ML Foundations: 18 flashcards

It's like teaching a computer by showing it many examples, so it can learn to do things on its own, just like a child learns to recognize animals by seeing pictures.

In traditional programming, we write rules and the computer follows them. In machine learning, we give the computer data and it finds the rules itself.

A feature is a piece of information about the data that helps the computer make a decision. For example, if we want to predict if it will rain, features might be humidity and temperature.

A label is the answer we want the computer to predict. For example, in a spam filter, the label is 'spam' or 'not spam'.

Training data is the set of examples we give the computer to learn from. It includes both the features and the labels (if supervised).

Test data is a separate set of examples the computer has never seen, used to check how well it learned. It's important because it shows if the computer can handle new situations, not just memorize.

Supervised learning is when we teach the computer with labeled examples, like showing it pictures of apples and oranges with the names written on them.

Unsupervised learning is when we give the computer data without labels, and it finds patterns on its own, like grouping similar items together.

Reinforcement learning is when the computer learns by trying things and getting rewards or penalties, like a dog learning tricks with treats.

A model is the result of learning from data. It's like a recipe the computer uses to make predictions on new data.

Overfitting is when the model learns the training data too well, including the noise, and fails on new data. It's like memorizing answers instead of understanding the subject.

Underfitting is when the model is too simple and doesn't learn enough from the training data, so it makes poor predictions even on the training data.

Accuracy is the percentage of correct predictions the model makes on the test data. For example, if it predicts 90 out of 100 correctly, accuracy is 90%.

A prediction is the output the model gives for new input data. For example, predicting whether an email is spam or not.

A spam filter in email is a good example. It learns from emails you mark as spam and not spam, and then predicts which new emails are spam.

Classification predicts a category (like 'cat' or 'dog'), while regression predicts a number (like the price of a house).

A dataset is a collection of examples, usually with features and sometimes labels, used to train and test a model.

We split data to make sure the model learns from one set and is evaluated on a different set. This helps us know if the model can generalize to new data, not just memorize.