| |June 20199CIOReviewUnderstanding bias and variance is critical for understanding the behavior of prediction modelson whom you talk with. In my sting with a MNC in Midwest - USA, early in my carrier, I was working as a Manager and had teams reporting into me. We were looking for bonding avenues as the team was forming. One of the recommendations from the team was to a potluck (meaning bring food from our respective houses and eat together in the office).I happen to bring in well-cooked Okra, properly stuffed with spices from the Indian subcontinent; I heard a comment from one of my team members that isn't Okra a `weed'. Not the weed which is bad for health but more of an unwanted plant types `weed'. To make it worst someone made a comment in the room telling me I should have brought in half-cooked walking or swimming creature from the eastern part of the world. The situation was becoming quite weird and I was thinking to myself "Jee" its only food and why should anyone be judged based on the food they eat. We are humans; biologically all of us need energy to survive so our body functions. During our Potluck, we had a decent discussion on multiple topics but the gentleman was stuck at refusing to accept it as vegetable and kept calling it an unwanted plant. Giving him a benefit of doubt, I asked him the question on "why is he saying what he is" additionally I asked him "why is it bothering him?" Based on his experience he has seen Okra growing on highways, along the unmanned roads. Anyways a Bias got formed.It would not be wrong to say we form opinions based on our experience. However there is a thin line between `Bias' and `opinions based on Bias'. A lot of us fall into this trap of making opinions based on Bias.When you have `high bias' you tend to look at `limited options'; If you don't have `bias' or `limit the bias' then your `variance is high', in today's world this could be looked from a psychological aspect as having `indecisive behaviour'. When choosing decision points within the machine learning world it is very important to ensure that you do take a good spread of points to get an even spread of results. However, a lot of us have habit to exclude the `outliers', decision points, which are very far from average results. We should look at the high bias and high variance data sets to train our models.The Low Bias and Low Variance data sets should eventually be used for analysis but our models needs to be trained on every type of data so the model also helps us recognize what would be a good choice of data to get the effect results we are seeking. Low Bias because we don't want to judge based on just a few events or matches; low variance because we do want to make some predictions in the end. Understanding bias and variance is critical for understanding the behavior of prediction models. As we coach the machine to focus on these; we should coach ourselves too and have Low Bias and Low Variance.
< Page 8 | Page 10 >