Introduction - Machine Learning

Field of study that gives computers the ability to learn without being explicitly programmed

• Traditional programming takes Input and Program to generate an Output
• Training set given to an Algorithm which generates a Model

• One vs All
  • Generate test cases equal to the number of classes
  • Generate classifier/model
  • Use test cases to obtain responses
  • Discard all negative responses
  • Response with highest probability is selected
• One vs One
  • If "N" classes then generate number of Classifier equal to N(N-1)/2
  • Classifier is generated for pair of Classes

• 
• False Positive (FP) => Type 1 Error
• False Negative (FN) => Type 2 Error
• Accuracy
  • Not a good metric to evaluate
• Precision
• Recall
• Precision Recall Tradeoff
  • If one increases then the other decreases
• F1 Score
  • 2 * Precision * Recall/Precision + Recall

• Dimension represents the number of Attributes/Features used to build the Model
• Accuracy of Model keeps on increasing with increase in Dimension upto a certain Threshold after that accuracy starts decreasing as irrelevant attributes will get used

• Content Based
  • Item similarity
• Collaborative Filtering
  • User similarity

• Batch Learning (Offline Learning)
  • System only learns once
  • Uses all available data, Generally done offline
  • Trained before being deployed, After Deploying does not learns more
• Online Learning
  • System learns incrementally, Generally done online
    • Continuous data flow
  • It is feeded with individual data or mini batches
    • Batch Size => Data considered to calculate Loss function
    • Mini Batch => Reduce batch size
  • Each learning step is fast, Keeps learning even after being deployed as new data arrives

• Euclidean Distance (l₂ norm)
  • 
• Manhattan Distance (l₁ norm)
• Generalized (l_n norm)
  •