Machine Learning Checklist
This post is a work in progress.
I am taking Stanford Univerisity's Machine Learning course with Andrew Ng's on Coursera, and beginning to learn about the various formulas available to a Machine Learning Engineer. This post will serve as my attempt to more quickly determine which formula might be a good fit.
Is it supervised learning?
- Can linear regression be applied?
- Is it unsupervised learning?
- If alpha (learning rate) is very small then gradient descent can be slow. If alpha is very large then gradient descent will be less accurate.
- Gradient descent can still converge to a local minimum, even with alpha (learning rate) fixed. This is because the derivative (the slope of the tangential line) is smaller each time.
- Gradient descent will scale better than the Normal Equations Method to solve for the global minimum of the cost function.
- A vector is a matrix that has only one column.
- Scalar multiplication is the act of multiplying a matrix by a number.
- You can only add 2 matrices of the same dimension.
- Matrix multiplication is used to apply your various hypothesis (get the prediction values) which can then be put into the cost function.
- Matrices that don't have an inverse are called 'singular' or 'degenerate'.
- Plot values first. Use a graph if just 2D or 3D (1 or 2 parameters).