pub trait Estimator {
fn n_features_in(&self) -> usize;
}
use crate::data::{DenseMatrix, MatrixView};
use super::ModelError;
pub trait Classifier: Estimator {
fn classes(&self) -> &[u8];
fn predict_into(&self, data: &MatrixView<'_>, output: &mut [u8]) -> Result<(), ModelError>;
fn predict(&self, data: &MatrixView<'_>) -> Result<Vec<u8>, ModelError> {
check_batch_width(self.n_features_in(), data)?;
let mut output = vec![0; data.rows()];
self.predict_into(data, &mut output)?;
Ok(output)
}
}
pub trait ProbabilisticClassifier: Classifier {
fn predict_proba_into(
&self,
data: &MatrixView<'_>,
output: &mut [f32],
) -> Result<(), ModelError>;
fn predict_proba(&self, data: &MatrixView<'_>) -> Result<Vec<f32>, ModelError> {
check_batch_width(self.n_features_in(), data)?;
let output_len = data.rows().checked_mul(self.classes().len()).ok_or(
ModelError::OutputShapeOverflow {
rows: data.rows(),
columns: self.classes().len(),
},
)?;
let mut output = vec![0.0; output_len];
self.predict_proba_into(data, &mut output)?;
Ok(output)
}
fn predict_class_proba_into(
&self,
data: &MatrixView<'_>,
class: u8,
output: &mut [f32],
) -> Result<(), ModelError>;
fn predict_class_proba(
&self,
data: &MatrixView<'_>,
class: u8,
) -> Result<Vec<f32>, ModelError> {
check_batch_width(self.n_features_in(), data)?;
let mut output = vec![0.0; data.rows()];
self.predict_class_proba_into(data, class, &mut output)?;
Ok(output)
}
}
pub trait Regressor: Estimator {
fn predict_into(&self, data: &MatrixView<'_>, output: &mut [f32]) -> Result<(), ModelError>;
fn predict(&self, data: &MatrixView<'_>) -> Result<Vec<f32>, ModelError> {
check_batch_width(self.n_features_in(), data)?;
let mut output = vec![0.0; data.rows()];
self.predict_into(data, &mut output)?;
Ok(output)
}
}
pub trait Transformer: Estimator {
fn n_features_out(&self) -> usize;
fn transform_into<'output>(
&self,
data: &MatrixView<'_>,
output: &'output mut [f32],
) -> Result<MatrixView<'output>, ModelError>;
fn transform(&self, data: &MatrixView<'_>) -> Result<DenseMatrix, ModelError> {
check_batch_width(self.n_features_in(), data)?;
let columns = self.n_features_out();
let output_len =
data.rows()
.checked_mul(columns)
.ok_or(ModelError::OutputShapeOverflow {
rows: data.rows(),
columns,
})?;
let mut output = vec![0.0; output_len];
let transformed = self.transform_into(data, &mut output)?;
validate_transformed_shape(data.rows(), columns, &transformed)?;
Ok(transformed.to_dense())
}
}
fn check_batch_width(expected: usize, data: &MatrixView<'_>) -> Result<(), ModelError> {
if data.columns() != expected {
return Err(ModelError::FeatureDimension {
expected,
actual: data.columns(),
});
}
Ok(())
}
pub(crate) fn validate_transformed_shape(
expected_rows: usize,
expected_columns: usize,
transformed: &MatrixView<'_>,
) -> Result<(), ModelError> {
if transformed.rows() != expected_rows {
return Err(ModelError::OutputLength {
expected: expected_rows,
actual: transformed.rows(),
});
}
if transformed.columns() != expected_columns {
return Err(ModelError::FeatureDimension {
expected: expected_columns,
actual: transformed.columns(),
});
}
Ok(())
}
pub trait HasParams {
type Params;
fn get_params(&self) -> &Self::Params;
}