My Progress

1. IntuitionEx) Cat Classification exampleWe have three different features(Ear shape, Face shape, Whiskers) of cats to determine whether the given image of an animal is a cat or not.  TerminologyDecision tree algorithm in machine learning shares the same knowledge of the tree data structure in computer science. The top node of a tree represents the root node. Nodes at the very bottom of the tree..

1. Machine Learning Development Process 1.1 Cycle of ML process Step 1: Choose Architecture (mode, data, etc) Step 2: Train Model Step 3: Diagnostics (bias, variance, and error analysis) Step 4: Deploy in production(Deploy, moinor and maintain system) 1.2 Example (Spam Classification Example) Example: Spam Classification Example Supervised Learning Algorithm x = feature of spam emails y = whethe..

1. Advice for applying machine learning 1-1. Intuition In order to make a efficient machine learning model, we have to make a good decision. Then, how do we make a good decision? Ex) Problem: We implemented a regularized linear regression on housing prices, but the accuracy of the model is low. Possible answer: More training examples, try smaller or bigger features, add polynomial features, incr..

1. Details of training #Step 1 : Specify how to compute output given x and parameters w, b (define model)#Math : f_w,b(x) = ?#Logistic Regressionz = np.dot(w,x) + b#Step 2 : Specify loss and cost#Math: L(f_w,b(x), y)# J(w,b) = 1/m sigma L (f_w,b(x(i)), y[i])f_x = 1 / (1 + np.exp(-z))loss = -y * np.log(f_x) - (1 - y) * np.log(1 - f_x)#Train on data to minimize J(w,b)w = w - alpha * dj_dwb = b- ..

1. Neurons and the brain Origins: Algorithms that try to mimic the brain Development RoadMap: speech -> images -> text(NLP) -> ... Since the development of Large Neural Network and GPU, we could perform many different task using the "Big Data". 2. Example - Demand Prediction We use sigmoid function to solve the classification problem. We called the function we use f(x). In the neural network, we..

This is the overview of supervised learning. Based on the characteristics of problems, we can divide problems into regression and classification. To solve the problem, we use linear regression for regression and logistic regression for classification problem. We have to get the best linear or logistic regression formula that fits the datasest. The way to find the formula is the cost function. By..

1. Intuition Regression Example Underfit / High bias : does not fit the training set well Overfit / High variance: Fits the training set extremely well Generalization: Fits training set pretty well Even though overfitting example might have a high accuracy on the training set, it will not have a high accuracy for new test dataset. 2. Addressing overfitting Option 1: Collecting more examples By c..

1. Cost function1.1 IntuitionWe use logistic regression / sigmoid function to estimate the data's label or category. How do we choose w and b?For linear regression, we used squared error cost.Linear regression is a convex form, so we could use standard gradient descent equation to figure out local minimum. Since logistic regression is a non-convex form, it has multiple local minimum. Thus, we ca..

1. IntuitionLinear Regression is not a good method for Classification Problem.Why?One outlier on the right changes the linear regression function. It moves decision boundary dramatically. 2. Logistic Regression2.1 FormulaLogistic function allows the function to be curved unlike the linear regression. This is a sigmoid function aka logistic function. x-axis represents the number z. It outputs a v..

1. What is Feature Engineering? Using intuition to design new features, by transforming or combinging original features. Example) House price. If we are given the dimension of the house such as length and width, we can create a new variable to include for price prediction of a house. For example, we create a area variable with length and width. 2. Feature engineering in Polynomial Regression We ..