An input is a measurable or encoded property supplied to a model so it can produce an output. In machine learning, inputs are often called features: quantifiable variables believed to bear on the outcome. A model represents an unknown target function by mapping input values x to an output y. section 7.2 section 1.2
Inputs must be numeric entities. A single property may be a scalar; several properties are commonly collected into a vector x = (x0, x1, …). One concrete set of values is an input instance. Subscripts can identify feature elements, while superscripts may distinguish separate instances. section 1.3.1 section 7.2
Input variables can be continuous or categorical. A continuous variable ranges over real values, such as a stock price; a categorical variable takes one of a finite set of values, including binary 0/1 choices. Both are expressed numerically before they enter the target function. section 7.2
A useful input workflow is:
- Identify quantifiable variables that can affect the desired outcome. section 7.2
- Encode the variables as scalars, vectors, matrices, or other numeric structures. section 7.2
- Optionally normalize a value from its known minimum–maximum interval into the range [0, 1]. section 1.3.1
- Pass the resulting input to a parameterized model, whose learned weights transform it into a prediction. section 1.1 section 1.2
Hardness and sharpness as inputs
A two-feature input can describe an object by hardness x0 and sharpness x1. A weighted-sum model combines those values with weights and a bias to produce a threat score. Low hardness and sharpness can yield “approach and purr,” intermediate values “ignore,” and high values “run away.” The example shows that the inputs are not rules by themselves; learned parameters determine how strongly each feature affects the output. section 1.1 section 1.3.1
| input: hardness = 0.01, sharpness = 0.02 → threat = —0.90 → decision: “approach and purr” input: hardness = 0.50, sharpness = 0.60 → threat = 0.01 → decision: “ignore” input: hardness = 0.99, sharpness = 0.97 → threat = 0.90 → decision: “run away” |
Compare three hardness-and-sharpness input instances with their computed threat scores and resulting decisions. section 1.1
Inputs play different roles across tasks. A stock-selling function can use purchase price, current price, and expert advice to choose between two outputs. A driving function can use speed limits and neighboring vehicles’ speeds to estimate a continuous speed. After training, the same model accepts new inputs during inferencing. section 7.2 section 7.2.3 section 1.3.6
