Reduce allocations (#350)

* capacity for dataset

* replace map with mapv_into

* replace map with mapv_into again

* fix clippy clone_from

* returned iter + map back

src/composing/platt_scaling.rs

@@ -449,7 +449,7 @@ mod tests {
                 y.len(),
                 "The number of data points must match the number of output targets."
             );
-            *y = self.reg_vals.clone();
+            y.clone_from(&self.reg_vals);
         }
 
         fn default_target(&self, x: &Array2<f32>) -> Array1<f32> {

src/dataset/impl_dataset.rs

@@ -603,13 +603,13 @@ where
     )> {
         let targets = self.as_targets();
         let fold_size = targets.len() / k;
-        let mut res = Vec::new();
 
         // Generates all k folds of records and targets
         let mut records_chunks: Vec<_> =
             self.records.axis_chunks_iter(Axis(0), fold_size).collect();
         let mut targets_chunks: Vec<_> = targets.axis_chunks_iter(Axis(0), fold_size).collect();
 
+        let mut res = Vec::with_capacity(k);
         // For each iteration, take the first chunk for both records and targets as the validation set and
         // concatenate all the other chunks to create the training set. In the end swap the first chunk with the
         // one in the next index so that it is ready for the next iteration
@@ -742,7 +742,7 @@ where
         let targets = self.ntargets();
         let tshape = self.targets.raw_dim();
 
-        let mut objs: Vec<O> = Vec::new();
+        let mut objs: Vec<O> = Vec::with_capacity(k);
 
         {
             let records_sl = self.records.as_slice_mut().unwrap();

src/metrics_regression.rs

@@ -32,7 +32,7 @@ pub trait SingleTargetRegression<F: Float, T: AsSingleTargets<Elem = F>>:
     fn mean_absolute_error(&self, compare_to: &T) -> Result<F> {
         self.as_single_targets()
             .sub(&compare_to.as_single_targets())
-            .mapv(|x| x.abs())
+            .mapv_into(|x| x.abs())
             .mean()
             .ok_or(Error::NotEnoughSamples)
     }
@@ -41,7 +41,7 @@ pub trait SingleTargetRegression<F: Float, T: AsSingleTargets<Elem = F>>:
     fn mean_squared_error(&self, compare_to: &T) -> Result<F> {
         self.as_single_targets()
             .sub(&compare_to.as_single_targets())
-            .mapv(|x| x * x)
+            .mapv_into(|x| x * x)
             .mean()
             .ok_or(Error::NotEnoughSamples)
     }
@@ -58,7 +58,7 @@ pub trait SingleTargetRegression<F: Float, T: AsSingleTargets<Elem = F>>:
         let mut abs_error = self
             .as_single_targets()
             .sub(&compare_to.as_single_targets())
-            .mapv(|x| x.abs())
+            .mapv_into(|x| x.abs())
             .to_vec();
         abs_error.sort_by(|a, b| a.partial_cmp(b).unwrap());
         let mid = abs_error.len() / 2;
@@ -74,8 +74,8 @@ pub trait SingleTargetRegression<F: Float, T: AsSingleTargets<Elem = F>>:
     fn mean_absolute_percentage_error(&self, compare_to: &T) -> Result<F> {
         self.as_single_targets()
             .sub(&compare_to.as_single_targets())
-            .div(&self.as_single_targets())
-            .mapv(|x| x.abs())
+            .div(self.as_single_targets())
+            .mapv_into(|x| x.abs())
             .mean()
             .ok_or(Error::NotEnoughSamples)
     }
@@ -95,7 +95,7 @@ pub trait SingleTargetRegression<F: Float, T: AsSingleTargets<Elem = F>>:
             - self
                 .as_single_targets()
                 .sub(&single_target_compare_to)
-                .mapv(|x| x * x)
+                .mapv_into(|x| x * x)
                 .sum()
                 / (single_target_compare_to
                     .mapv(|x| (x - mean) * (x - mean))
@@ -114,7 +114,7 @@ pub trait SingleTargetRegression<F: Float, T: AsSingleTargets<Elem = F>>:
         let mean_error = diff.mean().ok_or(Error::NotEnoughSamples)?;
 
         Ok(F::one()
-            - (diff.mapv(|x| x * x).sum() - mean_error)
+            - (diff.mapv_into(|x| x * x).sum() - mean_error)
                 / (single_target_compare_to
                     .mapv(|x| (x - mean) * (x - mean))
                     .sum()