2nd degree ML Thoughts

A blog post explains machine learning concepts, stating that feature scaling works in supervised learning because models learn patterns of X to Y, so normalizing input data (e.g., X/2) maps to normalized output (Y/2). It also describes logistic regression as wrapping linear regression in a sigmoid function to output probabilities between 0 and 1, with the model learning to minimize loss for classification.

2nd degree ML Thoughts Feature Scaling. As far as supervised learning, feature scaling works, it might be counter intuitive to reduce the training data set, to some none meaning number, because in ML learning, it was never to learn about input data, it was learning patterns of X → Y. so that next time X some it close to Y. if we teach model X → Y, or X^2 → Y^2 or X/2 → Y/2, next time for input X’, if we do the normilization . then say X’ = X’/2, then it will be mapped after learning to Y/2. How Logistics Regression works by.. by just wrapping a Linear Regression in sigmoid . ? Since sigmoid is a cleaver function who’s property is to output just 0 - 1 answer, we create our model, wrapping Linear Regression in simod which become’s our MODEL . so that model Learns to minimize the loss between this outputted y’hat and the classification, which is actually just 0 and 1. so our MODEL learns i.e set optimal paramters which are optimized to give near correct answer tof rnext input in classification. For next input, test time . we still get a probability between 0 → 1 for the Probability of Y, i.e if y == 1 then probability of 1 else 0. ← Back to All Blogs /blog