Summary: Ml/ai Intro: Coursera Class

Read the summary and the most important questions on ML/AI Intro: Coursera Class

• 1 Intro & Univariate Linear Regression

• 1.1 Intro

  This is a preview. There are 1 more flashcards available for chapter 1.1
  Show more cards here

• Example: Classifying email as spam/not spam

  
  Example: Suppose my email program watches which email I do or do not mark as spam, based on that learns how to better filter spam.
  
  
  – Classifying emails as spam or not spam is a Task T
  – # of emails correctly classified as spam/not spam is  performance measure P
  – Watching me label email as spam/not spam is an experience E

• Machine Learning Algorithms

  
  Machine Learning can be divided in multiple fields:
  
  
  Deciding which project as supervised learning vs supervised learning is an art, and it becomes better with practice.
  
  
  

  • Supervised Learning
  • • What is the right answer, given some input?
    • Here we give the “right answer” for each example in data to train the model.
    • Regression: Predict continuous valued output (ex: price)
    • Classification: Discrete valued output (e.g. 0 or 1)
  • Unsupervised Lerning
  • Others: 
  • • Reinforcement Learning
    • Recommend-er systems

• Machine Learning Mind Map

  Take [Training Data] ----> [Learning Algorithm] ----> [Outputs a hypothesis h]
  
  We use this hypothesis h to make prediction from test set

• 1.2.1 Gradient Descent Algo

  This is a preview. There are 4 more flashcards available for chapter 1.2.1
  Show more cards here

• Gradient Descent Algo Psuedo-Code:

  It's main goal is to minimize the cost function J: 
   
  
  
  
  
  Note 
  

  • := (is an assignment operator): For example: a := a +1. Here a gets updated with the value of a + 1. 
  •    = Learning rate (algorithm knows where to take small steps) 
  • gets updated simultaneously, and this is a crucial step 

   
  

• Question: Suppose,  (assume =0) is stuck at local optimum of , such as shown in the graph. If that happens, what do I think will happen to one step of gradient-descent?

  - In such case, will stay unchanged, as derivative is zero here.

• What is correct simultaneous update in Gradient Descent?

  Correct: Simultaneous update
  
  

• "Batch" : Gradient Descent

  "Batch": Each step of gradient descent uses all the training examples

• 2 Linear Regression with Multiple Variables

• 2.1 Multivariate Linear Regression

  This is a preview. There are 2 more flashcards available for chapter 2.1
  Show more cards here

• Hypothesis: Multivariate Liner regression

  
  
  
  Here    = vector of 1 (to calculate the bias unit)

• Gradient Descent Algo: Single vs Multi factors

  See attached image
  
  

• 2.1.1 Gradient Descent in Practice

  This is a preview. There are 2 more flashcards available for chapter 2.1.1
  Show more cards here

• Practical tricks for converging gradient descent I:

  How to adjust features: 
  
  

  • Feature Scaling Make sure features are on a similar scale. Get every feature into approximately  range.  

  
  

  • Mean Normalization: Replace with , to have feature approximately close to zero. 
  • • here S(i) = Either range of feature (max -min) or Std of data 

  
  
  

  • Feature Selection: Adjust features to make sure data represents meaningful information. 
  • • For example, if we have data on length and width of the house, it makes sense to compute the area of houses as a feature