| Supervised learning | Unsupervised learning | Reinforcement learning | Deep learning |
|---|---|---|---|
| Uses a labeled dataset to train machines to predict the output for future inputs | Works on datasets without labels, it is used to predict some features without any learning according to the slides | Like unsupervised, it predict some features however it apply some correction according to specific positive or negative feedback. Simply, the AI agent explores its surrounding by hitting and trial actions. It uses a reward system. | Based on Artifical neural networks, it is inspired by the structure of the brain and its function |
| Regression, Classification | Clustering, Association |
Regression Analysis
- Predicts continous/real values
- Helps us understand how the dependant variable changes when a change in an independant vairable happens
- Regression analysis is a statistical method to model the relationship between a dependent (target) and independent (predictor) variables with one or more independent variables.
Types:
- Linear Regression: fits a linear dataset
- Polynomial Regression: fits a non-linear dataset with a linear model
NOTE: a liner model is the a model where the prediction can be written as a linear combination of the features
Classification
The output is a category, the goal is to classify into groups, not output a real value
- Logistic regression uses the idea of threshold levels, values above the threshold are considered a 1 else a 0
- SVM creates the best boundary that divides the space into classes, remember the widest road
- KNN is considered a lazy learner, it does not learn, it stores all the available data
Decision Trees
The entropy is the amount of heterogeneity or impurity in my dataset, because it has a lot of information in it
We then calculate the Information gain by each split, this tells me how much information I will gain by doing this split
I first calculate the entropy of the output feature, then I subtract from it the entropy of my split, if the entropy of my split is still high, then that means i didnt gain alot
Random Forest
We have multiple decision trees on the subset of the data then take the average to improve the accuracy
Naive Bayes
- A probabilistic classifier
In the last part it is better to creatine a new distribution by doing
K-means
used in clustering
WSS: used to know best value for k
