Fit a line by gradient descent

Botanist Edgar Anderson measured 150 iris flowers in the 1930s, and the statistician Ronald Fisher made the data famous in 1936. You will use two of the measurements: petal length and petal width. They are strongly related, and your job is to learn the relationship as a line, $\hat{y}=wx+b$.

You'll write every piece: the model, the loss, its gradient (checked numerically), the training loop, and a comparison with the exact answer. Each step runs in the same Python session, so variables from earlier steps are still there.