Skip to main content

Linear Regression

caution

This design is not standardized and currently unstable.

We implement the least-squares linear regression algorithm. Linear regression is a popular machine learning algorithm that models the relationship between two variables as linear. In a simple linear regression, the linear regression line has an equation of the form Y = a + b ∗ X where X is an explanatory variable and Y is a dependent variable.

While more efficient implementations of least-squares linear regression can easily be achieved, we show a simple implementation for ease of readability. For convenience, we define a Point circuit to abstract the notion of a two-dimensional point, as shown in Figure 15. Figure 16 shows the code used to instantiate the least-squares linear regression circuit.

circuit Point {
x: i32,
y: i32,
function new(x: i32, y: i32) -> Self {
return Self { x, y };
}
}
circuit LinearRegression {
points: [Point; 5],
// Instantiates a linear regression circuit.
function new(points: [Point; 5]) -> Self {
return Self { points };
}
// Return the slope of the linear regression.
function slope(self) -> i32 {
let num_points = 5i32;
// Calculate the sums.
let x_sum = 0i32;
let y_sum = 0i32;
let xy_sum = 0i32;
let x2_sum = 0i32;
for i in 0..5 {
x_sum += self.points[i].x;
y_sum += self.points[i].y;
xy_sum += self.points[i].x * self.points[i].y;
x2_sum += self.points[i].x * self.points[i].x;
}
let numerator = (num_points * xy_sum) - (x_sum * y_sum);
let denominator = (num_points * x2_sum) - (x_sum * x_sum);
let slope = numerator / denominator;
return slope;
}
// Return the offset of the linear regression.
function offset(self, slope: i32) -> i32 {
let num_points = 5i32;
// Calculate the sum.
let x_sum = 0i32;
let y_sum = 0i32;
for i in 0..5 {
x_sum += self.points[i].x;
y_sum += self.points[i].y;
}
return (y_sum - slope * x_sum) / num_points;
}
}