Link: Machine learning

What’s supervised learning

Supervised learning algorithms are trained with labels, meaning the correct output is given in the train data E.g. Negative or Positive review

How does the model work

The neural networks learns by comparing its actual outputs vs correct outputs, and they adjust the models to reduce the error rate.

Usage

Good for predicting future events using historical data

Flowchart

Method of split data

Split data to 3 sets to avoid using test data to validate the result. Note: on this course, it’s simplified by only use single train/test split

Training data

Used to train model parameters

Validation data

To determine what model hyper-parameters to adjust

Test data

To get final performance metric. This also means we cannot go back to adjust model parameters after using model on the test set. Whatever the performance metric is, it’s the final measure.