group or in other word you can say that any object of one group should be quite dissimilar to any object of other group. Now think of if you decide just shape that you will just be bothered with the shape, you can actually make group like this. You can say P, A, B all are triangle, Q, X, Y, Z are all circle, R and Z are rectangle. So this is how you will make the group. Now if you are right now you are just bothered with the shape, so you can say all these are similar because all these are triangle, all these are circle, all these are rectangle and each object of this group is quite dissimilar to any object of this group. Each object of this group is similar to any object of this group. So within group you have homogeneity, between group you have heterogeneity. That's what is clustering. Now consider one more thing. You consider shape, you get this kind of cluster P, A, B came one place. If you would have considered color, you know you would have got a different cluster. Now if you think of some of the thing like which was here Q, X, Y, Z is in circle in one group. Now that has come in a different group, Q, X is here, Y, Z is here because their color is different. I mean now you are doing based on color, so you have one color here, one color here. So you got a different grouping all together or if you would have considered both shape and color, you would have got graph like this, right, group like this where you would have got Q, X one side, Y, Z one side because these are red as well as circle, these are blue and circle. So point that I am trying to say that group is, there is no dependent variable kind of thing. I mean if you consider what you are doing in classification tree, you had a dependent variable outcome Y, N and that is why there was a science that okay which are the variable you should be taking so that you can develop it in the best possible way. Here there is no objective function, there is no dependent variable. The only task that you have is that which are the observations which are similar or you can say which are the objects which are similar and you have the choice of taking with any variable. Whatever attribute you will select, whatever variable you will select, you will get group like that and these groups will be homogeneous on the selected attribute only. Right now it is appearing quite homogeneous, okay, this is circle as well as red, so Q and X are exactly similar but the moment you start taking third variable, for example if you take Q and X, you take the third variable, so they are dissimilar because one is metal, another is plastic. So what I am trying to say that you take different attribute, you will get different groups. These groups are homogeneous based on selected attributes only. Also the more attribute you will take, you will require more and more cluster. If you are taking just one attribute, say color, only two groups were sufficient. You are taking shape, only three groups were sufficient, the moment you took both of them, you got actually, if you think of you actually have got six groups because if there is any object which comes here below which is triangle and blue in color, you will have to create another group. So more the attribute that you will take, more the groups you will have to create and then each group will be homogeneous within and heterogeneous across.