A small demo to play for least squares regression models

lesson.land/mldeploy

Comments

Post a Comment

Popular posts from this blog

Welcome to MLDeploy

Machine learning has changed numerous applications and would continue to change with more availability of data. Once you look at the literature, you can find many different documents describing algorithms. However, one aspect that I feel needs a more comprehensive overview is the deployment of those algorithms. In this blog channel, I would like to go over alternatives for the deployment methods. Let me name a few: REST API backends Desktop applications Edge devices Mobile deployment Script Executables Web front ends No deployment There are many pros and cons for each method in terms of latency and resource usage. We will be discussing them in more detail.
Read more

Least Squares Regression

This is one of the favorite topics for people to start machine learning. It has roots more than a century. The basic problem of finding the rule behind a process lead people to think about how to create a rule from scientific measurements. Offcourse, linear relationships are the most intuitive ones. If you buy one apple it is 1$ if you buy 10 it is 10$. The fomulation in this case would be y=wx for single variable. Hoever, machine learning is reverse of the process we are not trying to find y or total amount in previous example instead we would like to find w from data. Lets extend on this apples example. If we have data like this 10 apples -> 9$ 8 apples -> 8$ 2 apples -> 3$ What would be your guess of w (unit price) if you assume linear relationship ? Least square estimate is a learning mechanism by optimizing squared discripancies between model and data. So lets formulate like this argmin_w \sum (wx-y_true)^2 In most minimization problems you tak...
Read more

Lets start coding for linear regression

As from previous post, linear regression has analytical solution. For a given dataset the optimization problem reduces to linear algebra psudeo-inverse problem.  ``` W = (XX')^-1XY``` This would be the training step for the algorithm. Python code would be very easy.  np.linalg.inv(X.dot(X.T)).dot(X).dot(Y) Later once the parameters W is defined runtime would be a single matrix multiplication as our linear model. Y' = WX  Could be written as  Y_est = W.dot(X)
Read more