stuti-16 / bias-variance-trade-off-for-model-selection

Here we assume an arbitrary function 𝑓(𝑥) and compute its value on 10,000 equally-spaced data points. We try to find out the optimal complexity of the regression model that can best estimate this function.

Steps to follow:

• Take a trigonometric function and a choose an error function (N(0, sigma-sq)). Generate data set of 10,000 instances. Fit polynomials of order 1 - 10 and estimate and plot total error, Bias, Variance,

• training and validation error for each using 10-fold cross validation. Create the folds using Python function and compute all errors using the equations given in textbook. Select the optimal model.

• For the selected model, estimate and plot total error, Bias, Variance, training and validation error for training set sizes 1K, 2K, ... 10K. Use 10-fold cross validation for each training set, and the functions coded earlier for estimating the error.