2.0.12

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "rstats"
-version = "2.0.11"
+version = "2.0.12"
 authors = ["Libor Spacek"]
 edition = "2021"
 description = "Statistics, Information Measures, Data Analysis, Linear Algebra, Clifford Algebra, Machine Learning, Geometric Median, Matrix Decompositions, PCA, Mahalanobis Distance, Hulls, Multithreading.."

README.md

@@ -326,7 +326,9 @@ Methods which take an additional generic vector argument, such as a vector of we
 
 ## Appendix: Recent Releases
 
-* **Version 2.0.11** - removed not so useful `variances`. Tidied up error processing in `vecveg.rs`. Added to it `serial_covar` and `serial_wcovar` for when heavy loading of all the cores may not be wanted.
+* **Version 2.0.12** - added `depth_ratio`.
+
+* **Version 2.0.11** - removed not so useful `variances`. Tidied up error processing in `vecvecg.rs`. Added to it `serial_covar` and `serial_wcovar` for when heavy loading of all the cores may not be wanted.
 
 * **Version 2.0.10** - Added to struct TriangMat `eigenvectors` (enabling PCA).
 
@@ -349,35 +351,3 @@ Methods which take an additional generic vector argument, such as a vector of we
 * **Version 2.0.1** - Added `TriangMat::dim()` and tidied up some comments.
 
 * **Version 2.0.0** - Renamed `MStats` -> `Params` and its variant `dispersion` -> `spread`. This may cause some backwards incompatibilities, hence the new major version. Added 'centre' as an argument to `dfdt`,`dvdt`,`wdvdt`, so that it does not have to be recomputed.
-
-* **Version 1.3.3** - Added `wdvdt`- individually weighted time series derivative (weighted arithmetic mean minus geometric median).
-
-* **Version 1.3.2** - Added `dvdt` - linearly weighted (approximate) time series derivative at the last point (present time). Similar to `dfdt` but works on vectors and returns a derivative vector. Changed error helper function `re_error` to return Result (Err variant), that can be more conveniently processed upstream with just the ? operator.
-
-* **Version 1.3.1** - Some more changes to the `hulls` fixed `wsigvec` to be consistent with `sigvec`.
-
-* **Version 1.3.0** - Renamed `t_stat -> tm_stat` and `t_statistic -> tm_statistic` to avoid potential confusion with classical t-statistic. Added `insideness` of `nd` points. Improved `hulls` algorithms and their tests. Changed `sigvec` and `dotsig`.
-
-* **Version 1.2.52** - Added explicit `inner_hull` and `outer_hull`.
-
-* **Version 1.2.51** - Upped dependency on `medians` to version 2.3.
-
-* **Version 1.2.50** - Upped dependency on `indxvec` to version 1.8. Added error checking to 'contribution' methods in trait `Vecg`.
-
-* **Version 1.2.49** - Added `wradii`. Some more code rationalizations.
-
-* **Version 1.2.48** - Added also weighted `scalar_wfn` and `vector_wfn` to trait `VecVecg`. Also `wdivsmed`.
-
-* **Version 1.2.47** - Added `scalar_fn` and `vector_fn` to trait `VecVec`. These apply arbitrary scalar valued or vector valued closures to all vectors in self. This generality allows some code rationalization.
-
-* **Version 1.2.45** - Completed trait bounds relaxation and simplification. Some minor documentation improvements.
-
-* **Version 1.2.44** - Swapped the sign of `wedge` so it agrees with convention.
-
-* **Version 1.2.43** - Removed `pseudoscalar` method. The `sine` method now computes the correct oriented magnitude of the 2-blade directly from the wedge product. Added geometric product `geometric`. Added some methods to `struct TriangMat` for completeness. In particular, `eigenvalues` and `determinant`, which are both easily obtained after successful Cholesky decomposition.
-
-* **Version 1.2.42** - Added `wedge` (product of Exterior Algebra), `pseudoscalar` and `sine` to trait Vecg. The sine method now always returns the correct anti reflexive sign, in any number of dimensions. The sign flips when the order of the vector operands is exchanged.
-
-* **Version 1.2.41** - Added `anglestat` to `VecVecg` trait. Added convenience function `re_error`. Relaxed trait bounds in `Vecg` trait: `U:Copy -> U:Clone`. Renamed `tukeydot`,`tukeyvec`,`wtukeyvec` to more descriptive `sigdot`,`sigvec`,`wsigvec` and made them include orthogonal points.
-
-* **Version 1.2.40** - Fixed dependencies in `times 1.0.10` as well.

