L7 - Classifiers, Perceptrons, MLP Flashcards
What is the difference between linear regression and logisitic regression?
Linear deals with predicting a continuous value. Logistic is a classification
process that classifies data into a category via binary classification.
In image classification, what training data issue might cause poor classification correctness?
Narrow training data. I.e training data where each data looks very similar
What type of learning is classification?
Supervised.
What is Logistic Regression? What does X and Y represent?
A binary classification algorithm whereby the model generates a probability of a data being either 1 or 0.
X represents continuous data, Y represents binary classification 1 or 0.
Explain how image classification using logisic regression would work to identify a start? Luminosity***
X is luminosity, Y is the classification. Depending on the luminosity, the model would predit whether the image
is a star.
What is the probability theshold in classification models?
A probability threshold hyperparameter we set that determines the certainty with which predictions are made. For example,
if we choose a threshold of 0.8, every data above 0.8 will be assigned 1, and all below will be 0.
Give examples of when we would specifically set a high or low probability threshold in logistic regression?
- High if we need high certainty, for example, detecting if cells are cancerous.
- Low if we can deal with less certainty.
Why do we tune the cost function in logistic regression?
Because we want to minimise the model error, resulting in greater classifcation capabilities.
How do we score the logistic regresion models accuracy?
Through a confusion matrix we can find the accuracy.
What is the equation for accuracy in a classification model?
TP + TN / TP + TN + FP + FN
If we have a graph area populated with data points, what is the ideal scenario for effective classification?
That the data points are linearly separable, enabling accurate classification.
What is the Perceptron? What is the inspiration for it?
It’s an algorithm used for classification.
Inspired by the way in which neurons operate.
How does the Perceptron algorithm work?
Takes K X data points as input. Multiplies each X by an associated weight and feeds into a threshold function.
Threshold function operates on the inputs. E.g Sum or product etc.
Outputs 1 or 0 based on whether the threshold function outputs a Y above or below the chosen threshold hyperparameter.
What does learning the weights in the Perceptron algorithm enable us to do?
Find the best classification line that separates the data.
What is the weakness of the Perceptron? What is the solution to this?
Doesn’t perform well on data that can’t be linearly separated.
Solution is the Multi-layer Perceptron.
