work/smartcore
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'development' into kmeans-with-fastpair

Browse Source
This commit is contained in:
Lorenzo (Mec-iS)
2023-05-04 17:19:01 +01:00
parent bf65fe3753 2d7c055154
commit 13bb222ca7
8 changed files with 96 additions and 95 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.github/CONTRIBUTING.md
+2
View File
@@ -37,6 +37,8 @@ $ rust-code-analysis-cli -p src/algorithm/neighbour/fastpair.rs --ls 22 --le 213
```
* find more information about what happens in your binary with [`twiggy`](https://rustwasm.github.io/twiggy/install.html). This need a compiled binary so create a brief `main {}` function using `smartcore` and then point `twiggy` to that file.
* Please take a look to the output of a profiler to spot most evident performance problems, see [this guide about using a profiler](http://www.codeofview.com/fix-rs/2017/01/24/how-to-optimize-rust-programs-on-linux/).
## Issue Report Process
1. Go to the project's issues.
Cargo.toml
+1 -1
View File
@@ -2,7 +2,7 @@
name = "smartcore"
description = "Machine Learning in Rust."
homepage = "https://smartcorelib.org"
version = "0.3.1"
version = "0.3.2"
authors = ["smartcore Developers"]
edition = "2021"
license = "Apache-2.0"
src/linalg/basic/matrix.rs
+6 -6
View File
@@ -431,9 +431,9 @@ impl<T: Number + RealNumber> SVDDecomposable<T> for DenseMatrix<T> {}
impl<'a, T: Debug + Display + Copy + Sized> Array<T, (usize, usize)> for DenseMatrixView<'a, T> {
fn get(&self, pos: (usize, usize)) -> &T {
if self.column_major {
&self.values[(pos.0 + pos.1 * self.stride)]
&self.values[pos.0 + pos.1 * self.stride]
} else {
&self.values[(pos.0 * self.stride + pos.1)]
&self.values[pos.0 * self.stride + pos.1]
}
}
@@ -495,9 +495,9 @@ impl<'a, T: Debug + Display + Copy + Sized> ArrayView1<T> for DenseMatrixView<'a
impl<'a, T: Debug + Display + Copy + Sized> Array<T, (usize, usize)> for DenseMatrixMutView<'a, T> {
fn get(&self, pos: (usize, usize)) -> &T {
if self.column_major {
&self.values[(pos.0 + pos.1 * self.stride)]
&self.values[pos.0 + pos.1 * self.stride]
} else {
&self.values[(pos.0 * self.stride + pos.1)]
&self.values[pos.0 * self.stride + pos.1]
}
}
@@ -519,9 +519,9 @@ impl<'a, T: Debug + Display + Copy + Sized> MutArray<T, (usize, usize)>
{
fn set(&mut self, pos: (usize, usize), x: T) {
if self.column_major {
self.values[(pos.0 + pos.1 * self.stride)] = x;
self.values[pos.0 + pos.1 * self.stride] = x;
} else {
self.values[(pos.0 * self.stride + pos.1)] = x;
self.values[pos.0 * self.stride + pos.1] = x;
}
}
src/linalg/basic/vector.rs
+20
View File
@@ -15,6 +15,25 @@ pub struct VecView<'a, T: Debug + Display + Copy + Sized> {
ptr: &'a [T],
}
impl<T: Debug + Display + Copy + Sized> Array<T, usize> for &[T] {
fn get(&self, i: usize) -> &T {
&self[i]
}
fn shape(&self) -> usize {
self.len()
}
fn is_empty(&self) -> bool {
self.len() > 0
}
fn iterator<'b>(&'b self, axis: u8) -> Box<dyn Iterator<Item = &'b T> + 'b> {
assert!(axis == 0, "For one dimensional array `axis` should == 0");
Box::new(self.iter())
}
}
impl<T: Debug + Display + Copy + Sized> Array<T, usize> for Vec<T> {
fn get(&self, i: usize) -> &T {
&self[i]
@@ -46,6 +65,7 @@ impl<T: Debug + Display + Copy + Sized> MutArray<T, usize> for Vec<T> {
}
impl<T: Debug + Display + Copy + Sized> ArrayView1<T> for Vec<T> {}
impl<T: Debug + Display + Copy + Sized> ArrayView1<T> for &[T] {}
impl<T: Debug + Display + Copy + Sized> MutArrayView1<T> for Vec<T> {}
