Machine Learning

ML Key Terms

Labels are the thing we're predicting.
Features are the input variable.
Models define the relationship between features and labels.
Two phases of a model's life:
- Training means creating or learning the model.
- Inference means applying the trained model to unlabeled examples.

	Supervised	Unsupervised	Semi-Supervised	Reinforcement
Overview	Majority of algorithms. The machine is trained using well-labeled data; inputs and outputs are matched.	Unlabeled data (inputs only) is analyzed. Learning happens without supervision	Some data is labeled, some not. The goal is to get better results than labeled data alone. Good for real-world data.	An agent interacts with its environment by performing an action and learning from errors or rewards.
Process	Mapping function takes inputs and matches to outputs, creating a target function.	Inputs are used to create a model of the data.	Combination of the supervised and unsupervised processes.	Follows trial-and-error method.
Subtypes	Classification, Regression	Clustering, Association, Dimension Reduction	Classification, Regression, Clustering, Association	Reward-based
Algorithms	Linear Regression, Random Forest, SVM	PCA, K-Means	Mixture models, Semi-supervised SVM	Q-Learning

	Classification	Regression
Overview	The task of predicting a discrete class label.	The task of predicting a continuous quantity.
Subtypes	Binary Classification: The data is classified into two classes/labels. Multiclass Classification: The data is classified into more than 2 classes/labels.	Single Regression: The data can have one dependent (y) variable with one independent (x) variables. Multivariate Regression: The data can have one independent (y) variable and two or more dependent (x) variables.
Algorithms	Naive Bayes, Logistic Regression	Linear Regression
Type of Output	Categorical	Continuous
Assessment of Model	Accuracy/Confusion Matrix	Root Mean Squared Error
Examples	Binary Classification: detecting if given email is spam or not (classes - yes/no) Multiclass Classification: predicting what dog breed is presented to the model (classes - pug, labrador, golden retriever, etc.)	Single Regression: predicting the price of a house (y) given the feature of square feet (x). Multivariate Regression: predicting the price of a house (y) given the features of square feet, number of rooms, schools in the area, etc. (x).

	Clustering	Association
Overview	Grouping a set of objects in such a manner that objects in the same group are more similar than to those object belonging to other groups.	Finding associations amongst items within large commercial databases.
Subtypes	There are 5 types of clustering methods: Partioning, Heirarchical, Fuzzy, Density-Based, Model-Based	f
Algorithms	K-Means, Agglomerative	Apiori
Examples	Split up similar clothing into stacks.	Find what clothes one often wears together.