use super::Classifier;
use crate::prelude::Predictable;
use crate::prelude::Trainable;
use linfa::prelude::{Fit, Predict};
use linfa::Dataset;
use linfa::Float;
use linfa::Label;
use linfa_bayes::GaussianNb;
use ndarray::Array1;
use ndarray::ArrayView1;
use ndarray::ArrayView2;
#[derive(Clone, Default, Debug, PartialEq)]
pub struct BayesClassifier<X, Y>
where
X: Float,
Y: Label,
{
model: Option<GaussianNb<X, Y>>,
}
impl<X, Y> BayesClassifier<X, Y>
where
X: Float,
Y: Label,
{
pub fn new() -> Self {
BayesClassifier { model: None }
}
}
impl<X, Y> Trainable<X, Y> for BayesClassifier<X, Y>
where
X: Float,
Y: Label,
{
fn fit(&mut self, x: &ArrayView2<X>, y: &ArrayView1<Y>) -> Result<(), String> {
let dataset = Dataset::new(x.to_owned(), y.to_owned());
let model = GaussianNb::<X, Y>::params().fit(&dataset).unwrap();
self.model = Some(model);
Ok(())
}
}
impl<X, Y> Predictable<X, Y> for BayesClassifier<X, Y>
where
X: Float,
Y: Label,
{
fn predict(&self, x: &ArrayView2<X>) -> Result<Array1<Y>, String> {
Ok(self.model.as_ref().unwrap().predict(x))
}
}
impl<X, Y> Classifier<X, Y> for BayesClassifier<X, Y>
where
X: Float,
Y: Label,
{
}
#[cfg(test)]
mod tests {
use super::*;
use linfa::prelude::ToConfusionMatrix;
use ndarray::{arr1, arr2};
#[test]
fn test_bayes_classifier() {
let mut classifier = BayesClassifier::new();
let x = arr2(&[[1., 1.], [1., 2.], [2., 1.], [2., 2.]]);
let y = arr1(&[1, 1, 0, 0]);
assert!(classifier.fit(&x.view(), &y.view()).is_ok());
let x_pred = arr2(&[[1., 1.], [1., 2.], [2., 1.], [2., 2.]]);
let y_pred = arr1(&[1, 1, 0, 0]);
let y_pred_res = classifier.predict(&x_pred.view()).unwrap();
assert_eq!(y_pred, y_pred_res);
let predicted_y = classifier.predict(&x.view()).unwrap();
let cm = predicted_y.confusion_matrix(&y).unwrap();
println!("{:?}", cm);
assert!(cm.precision() > 0.8);
assert!(cm.accuracy() > 0.5);
}
#[test]
fn test_partial_eq() {
let classifier = BayesClassifier::<f32, usize>::new();
let classifier2 = classifier.clone();
assert!(classifier.eq(&classifier2));
}
}