src/model_selection/kfold.rs
+2 -6
View File
@@ -283,9 +283,7 @@ mod tests {
(vec![0, 1, 2, 3, 7, 8, 9], vec![4, 5, 6]),
(vec![0, 1, 2, 3, 4, 5, 6], vec![7, 8, 9]),
];
for ((train, test), (expected_train, expected_test)) in
k.split(&x).into_iter().zip(expected)
{
for ((train, test), (expected_train, expected_test)) in k.split(&x).zip(expected) {
assert_eq!(test, expected_test);
assert_eq!(train, expected_train);
}
@@ -307,9 +305,7 @@ mod tests {
(vec![0, 1, 2, 3, 7, 8, 9], vec![4, 5, 6]),
(vec![0, 1, 2, 3, 4, 5, 6], vec![7, 8, 9]),
];
for ((train, test), (expected_train, expected_test)) in
k.split(&x).into_iter().zip(expected)
{
for ((train, test), (expected_train, expected_test)) in k.split(&x).zip(expected) {
assert_eq!(test.len(), expected_test.len());
assert_eq!(train.len(), expected_train.len());
}
src/readers/csv.rs
+3 -9
View File
@@ -83,7 +83,7 @@ where
Matrix: Array2<T>,
{
let csv_text = read_string_from_source(source)?;
let rows: Vec<Vec<T>> = extract_row_vectors_from_csv_text::<T, RowVector, Matrix>(
let rows: Vec<Vec<T>> = extract_row_vectors_from_csv_text(
&csv_text,
&definition,
detect_row_format(&csv_text, &definition)?,
@@ -103,12 +103,7 @@ where
/// Given a string containing the contents of a csv file, extract its value
/// into row-vectors.
fn extract_row_vectors_from_csv_text<
'a,
T: Number + RealNumber + std::str::FromStr,
RowVector: Array1<T>,
Matrix: Array2<T>,
>(
fn extract_row_vectors_from_csv_text<'a, T: Number + RealNumber + std::str::FromStr>(
csv_text: &'a str,
definition: &'a CSVDefinition<'_>,
row_format: CSVRowFormat<'_>,
@@ -305,12 +300,11 @@ mod tests {
}
mod extract_row_vectors_from_csv_text {
use super::super::{extract_row_vectors_from_csv_text, CSVDefinition, CSVRowFormat};
use crate::linalg::basic::matrix::DenseMatrix;
#[test]
fn read_default_csv() {
assert_eq!(
extract_row_vectors_from_csv_text::<f64, Vec<_>, DenseMatrix<_>>(
extract_row_vectors_from_csv_text::<f64>(
"column 1, column 2, column3\n1.0,2.0,3.0\n4.0,5.0,6.0",
&CSVDefinition::default(),
CSVRowFormat {
src/svm/svc.rs
+55 -64
View File
@@ -322,19 +322,26 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX> + 'a, Y: Array
let (n, _) = x.shape();
let mut y_hat: Vec<TX> = Array1::zeros(n);
let mut row = Vec::with_capacity(n);
for i in 0..n {
let row_pred: TX =
self.predict_for_row(Vec::from_iterator(x.get_row(i).iterator(0).copied(), n));
row.clear();
row.extend(x.get_row(i).iterator(0).copied());
let row_pred: TX = self.predict_for_row(&row);
y_hat.set(i, row_pred);
}
Ok(y_hat)
}
fn predict_for_row(&self, x: Vec<TX>) -> TX {
fn predict_for_row(&self, x: &[TX]) -> TX {
let mut f = self.b.unwrap();
let xi: Vec<_> = x.iter().map(|e| e.to_f64().unwrap()).collect();
for i in 0..self.instances.as_ref().unwrap().len() {
let xj: Vec<_> = self.instances.as_ref().unwrap()[i]
.iter()
.map(|e| e.to_f64().unwrap())
.collect();
f += self.w.as_ref().unwrap()[i]
* TX::from(
self.parameters
@@ -343,13 +350,7 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX> + 'a, Y: Array
.kernel
.as_ref()
.unwrap()
.apply(
&x.iter().map(|e| e.to_f64().unwrap()).collect(),
&self.instances.as_ref().unwrap()[i]
.iter()
.map(|e| e.to_f64().unwrap())
.collect(),
)
.apply(&xi, &xj)
.unwrap(),
)
.unwrap();
@@ -472,14 +473,12 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
let tol = self.parameters.tol;
let good_enough = TX::from_i32(1000).unwrap();
