Into Flashcards
(19 cards)
What is ML?
ML subset of AI techniques which learn from data
What is Supervised Learning?
Algorithm that learns a function using examples of inputs and outputs.
What is needed for supervised learning
labelled training data
Examples of supervised learning
- classification
- regression
What is unsupervised learning?
Algorithm that finds paterns in data without using labelled outputs
What does
Example of unsupervised learning
Clustering
What is classification
Predict discrete lavels or categories based on a group of features.
It has only two options (yes or no)
What is regression
Predict a continous calue based on a group of features
WHat are features?
- Measurable property or characteristic of a dataset
- they are inputs to ML models
List 2 learning strategies
- Lazy learning algorithms
- Eager learning algorithms
Lazy learning algorithms
- Computations are delayed until prediction.
- New data stored during training.
- Minimal training time, but prediction time is slow.
- e.g. K Nearest Neighbours
Eager learning algorithms
- A full model is built during the training phase.
- More time needed to train model, but prediction is very fast.
- e.g. Decision trees, Perceptron, SVM
What is overfitting?
implies that a model is too complex for the data
what is underfitting?
implies that a model is too simple for the data.
Which algorithm should I use?
- Number of input examples, N.
- Number of features (dimensions), D, which represent each
input example. - Number of target classes, M.
- Complexity: Simple models (linear or nearest neighbours) →
Bagging → Boosting → Neural networks - Predictive power
- Interpretability
- Stability
- Assumptions
Explain the difference between supervised and unsupervised learning?
Supervised:
* learns using examples of input or output
* required labelled data
Unsupervised:
* Finds patterns in data without using labelled data
What is the difference between an eager learning algorithm and a lazy learning algorithm?
Eager:
* Full model built during training phase
* Time needed to train model
* Prediction is fast
* E.g: SVM, perceptron, decision trees
Lazy:
* Computations delayed till training phase
* new data stored during training
* minimum training time, but predition time is slow
* E.g: KNN
Explain overfitting and underfitting.
Overfitting:
* Model is too complex for data
Underfitting:
* model is too simple for the data
