coursera-machine-learning

Intro to Machine Learning

Supervised Learning

99% of values created by ML.
supervised learning refers to learn from input to output mapping.
give learning algo examples that learn from right answers based on labels.
it’s by seeing correct input x and output y.
Lab 1: Python Jupyter
Practice Quiz

1. Regression

predict number from many possible numbers. For example, house price prediction

Univariate linear regression is one input feature linear regression. If there are more than on3 features

Gradient descent

On each step update the parameter by calculating slope. Learnings rate is the rate which determines the step taken to reach the optimal minima.

Gradient Descent depends on Learning rate. If learning rate is too large, it may overshoot local minima and if it’s too small, it may slowly converge to the minima.

2. Classification

trying to categorize data instead of finding number from many possible options
small finite numbr of possible output categories (classes)
spam detection
breast cancer detection

Unsupervised Learning

In this case data is not associated with any output label y.
find some pattern or structure around data
data can be grouped into some clusters
data comes with inputs x but not output labels y. Algorithm has to find structure in the data.

Clustering Algorithm

group data into multiple clusters
grouping news stories or recommendation

Anomaly Detection

used to detect unusual data points such as fraud detection

Dimensionality Reduction

compress big data set to smaller dataset.

Terminologies

Training set: dataset used to train the model.
Feature: Input values like x like size in square feet.
Target: this is the output you want to predict y like house of price.
Parameters: Parameters are machine learning function coefficients. They are also called weight. This are adjusted during model training to reach optimal function.
Cost function: is the function used to determine the best algorithm. The goal is to minimize the cost function.