Chapter 6
- The main components of a typical machine learning pipeline are the predictive model (model assumptions, parameters), the loss function, and the optimization of the model parameters to minimize the loss function.
- The loss function may be minimized using gradient descent, in which we compute the gradient of the loss with the current values of the model parameters and adjust those values in the opposite direction of the gradient. JAX can compose this gradient loss function using
grad
. - In our example, we utilized kernels to create nonlinear features from the features we were given. Naïve implementations of kernels are hard to vectorize, which may lead to inefficient for loops. JAX can automatically vectorize these kernels to address this problem.