Kernel API simplifications and SVMs improvements (#80)

* moved kernel transform inside fit for svms

* removed dataset from kernel, moved w_sum in svm

* svm stores only support vectors

* fixed alpha calculation for linear kernel regression

* keep only kernel method in svm

* adapted other crates to new kernel definition

* addresses pr comments

linfa-hierarchical/src/lib.rs

@@ -57,13 +57,13 @@ impl<F: Float> HierarchicalCluster<F> {
     }
 }
 
-impl<'b: 'a, 'a, F: Float> Transformer<Kernel<'a, F>, DatasetBase<Kernel<'a, F>, Vec<usize>>>
+impl<F: Float> Transformer<Kernel<F>, DatasetBase<Kernel<F>, Vec<usize>>>
     for HierarchicalCluster<F>
 {
     /// Perform hierarchical clustering of a similarity matrix
     ///
     /// Returns the class id for each data point
-    fn transform(&self, kernel: Kernel<'a, F>) -> DatasetBase<Kernel<'a, F>, Vec<usize>> {
+    fn transform(&self, kernel: Kernel<F>) -> DatasetBase<Kernel<F>, Vec<usize>> {
         // ignore all similarities below this value
         let threshold = F::from(1e-6).unwrap();
 
@@ -129,17 +129,14 @@ impl<'b: 'a, 'a, F: Float> Transformer<Kernel<'a, F>, DatasetBase<Kernel<'a, F>,
     }
 }
 
-impl<'a, F: Float, T: Targets>
-    Transformer<DatasetBase<Kernel<'a, F>, T>, DatasetBase<Kernel<'a, F>, Vec<usize>>>
+impl<F: Float, T: Targets>
+    Transformer<DatasetBase<Kernel<F>, T>, DatasetBase<Kernel<F>, Vec<usize>>>
     for HierarchicalCluster<F>
 {
     /// Perform hierarchical clustering of a similarity matrix
     ///
     /// Returns the class id for each data point
-    fn transform(
-        &self,
-        dataset: DatasetBase<Kernel<'a, F>, T>,
-    ) -> DatasetBase<Kernel<'a, F>, Vec<usize>> {
+    fn transform(&self, dataset: DatasetBase<Kernel<F>, T>) -> DatasetBase<Kernel<F>, Vec<usize>> {
         //let Dataset { records, .. } = dataset;
         self.transform(dataset.records)
     }

linfa-kernel/src/inner.rs

@@ -6,6 +6,8 @@ use serde_crate::{Deserialize, Serialize};
 use sprs::{CsMat, CsMatView};
 use std::ops::Mul;
 
+/// Specifies the methods an inner matrix of a kernel must
+/// be able to provide
 pub trait Inner {
     type Elem: Float;
 
@@ -18,6 +20,8 @@ pub trait Inner {
     fn diagonal(&self) -> Array1<Self::Elem>;
 }
 
+/// Allows a kernel to have either a dense or a sparse inner
+/// matrix in a way that is transparent to the user
 pub enum KernelInner<K1: Inner, K2: Inner> {
     Dense(K1),
     Sparse(K2),