How much of machine learning is statistics and vice versa?
Learning using https://www.coursera.org/learn/machine-learning/home/welcome
- machine learning = teaching a computer to learn concepts using data — without being explicitly programmed.
- Supervised learning = "right answers" given
- Regression problem
- continuous valued output
- deduce the function for a given data set and predict other values
- "in regression problems, we are taking input variables and trying to map the output onto a continuous expected result function."
- Univariate linear regression is used when you want to predict a single output value from a single input value.
- From https://en.wikipedia.org/wiki/Regression: Regression analysis is a statistical technique for estimating the relationships among variables.
- Classification problem
- discrete valued output
- From https://en.wikipedia.org/wiki/Statistical_classification: the problem of identifying to which of a set of categories (sub-populations) a new observation belongs, on the basis of a training set of data containing observations (or instances) whose category membership is known.
- Feature = attribute
- used to predict others
- Unsupervised learning = all data is without labels or have the same label
- Unstructured data
- Can you find some structure in the dataset?
- An Unsupervised Learning algorithm can group data into partitions called clusters
- We're not telling the algorithms about structure of the data set or right answers/examples = we know nothing in advance
- Clustering algorithm
- Cocktail party algorithm
- The Hypothesis Function
- measure the accuracy of our hypothesis function by using a cost function.
- training examples
- an iterative algorithm of linear regression
- minimize the cost function
$J(\theta)$
- partial derivative - what's that? Why is this important for the algorithm?
$\alpha$
- learning rate- feature scaling = converge quicker
- Works well for smaller set of features, e.g.
n < 10^5