Original article can be found here (source): Artificial Intelligence on Medium
A game of darts in Bias and Variance
Are you a machine learning enthusiast and also want to play darts? Well, the right time on the right page.
Let us keep this article clear and simple.
It is the simplistic assumption made by the model. It means the model is learning the wrong way to predict without learning all of the information from the data. This can possibly lead the model to the condition of underfitting for unseen data.
It is the degree of complexity undergone by the model. It means the model has a tendency of sensitivity towards the data. This also tells how scattered is the predicted value from the actual value. More sparse the data, the more the variance. The more the variance, the more is the complexity undergone by the model.
A game of darts in Bias and Variance?
Let us assume there are four players in darts who belong to different levels as said below:
Here the player is an ideal learner with high precision performance. In other terms, the learner has low bias and low variance to give an exemplary performance.
Here the player has a methodological approach that results in a good performance. The learner is said to have good stability through less bias but his throws are distributed and less precise leading to high variance.
Here the player is considered to be terrible. The learner has no good aim or stability and always misses the point on the board because of the bad throw. This is how the model performs in the condition of high bias and high variance.
Here the player has good precision and but less stability. The learner has a good aim but not a proper way of approach which often leads to missing or underperforming. The condition remains the same for the model during high bias and less variance.
Here we can relate the performance of the player to the performance of the model and also observe the role played by variance and bias in determining its potential.
Everybody wants to become Phil Tyson! why not? When we can make a trade-off between Bias and Variance, we eventually get a good learner if not an ideal one.
Happy learning and 180!