tag:blogger.com,1999:blog-3366935554564939610.post3759896371544222104..comments2019-08-20T11:36:29.313-04:00Comments on Data Miners Blog: Doing the Right Thing: Are your measures correct?Michael J. A. Berryhttp://www.blogger.com/profile/06077102677195066016noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-3366935554564939610.post-3485355646332366482014-04-22T23:33:36.441-04:002014-04-22T23:33:36.441-04:00I agree with the comment above--more than just und...I agree with the comment above--more than just understanding the data and what the data tells us about the world, we must understand the underlying problem which we are trying to solve.<br /><br />Data mining and other learning problems typically seek to build a representation, or model, of an unknown system. To do this, these problems require three things: data (as described in the article), a class of models from which we can choose a solution, and a metric by which we can judge which of the models in the class of models best represents the data. <br /><br />Once a problem is formulated with the above elements, we can then choose a solution, such as a data mining algorithm, through which we build the model. This solution, however, is meaningless unless we, as the post suggests, have an understanding of how to map the solution back into the real world.<br /><br />In brief, we need to know more than just how the data relates to the world. We need to have a solid understanding of the problem we are seeking to solve and how the entire data mining process (data collection, formulation, and solution generation) map to this problem and its respective solution.Nathan Woodburyhttps://www.blogger.com/profile/00525334934770662624noreply@blogger.comtag:blogger.com,1999:blog-3366935554564939610.post-86076319978513589982014-04-17T19:16:20.995-04:002014-04-17T19:16:20.995-04:00This is a fantastic data mining example of a princ...This is a fantastic data mining example of a principle I believe extends beyond the domain of just data mining – solving the problem isn't the problem, understanding the problem is the problem. As a computer scientist myself, I can sympathize with the desire to just blindly follow the spec, and hash out an answer to the problem as stated. That, however, is a pitfall for the unwary. The key to the process of correct problem solving begins and ends with problem formulation. A proper understanding of what it is you're really trying to do is essential to the design of a correct solution. Are you solving the problem as stated, or are you solving the problem you really have? (In this context, are you performing arithmetic with the given numbers - which were flawed in this case - or are you solving the actual problem you had, determining the cost-benefit ratio of the ad campaigns?)<br /><br />The ability to solve the problem you really have, and not just the one that's stated in the text, is what I believe we refer to when we talk about “thinking outside the box”. That point is well illustrated with an example from one Laura Gilchrist, a teacher. She tasked her students with writing their “first name in as many ways as [they could]”. They proceeded to perform the task as they perceived it: simply rewriting their names in different styles. After being prompted by their teacher, “I notice you're all still sitting in your chairs writing your names,” however, the students realized that the bounds they had perceived to be inherent in the problem were instead artificially imposed, and the classroom quickly became a flurry of activity. <br /><br />This all begs the question, however; assuming there is a real problem hidden behind our stated one, how do we go about discovering what our real problem is? This is the precise idea behind a powerful technique used originally by Toyota known as the “5 whys”. The idea is to iteratively repeat the question “why” to the problem statement in order to explore the true cause-and-effect relationships that underlie the problem to be solved. As I lack the expertise to discuss this process in any greater detail, I would point the interested reader to its Wiki page, which is both informative and serves as a good starting ground for further information on this and related techniques.Justin Wunoreply@blogger.com