let mut x = Vec::with_capacity(n);
for _ in 0..self.parameters.epoch {
for i in self.permutate(n) {
self.process(
i,
Vec::from_iterator(self.x.get_row(i).iterator(0).copied(), n),
*self.y.get(i),
&mut cache,
);
x.clear();
x.extend(self.x.get_row(i).iterator(0).take(n).copied());
self.process(i, &x, *self.y.get(i), &mut cache);
loop {
self.reprocess(tol, &mut cache);
self.find_min_max_gradient();
@@ -511,24 +510,17 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
let mut cp = 0;
let mut cn = 0;
let mut x = Vec::with_capacity(n);
for i in self.permutate(n) {
x.clear();
x.extend(self.x.get_row(i).iterator(0).take(n).copied());
if *self.y.get(i) == TY::one() && cp < few {
if self.process(
i,
Vec::from_iterator(self.x.get_row(i).iterator(0).copied(), n),
*self.y.get(i),
cache,
) {
if self.process(i, &x, *self.y.get(i), cache) {
cp += 1;
}
} else if *self.y.get(i) == TY::from(-1).unwrap()
&& cn < few
&& self.process(
i,
Vec::from_iterator(self.x.get_row(i).iterator(0).copied(), n),
*self.y.get(i),
cache,
)
&& self.process(i, &x, *self.y.get(i), cache)
{
cn += 1;
}
@@ -539,7 +531,7 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
}
}
fn process(&mut self, i: usize, x: Vec<TX>, y: TY, cache: &mut Cache<TX, TY, X, Y>) -> bool {
fn process(&mut self, i: usize, x: &[TX], y: TY, cache: &mut Cache<TX, TY, X, Y>) -> bool {
for j in 0..self.sv.len() {
if self.sv[j].index == i {
return true;
@@ -551,15 +543,14 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
let mut cache_values: Vec<((usize, usize), TX)> = Vec::new();
for v in self.sv.iter() {
let xi: Vec<_> = v.x.iter().map(|e| e.to_f64().unwrap()).collect();
let xj: Vec<_> = x.iter().map(|e| e.to_f64().unwrap()).collect();
let k = self
.parameters
.kernel
.as_ref()
.unwrap()
.apply(
&v.x.iter().map(|e| e.to_f64().unwrap()).collect(),
&x.iter().map(|e| e.to_f64().unwrap()).collect(),
)
.apply(&xi, &xj)
.unwrap();
cache_values.push(((i, v.index), TX::from(k).unwrap()));
g -= v.alpha * k;
@@ -578,7 +569,7 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
cache.insert(v.0, v.1.to_f64().unwrap());
}
let x_f64 = x.iter().map(|e| e.to_f64().unwrap()).collect();
let x_f64: Vec<_> = x.iter().map(|e| e.to_f64().unwrap()).collect();
let k_v = self
.parameters
.kernel
@@ -701,8 +692,10 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
let km = sv1.k;
let gm = sv1.grad;
let mut best = 0f64;
let xi: Vec<_> = sv1.x.iter().map(|e| e.to_f64().unwrap()).collect();
for i in 0..self.sv.len() {
let v = &self.sv[i];
let xj: Vec<_> = v.x.iter().map(|e| e.to_f64().unwrap()).collect();
let z = v.grad - gm;
let k = cache.get(
sv1,
@@ -711,10 +704,7 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
.kernel
.as_ref()
.unwrap()
.apply(
&sv1.x.iter().map(|e| e.to_f64().unwrap()).collect(),
&v.x.iter().map(|e| e.to_f64().unwrap()).collect(),
)
.apply(&xi, &xj)
.unwrap(),
);
let mut curv = km + v.k - 2f64 * k;
@@ -732,6 +722,12 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
}
}
let xi: Vec<_> = self.sv[idx_1]
.x
.iter()
.map(|e| e.to_f64().unwrap())
.collect::<Vec<_>>();
idx_2.map(|idx_2| {
(
idx_1,
@@ -742,16 +738,12 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
.as_ref()
.unwrap()
.apply(
&self.sv[idx_1]
.x
.iter()
.map(|e| e.to_f64().unwrap())
.collect(),
&xi,
&self.sv[idx_2]
.x
.iter()
.map(|e| e.to_f64().unwrap())
.collect(),
.collect::<Vec<_>>(),
)
.unwrap()
}),
@@ -765,8 +757,11 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
