Basics

1. Basics
2. Overfitting

Basics

• Supervised Learning

Based on historical data that has labels, train a model to predict the values of labels (dependent variable) based on features. Example: Predict person income (labels) based on the age (feature)

• Two groups of supervised learning

  • Classification: train an algorithm to predict a dependent variable that is categorical (beloging to discrete, finite set of values)
    • Binary classification
    • Multiple classification
  • Regression: predict on continuous variable

• Recommendation engines:

  • Suggest based on the behavior
  • Ex: based on the rating, suggest to customer the product

• Unsupervised Learning: try to find a patterns or discover the underlying structure in a given data set

  • Clustering
  • Anomaly detection
  • Topic modeling

Overfitting

• Model is unable to generalize with new data

• Causes

  • Not enough training data
  • Too many feature (Too complex model)
  • Model fited too necessary features Noise

• Solutions

  • Mode data
  • Make model less complex
  • Remove noise from the model by increasing the regularization parameters

• What's regularization ?

• Wide neural and Deep neural ?

  • Wide models are used for memorization. Deep models are for generalization
  • Deep and wide models are ideal for a recommendation application.