Machine Learning

Select the classification application:

Predict stock prices

find the largest number in the array

image recognition

calculate your tax return based on your annual income

image recognition

Please select the regression application:

predict the stock price

find the largest number in the array

image recognition

calculate your tax return based on your annual income 

predict the stock price

in what type of learning is the data labeled?

Supervised

in what type of learning is the training progress very simple but you need to use the whole dataset for testing

nonparameter

true or false: all dataset contains data and the corresponding label

false

Loss function and why we need it?

measures how well a machine learning models predictions match the actual target values
- gives model direction to move because it minimizes loss the model updates it parameters to make better predictions

• provides measurable goal. loss value will tell us how well or porrly model is performing during training

• drives optimization algorithms like gradient descent rely on the loss functons gradients to adjust parameters in right direction

describe gradient descent and why we need it?

optimization algorithm used to minimize the loss function in machine learning works by iteratively adjusting model parameters in direction that reduces the lss of the most - that direction is given by the negative gradient loss function

we need it because it proves an efficient way to find that minimum even for complex models

what is the difference between gradient descent and derivative?

Derivative tells you which way to move, the gradient descent uses that ignore to actually move and find best parameters

Explain the difference between supervised and unsupervised learning 

differ in whether labeled data us provided

What is an example of unsuprervised learning?

K-means

what is an example of supervised learning

knn

What is the difference between model and hyperparameters?

Both are important in machine learning but play different roles in how model learns and performs

define the training set

used to train model to learn model params

define a validation set

used to tune hyperparameter and evaluate model performance during training

define the test dataset

used after training and tuning are complete to asses final performance of model

why do we need test, validation and training datasets?

using all three helps ensure the model performs well mot jsut on tainging data but also on unseen real-world data

Matrix multiplication rules

• number of columsn in first factor must equal the number of rows in the second

  • A(m x n) * B (n x p) validation: n = n and result dimensions = m x p