ML Flashcards
What are features and labels?
Features: Values that usefully characterize the things we wish to classify.
Labels: The one feature that is the outcome of the training data.
Name the three types of Machine Learning (plus a very short explanation).
-> Supervised machine learning algorithms
° the possible outcomes are already known and the training data is labeled with the correct answers
-> Unsupervised machine learning algorithms
° the training data will have no correct answer or no specific outcome or label. Algorithms help to discover interesting patterns in data.
->Reinforcement machine learning algorithms
° the machine is exposed to an environment where it trains itself continually using the trial and error method to make accurate decisions
What are the seven steps of Machine Learning (plus a very short explanation).
- Data collection
collecting training data - Data preparation
de-duplicated and errors need to be removed - Choosing a model
the model needs to meet the business goal - Training
use your training data and incrementally improve the predictions of the model - Evaluation
testing the machine learning against an unused control dataset - Parameter tuning
test the originally set parameters to improve the AI - Prediction
answer questions using predictions
What are the two kinds of problems that can be solved using supervised learning (plus a very short explanation)?
-> Classification
we have categorized output such as “black”, “teaching”, “non-teaching”
-> Regression
we have real value output such as “distance”, “kilogram”
Give al ML techniques and there kind of problems
-> supervised learning
- classification
- regression
-> unsupervised learning
- Clustering (simular data)
- Anomaly dectection (unusual detection)
- Association (interesting relations)
Using linear regression, the system estimates a regression function with the equation 𝑓(𝑥) = 𝑏 + m𝑥. Can you explain the values of b and m?
b
the point where the estimated regression line crosses the 𝑦 axis
m
determines the slope of the estimated regression line
Can you explain the terms bias and variance? Is bias, resp. variance low or high for the straight line / squiggly line?
bias: it describes how well the model matches the training data set
variance: the changes in the model when using different portions of the training data set
straight line: bias high and variance low
squiggly line: bias low and variance high
Name the three types of recommender systems (plus a short explanation).
- Popularity based
houd enkel rekening met de populariteit in het algemeen om een aanbeveling te doen - Content based
it analyses the content and finds similar content - Collaborative filtering
find similar users and recommend something the other user liked/watched
Name three companies that use recommender systems intensively. What are the items they recommend?
Netflix -> movies
Amazon -> products to buy eg. books
facebook -> people you might know: friends
Name the three challenges of collaborative filtering. Can you explain them?
Data sparsity
-> Users in general rate only a limited number of items
Cold start
-> Difficulty in recommendations in new users or new items
Scalability
-> Increase in number of users or items
Explain the following term:
entropy
een maat voor de waarschijnlijkheid van een bepaalde verdeling
Explain the following term:
information gain
the amount of information gained about a random variable
Explain the following term:
leaf node
end node
Explain the following term:
decision node
node waarin een beslissing van verdeling gemaakt wordt
Explain the following term:
root node
Eerste node
Name one advantage and one disadvantage of decision trees.
+ easy to understand
- overfitting is quite common
What are the four steps in the Random Forest algorithm?
- Select random samples from a given dataset (= bootstrapped datasets).
- Construct a decision tree for each sample (using only a random subset of variables) and get a prediction result from each decision tree.
- Perform a vote for each predicted result.
- Select the prediction result with the most votes as the final prediction.
Explain, using the dog analogy, how reinforcement learning works.
- een hond is een agent in een omgeving. De omgeving kan je huis zijn.
- De situaties die de hond tegenkomt is een state. vb: een hond staat en er word een bepaald commando op een bapaalde toon gegeven in de living.
- De agent reageerd door een actie uit te voeren om van de ene state over te gaan naar een andere state, de hond gaat bijvoorbeeld van staan naar zitten.
- Na de overgang kan de agent een beloning of een straf terugkrijgen. De hond krijgt een traktatie of een nee
- Het beleid is de strategie van het kiezen van een actie gegeven een state in de verwachting van betere uitkomsten.
Why are there 500 different states in the taxi problem?
5×5×5×4 = 500
-5x5 grid
-5 possible locations for our passenger
-4 possible locations where we can drop of our passenger
What rewards and/or penalties are involved in the taxi problem?
- high positive reward for a successful dropoff
- penalized if it tries to drop off a passenger in wrong locations
- slight negative reward for not making it to the destination after every time-step
What are the six different actions that can be taken in the taxi problem?
- south
- north
- east
- west
- pickup
- dropoff