src/lib.rs

@@ -334,7 +334,9 @@ pub trait VecVec<T> {
     /// Inner hull subscripts from their square radii and their sort index.  
     fn inner_hull(self, sqrads: &[f64], sindex: &[usize]) -> Vec<usize>; 
     /// Measure of likelihood of zero median point **p** belonging to zero median data cloud `self`.
-    fn depth(self, descending_index: &[usize], p: &[f64]) -> Result<f64,RE>; 
+    fn depth(self, descending_index: &[usize], p: &[f64]) -> Result<f64,RE>;
+    /// The proportion of points outside of the normal plane through **p** 
+    fn depth_ratio(self, descending_index: &[usize], p: &[f64]) -> f64;
     /// Collects indices of outer and inner hull points, from zero median data    
     fn hulls(self) -> (Vec<usize>, Vec<usize>);
     /// Geometric median's residual error

src/vecvec.rs

@@ -359,20 +359,35 @@ where
     }
 
     /// Likelihood of zero median point **p** belonging to zero median data cloud `self`,
-    /// based on the points outside of normal plane through **p**. 
+    /// based on the points outside of the normal plane through **p**. 
     /// Returns the sum of unit vectors of its outside points, projected onto unit **p**. 
     /// Index should be in the descending order of magnitudes of self points (for efficiency).
     fn depth(self, descending_index: &[usize], p: &[f64]) -> Result<f64,RE> {
-        let p2 = p.vmagsq();
+        let psq = p.vmagsq();
         let mut sumvec = vec![0_f64;p.len()]; 
         for &i in descending_index {
             let s = &self[i];
             let ssq = s.vmagsq();
-            if ssq <= p2 { break; }; // no more outside points
-            if s.dotp(p) > p2 { sumvec.mutvadd(&s.smult(1.0/(ssq.sqrt()))) };
+            if ssq <= psq { break; }; // no more outside points
+            if s.dotp(p) > psq { sumvec.mutvadd(&s.smult(1.0/(ssq.sqrt()))) };
         };
         Ok(sumvec.dotp(&p.vunit()?))
     }
+
+    /// Likelihood of zero median point **p** belonging to zero median data cloud `self`,
+    /// based on the proportion of points outside of the normal plane through **p**. 
+    /// Index should be in the descending order of magnitudes of self points (for efficiency).
+    fn depth_ratio(self, descending_index: &[usize], p: &[f64]) -> f64 {
+        let psq = p.vmagsq();
+        let mut num = 0_f64; 
+        for &i in descending_index {
+            let s = &self[i];
+            let ssq = s.vmagsq();
+            if ssq <= psq { break; }; // no more outside points
+            if s.dotp(p) > psq { num += 1.0; };
+        };
+        num/(self.len() as f64)
+    }
 
     /// Collects indices of inner hull and outer hull, from zero median points in self.
     /// We put a plane trough data point A, normal to its zero median vector **a**.     

tests/tests.rs

@@ -458,13 +458,20 @@ fn hulls() -> Result<(), RE> {
     let sqradii = zeropts.scalar_fn(|p|Ok(p.vmagsq()))?; 
     let mut radindex = sqradii.mergesort_indexed();
     radindex.reverse();
-    println!("Depths of innerhull points: {}",
+    println!("Depths of innerhull points:\n{}",
     innerhull
         .iter()
         .map(|&p| zeropts.depth(&radindex,&zeropts[p]))
         .collect::<Result<Vec<f64>,RE>>()?
         .gr()
     );
+    println!("Depths ratios of innerhull points:\n{}",
+    innerhull
+        .iter()
+        .map(|&p| zeropts.depth_ratio(&radindex,&zeropts[p]))
+        .collect::<Vec<f64>>()
+        .gr()
+    );
 
 
     let sigvec = zeropts.sigvec(&innerhull)?;
@@ -501,15 +508,6 @@ fn hulls() -> Result<(), RE> {
         .collect::<Result<Vec<f64>,RE>>()?
         .gr()
     );
-    /*
-    println!("Depths of all points: {}",
-    (0..pts.len()) 
-        .map(|p| zeropts.depth(&radindex,&zeropts[p]))
-        .collect::<Result<Vec<f64>,RE>>()?
-        .gr()
-    );
-    */
-
     let sigvec = zeropts.sigvec(&outerhull)?;
     println!(
         "Outer hull sigvec: {}",