let km = sv2.k;
let gm = sv2.grad;
let mut best = 0f64;
let xi: Vec<_> = sv2.x.iter().map(|e| e.to_f64().unwrap()).collect();
for i in 0..self.sv.len() {
let v = &self.sv[i];
let xj: Vec<_> = v.x.iter().map(|e| e.to_f64().unwrap()).collect();
let z = gm - v.grad;
let k = cache.get(
sv2,
@@ -775,10 +770,7 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
.kernel
.as_ref()
.unwrap()
.apply(
&sv2.x.iter().map(|e| e.to_f64().unwrap()).collect(),
&v.x.iter().map(|e| e.to_f64().unwrap()).collect(),
)
.apply(&xi, &xj)
.unwrap(),
);
let mut curv = km + v.k - 2f64 * k;
@@ -797,6 +789,12 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
}
}
let xj: Vec<_> = self.sv[idx_2]
.x
.iter()
.map(|e| e.to_f64().unwrap())
.collect();
idx_1.map(|idx_1| {
(
idx_1,
@@ -811,12 +809,8 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
.x
.iter()
.map(|e| e.to_f64().unwrap())
.collect(),
&self.sv[idx_2]
.x
.iter()
.map(|e| e.to_f64().unwrap())
.collect(),
.collect::<Vec<_>>(),
&xj,
)
.unwrap()
}),
@@ -835,12 +829,12 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
.x
.iter()
.map(|e| e.to_f64().unwrap())
.collect(),
.collect::<Vec<_>>(),
&self.sv[idx_2]
.x
.iter()
.map(|e| e.to_f64().unwrap())
.collect(),
.collect::<Vec<_>>(),
)
.unwrap(),
)),
@@ -895,7 +889,10 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
self.sv[v1].alpha -= step.to_f64().unwrap();
self.sv[v2].alpha += step.to_f64().unwrap();
let xi_v1: Vec<_> = self.sv[v1].x.iter().map(|e| e.to_f64().unwrap()).collect();
let xi_v2: Vec<_> = self.sv[v2].x.iter().map(|e| e.to_f64().unwrap()).collect();
for i in 0..self.sv.len() {
let xj: Vec<_> = self.sv[i].x.iter().map(|e| e.to_f64().unwrap()).collect();
let k2 = cache.get(
&self.sv[v2],
&self.sv[i],
@@ -903,10 +900,7 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
.kernel
.as_ref()
.unwrap()
.apply(
&self.sv[v2].x.iter().map(|e| e.to_f64().unwrap()).collect(),
&self.sv[i].x.iter().map(|e| e.to_f64().unwrap()).collect(),
)
.apply(&xi_v2, &xj)
.unwrap(),
);
let k1 = cache.get(
@@ -916,10 +910,7 @@ impl<'a, TX: Number + RealNumber, TY: Number + Ord, X: Array2<TX>, Y: Array1<TY>
.kernel
.as_ref()
.unwrap()
.apply(
&self.sv[v1].x.iter().map(|e| e.to_f64().unwrap()).collect(),
&self.sv[i].x.iter().map(|e| e.to_f64().unwrap()).collect(),
)
.apply(&xi_v1, &xj)
.unwrap(),
);
self.sv[i].grad -= step.to_f64().unwrap() * (k2 - k1);
src/svm/svr.rs
+7 -9
View File
@@ -248,19 +248,20 @@ impl<'a, T: Number + FloatNumber + PartialOrd, X: Array2<T>, Y: Array1<T>> SVR<'
let mut y_hat: Vec<T> = Vec::<T>::zeros(n);
let mut x_i = Vec::with_capacity(n);
for i in 0..n {
y_hat.set(
i,
self.predict_for_row(Vec::from_iterator(x.get_row(i).iterator(0).copied(), n)),
);
x_i.clear();
x_i.extend(x.get_row(i).iterator(0).copied());
y_hat.set(i, self.predict_for_row(&x_i));
}
Ok(y_hat)
}
pub(crate) fn predict_for_row(&self, x: Vec<T>) -> T {
pub(crate) fn predict_for_row(&self, x: &[T]) -> T {
let mut f = self.b;
let xi: Vec<_> = x.iter().map(|e| e.to_f64().unwrap()).collect();
for i in 0..self.instances.as_ref().unwrap().len() {
f += self.w.as_ref().unwrap()[i]
* T::from(
@@ -270,10 +271,7 @@ impl<'a, T: Number + FloatNumber + PartialOrd, X: Array2<T>, Y: Array1<T>> SVR<'
.kernel
.as_ref()
.unwrap()
.apply(
&x.iter().map(|e| e.to_f64().unwrap()).collect(),
&self.instances.as_ref().unwrap()[i],
)
.apply(&xi, &self.instances.as_ref().unwrap()[i])
.unwrap(),
)
.unwrap()