Week 3 Lab 1: Understanding Planar dataset

[mrbeast1]

Hi, Sorry in advance if I’m asking silly question, but while solving Week 3 programming assignment and I didn’t understand exact problem statement we are trying to address here.

What I understood, we want to fit the given dataset and create a model such that, any new given data-point (which have two values which are co-ordinates of the data visualisation plot) is supposed to be classified into two categories which are ‘red’ and ‘blue’(colour of flower). Since in given dataset, we do not have one category in one section of graph and one in another section, so why do we try to draw line which separates them in two categories?

If these training data-points are random, why do we see flower like pattern here? What do these input co-ordinate mean if output is red or blue flower?

Thanks in advance!

[paulinpaloalto]

The points are not random. They are a pattern in the shape of a flower with a mixture of red and blue dots forming the shapes of the “petals”. Now the question is whether you can build a neural network that has a decision boundary complex enough to “fit” that flower and predict where the blue and red dots are. In Week 2, we were working with Logistic Regression and that has the limitation that it can only create linear decision boundaries. In fact, they show us early in the Planar Data notebook a demo that LR is basically useless for this problem.

But a real Neural Network, even one with only 2 layers can actually handle it. If you look at the end of the notebook, they give you some other input datasets where the patterns to model are not a flower but other things. Have a look!