There’s No Free Lunch, Stupid

“Tea is an act complete in its simplicity.
When I drink tea, there is only me and the tea.
The rest of the world dissolves” – Thich Nhat Hanh

A picture is worth a thousand words, and numbers have the capacity to summarize a picture with just a few statistics, especially in today’s data driven world. The right perspective is necessary for the right kind of analysis. It is not just employing the right technique , but rather, it’s implementation  determines the efficacy of the analysis and the relevance of the insight.

An effective analysis usually aligns with the fundamentals of logic. These fundamentals are always in play, and some of them have been realized and immortalized for centuries in the form of theorems and principles. Here are two, quite old and deeply ingrained concepts in our culture. Below examines their relevance to the field of data science.

The No Free Lunch Theorem

‘There ain’t no such thing as a free lunch’ is an old maxim that originated sometime in the nineteenth century. Bars in New York offered free lunch with the purchase of a drink. However, onlookers were quick to notice that the snack offerings made available were usually of the salty, starchy variety, thus the consumer kep ordering drinks in order to facilitate consumption. The phrase gained popularity in the local dialect and gained widespread circulation through literature and cinema.

It is not just employing the right technique , but rather, it’s implementation determines the efficacy of the analysis and the relevance of the insight.

Economists such as Milton Friedman and Campbell McConnel have used this idea as a central theme for some of their works.

This concept is strongly relevant to the field of analytics, especially when dealing with complexities and growing computational costs and trade-offs. Mathematician David Wolpert and William G. Macready posit that for resolving a mathematical problem, the average computational cost of using different techniques for a list of problems, within a class of problems, is the same. No technique is better or worse than the rest, when sampled across a group of similar problems, in terms of computational costs. One of the most valued costs of computing is time. The theorem infers that no single method is going to get results sooner when compared to other methods and when applied to a pool of similar problems. Any deviations to this observation would be due to the deviant nature of the problem itself. In essence, a practicing analyst or data scientist will get familiar with techniques that are more effective for problems of their domain.

In one of our recent projects of classifying users, we realized that using Clustering didn’t yield any interesting insights. The dataset didn’t have enough usage parameters, and k-means could not separate users based on the available dimensions. We then used Social Network Analysis (SNA) to study the transition of the users’ search queries. Using SNA, we were able to discern differences in user’s search behaviors that classification algorithms failed to capture. We have an excellent article about this project that explains the application of SNA for studying search behavior in detail. Here’s the link. Do check it out: http://blogs.fiveelementanalytics.com/tracing-search-behavior-using-social-network-analysis/

The no free lunch theorem is a general theorem that helps dissolve any biases that the practicing data scientist or analyst might have towards techniques. It is liberating to know there are many different approaches to solving complex problems. This understanding paves the way  for the exploration and application of new techniques.

As the science of data analytics gains popularity, knowing multiple techniques used for problem solving is helpful to the data science community as a whole.

Occam’s Razor

William of Occam was an English philosopher and theologian, who was born in the 12th century. The Occam’s Razor is a principle that  held that amongst two competing hypothesis, the one with the fewest assumptions should be chosen. This principle can be applied in all sciences, and has a direct relevance to the field of data science and analytics. The rule of parsimony is held highly by statisticians, because it is a sure way of avoiding excessive effort and computation, thus reducing the likelihood of human error and oversight in today’s multitasking oriented environment.

Fun Fact:
the Adjusted R2 statistic penalizes non parsimonious models.

The principle is used by doctors, detectives, and countless other professionals. The simplest of the observations hold the most fundamental truths, for they require the least number of assumptions. This simple philosophical concept has been mentioned in different ways by many scientists and revolutionary personalities throughout time. In the words of Sir Isaac Newton, ‘We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances. Therefore, to the same natural effects we must, as far as possible, assign the same causes’.

The Occam’s razor has strong empirical validation for it has been applied in various discoveries, knowingly or unknowingly. Simple solutions are easy to implement, produce effective results, and lead to insights that are actionable. ‘Keep it simple, stupid’ is a design principle used by the United States Navy that talks about the same principle but with a different verbiage. It can be argued that several of the current and upcoming organizations have successfully applied this principe to theirs and their consumers’ advantages. The beauty of striking simplicity is that it requires a sound knowledge of the fundamentals. The Occam’s Razor will not only keep the analysis free from unwieldy assumptions, it will also push the analyst to strengthen their knowledge of the techniques and the domain, in their attempt to generate simpler and stronger solutions.

The beauty of striking simplicity is that it requires a sound knowledge of the fundamentals

The Church of England consider’s April 10th as the day of William Occam’s commemoration.

In summary, both of these timeless principles will make the process of analysis more effective, and will also help the insights to be more approachable. Very often we hear about ambitious plans by organizations stuck in the prototype phase because the initiative didn’t have a realistic application.  Large organizations can afford this experimentation in creativity, but it does take its toll on organizations and teams of all sizes.

If decisions or plans are based on realistic assumptions, and are allocated the right amount of resources, their application will be much more effective. In the long run, several such effective implementations could enhance the insight of the team or the individual, and avoid burnout.

There might not be a free lunch, but a cup of tea is simple and cheap.

Apurv currently works as a statistical analyst for Five Element Analytics, an analytics firm based in NY. He graduated from Hofstra University in 2015 with an MBA in Business Analytics.