[Machine Learning]공부해보기

 

 

harvard university

 

Notice that these terms vary among courses, textbooks, Often prediction is used for both categorical and continuous outcomesand the term regression is sometimes used for the continuous case.

예측은 범주형 결과와 연속형 결과 모두에 사용되는 경우가 많으며 회귀라는 용어는 연속적인 경우에 사용되기도 합니다.

 

Categorical outcome

Y can be any of K classes.
The number of classes can vary.

For example with digit readers K is 10.
The classes besing the digits 0, 1, 2, 3, all the way up to 9.

When the outcome is categorical, we refer to the machine learning task as classification.

 

When the outcome is categorical, we refer to the machine learning task as classification.
And the main output of the model will be a decision rule
which prescribes which of the k classes we should predict.
In this scenario, most models provide a function of the predictors for each class.
These are used to make the decision.
When the data is binary, a typical decision rule looks like this.
If f1 of x1 through xp is bigger than some constant,
some pre-defined constant C, we predict category 1.
And if it's less than C, then we predict category 0.
Because the outcomes are categorical, our predictions
will be either right or wrong.

Contiunious outcome

We have a series of features and an unknown outcome we want to predict.
When the output is continuous, we refer to the machine learning task as prediction.

 

So we want the prediction y hat to match the actual outcome y as close as possible.
Because our outcome is continuous, our prediction y hat will not be either exactly right or wrong, but instead we'll determine an error defined as the difference between the prediction and the actual outcome.

 

 

 

 

 

솔직히 이해 안됨

 

From

Harvard University