Week 1 Lecture 5: Tutorial 1

Week 1 Lecture 5: Tutorial 1

Agenda

• Supervised vs Unsupervised Learning
• Different types of Features: Categorical vs Contiious Features
• Supervised Learning
  • Regression vs Classification
• Bias vs Varience
• Generalisation Performance of A Learning Algorithm

Supervised vs Unsupervised Learning

Basic algorithms to follow Suppose we have a huge number of images (1 Million, say)

1. Clustering Done -> Get a broad idea about what are the different kinds of images
2. Classifier is run for each of these clusters to discover intricacies in this data

So step 1 is unsupervised, while step 2 may be supervised or unsupervised.

Categorical vs Continious

Categorical

Finite number of Values, or indicating presence and absence of something

//Imp for exam purpose these categories

Continious

Can theoretically take infinie number of values, eg Height, weight, price, etc

Types of Supervised Learning Algo

Dependent on type of output variable

(Check if its discrete or continious)

• Regression
• Given certain features of car, predict price of car
• Classification
• Is it categorical or discrete?
• examples
• given images of animals predict species of animal
• given CT Scans, predict malignant tumor

Bias vs Variance

Bias

• Set of erroneous assumptions in the learning algorithm
• This is only due to the learning algo, and not the training examples

Variance

• Due to sensitivity of learning towards noise as opposed to the features and relationship of input/output
• Happens if too many features are being considered
• Happens if too less training data

Generalisation of Performance

How good does the LA perform when given new training examples? This can be controlled by controlling the bias and variance.

Programming Assignment 1 [week2] of Ng

Solution by Suvam: https://github.com/YedMavus/ML-Programs/blob/main/AndrewNg-ex1/figure1.gif