diff options
Diffstat (limited to 'SI/Resource/Data Science/Machine Learning/Machine Learning.md')
| -rw-r--r-- | SI/Resource/Data Science/Machine Learning/Machine Learning.md | 81 |
1 files changed, 42 insertions, 39 deletions
diff --git a/SI/Resource/Data Science/Machine Learning/Machine Learning.md b/SI/Resource/Data Science/Machine Learning/Machine Learning.md index 0e1d5bc..7d39736 100644 --- a/SI/Resource/Data Science/Machine Learning/Machine Learning.md +++ b/SI/Resource/Data Science/Machine Learning/Machine Learning.md @@ -1,10 +1,13 @@ --- -id: 2023-12-17 -aliases: December 17, 2023 +id: "2023-12-17" +aliases: + - December 17, 2023 + - Machine Learning tags: -- Machine-Learning -- Data-Science + - Machine-Learning + - Data-Science --- + # Machine Learning Machine Learning (ML) is a subset of artificial intelligence (AI) that focuses on the development of systems and algorithms that can learn from and make decisions or predictions based on data. The core idea is to enable a machine to make intelligent decisions or predictions without being explicitly programmed to perform the task. @@ -23,9 +26,9 @@ Learning is to find the best model represented data, meaning optimization of par [[Optimization]]: Find optimal parameters -- Optimal: is the best representation of data - - A model with the smallest difference between predictions $\hat{y}$ and actual values $y$ - - A model parameter makes the smallest loss +- Optimal: is the best representation of data + - A model with the smallest difference between predictions $\hat{y}$ and actual values $y$ + - A model parameter makes the smallest loss ## Supervised Learning @@ -39,9 +42,9 @@ Learning is to find the best model represented data, meaning optimization of par - [[Support Vector Machine]] ([[Support Vector Machine |SVM]]) - [[Decision Tree]] - [[Linear Discriminant Analysis]] ([[Linear Discriminant Analysis |LDA]]) - 1. [[Ensemble]] - - [[Bagging]] - - [[Boosting]] + 1. [[Ensemble]] + - [[Bagging]] + - [[Boosting]] ## Unsupervised Learning @@ -57,43 +60,43 @@ Learning is to find the best model represented data, meaning optimization of par ## Notations - Data Properties - - Features (= attributes, independent variables): X - - characteristics of data or items - - N: # of data sample - - D: # of features - - Label (dependent variables): y - - if there is a label, it is supervised. Otherwise, it is unsupervised - - Parameter (=weight): learnable parameters that a model have, not given data - - [[Hyperparameter]]: parameters that human has to decide + - Features (= attributes, independent variables): X + - characteristics of data or items + - N: # of data sample + - D: # of features + - Label (dependent variables): y + - if there is a label, it is supervised. Otherwise, it is unsupervised + - Parameter (=weight): learnable parameters that a model have, not given data + - [[Hyperparameter]]: parameters that human has to decide - Input vs. Output - - Input ($X$): values, parts of features, are put into a model - - Output ($\hat{y}$): values of prediction derived from model + - Input ($X$): values, parts of features, are put into a model + - Output ($\hat{y}$): values of prediction derived from model - Linear vs. Nonlinear - - Linear regression: a model can be implemented by a linear function - - Simple Linear Regression: Involves two variables — one independent variable and one dependent variable. The relationship between these variables is modeled as a straight line. - - Multiple Linear Regression: Uses more than one independent variable to predict a dependent variable. The relationship is still linear in nature, meaning it assumes a straight-line relationship between each independent variable and the dependent variable. - - ex) $y = w_0 + w_1*x_1 + w_2*x_2 + \dots + w_D*x_D, y = w_0 + w_1*x_1 + w_2*x^2$ - - Non-linear regression: a model can't be implemented by a linear function - - ex) $log(y) = w_0 + w_1*log(x), y = max(x, 0)$ + - Linear regression: a model can be implemented by a linear function + - Simple Linear Regression: Involves two variables — one independent variable and one dependent variable. The relationship between these variables is modeled as a straight line. + - Multiple Linear Regression: Uses more than one independent variable to predict a dependent variable. The relationship is still linear in nature, meaning it assumes a straight-line relationship between each independent variable and the dependent variable. + - ex) $y = w_0 + w_1*x_1 + w_2*x_2 + \dots + w_D*x_D, y = w_0 + w_1*x_1 + w_2*x^2$ + - Non-linear regression: a model can't be implemented by a linear function + - ex) $log(y) = w_0 + w_1*log(x), y = max(x, 0)$ ## Basic Math for ML - Function - - Relationships or rules between two groups - - $y = f(x)$, $x$ = input, $y$ = output + - Relationships or rules between two groups + - $y = f(x)$, $x$ = input, $y$ = output - Linear function - - $y = a*x +b$, $(a \ne 0)$ - - a: coefficient (slope), b = intercept + - $y = a*x +b$, $(a \ne 0)$ + - a: coefficient (slope), b = intercept - Instantaneous Rate of Change - - is **the change in the rate at a particular instant**, and it is same as the change in the derivative value at a specific point. - - Slope (coefficient) where a $x (=a)$ meets a graph + - is **the change in the rate at a particular instant**, and it is same as the change in the derivative value at a specific point. + - Slope (coefficient) where a $x (=a)$ meets a graph - Derivation - - finds the instantaneous rate of change - - $f'(x)$ or $\dfrac{d}{dx}f(x)$ + - finds the instantaneous rate of change + - $f'(x)$ or $\dfrac{d}{dx}f(x)$ - Minimum of function - - **The instantaneous rate of change at the minimum of function is always *0*.** - - Using this property, can find optimal parameter value + - **The instantaneous rate of change at the minimum of function is always _0_.** + - Using this property, can find optimal parameter value - Exponential - - $e = \lim_{n \to \infty} (1 + \dfrac{1}{n})^n$ - - a function or growth pattern that increases at a rate proportional to its current value - - $\dfrac{d}{dx} e^x = e^x$ + - $e = \lim_{n \to \infty} (1 + \dfrac{1}{n})^n$ + - a function or growth pattern that increases at a rate proportional to its current value + - $\dfrac{d}{dx} e^x = e^x$ |