addition: gauss-jordan-elimination

src/lib.rs

@@ -1,4 +1,4 @@
-use fractions::Fraction;
+use fractions::{Fraction, FractionError};
 use std::fmt::{Debug, Display};
 use std::ops::Add;
 use std::ops::Mul;
@@ -15,6 +15,15 @@ pub enum MatrixError {
     InvalidSizeForSub,
     InvalidSizeForMul,
     ZeroSize,
+    FailedGauss,
+    FailedGaussJordan,
+    FractionError(FractionError),
+}
+
+impl From<FractionError> for MatrixError {
+    fn from(err: FractionError) -> Self {
+        MatrixError::FractionError(err)
+    }
 }
 
 #[derive(PartialEq, Eq, Debug)]
@@ -148,6 +157,37 @@ impl Matrix {
         None
     }
 
+    fn partial_pivoting(&mut self, col: usize, sign: &mut Fraction) -> Result<bool, MatrixError> {
+        if col >= self.columns {
+            return Err(MatrixError::ColumnOutOfRange);
+        }
+
+        let mut max_row = col;
+        let mut max_value = self.get(col, col).unwrap().abs();
+
+        for r in (col + 1)..self.rows {
+            let val = self.get(r, col).unwrap().abs();
+            if val > max_value {
+                max_value = val;
+                max_row = r;
+            }
+        }
+
+        if max_value.is_zero() {
+            return Ok(false);
+        }
+
+        if max_row != col {
+            match self.exchange_rows(col, max_row) {
+                Some(err) => return Err(err),
+                None => {}
+            };
+            *sign = -*sign;
+        }
+
+        Ok(true)
+    }
+
     pub fn gaussian_elimination(&self) -> Result<(Matrix, Fraction), MatrixError> {
         let mut trig_matrix = Matrix {
             columns: self.columns,
@@ -157,28 +197,13 @@ impl Matrix {
         let mut sign = Fraction::new(1, 1).unwrap();
 
         for i in 0..self.columns {
-            let mut max_row = i;
+            // We do partial pivoting for better efifiency and security
-            let mut max_value = trig_matrix.get(i, i).unwrap().abs();
+            let pivot_exists = trig_matrix.partial_pivoting(i, &mut sign)?;
-
-            // We do parcial pivoting to avoid getting insane
-            // numbers that may result in overflow with fractions
-            for r in (i + 1)..self.rows {
-                let val = trig_matrix.get(r, i).unwrap().abs();
-                if val > max_value {
-                    max_value = val;
-                    max_row = r;
-                }
-            }
 
             // If there ain't no other thing but 0 then we're
             // fucked, determinant is zero
-            if max_value.is_zero() {
+            if !pivot_exists {
-                return Ok((trig_matrix, Fraction::new(0, 1).unwrap()));
+                return Err(MatrixError::FailedGauss);
-            }
-
-            if max_row != i {
-                trig_matrix.exchange_rows(i, max_row);
-                sign = -sign;
             }
 
             let pivot = *trig_matrix.get(i, i).unwrap();
@@ -203,13 +228,65 @@ impl Matrix {
         Ok((trig_matrix, sign))
     }
 
+    pub fn gauss_jordan_elimination(&self) -> Result<Matrix, MatrixError> {
+        let mut new_matrix = Matrix {
+            columns: self.columns,
+            rows: self.rows,
+            data: self.data.clone(),
+        };
+
+        let mut dummy = Fraction::from(1);
+
+        for i in 0..self.columns {
+            let pivot_exists = new_matrix.partial_pivoting(i, &mut dummy)?;
+
+            if !pivot_exists {
+                return Err(MatrixError::FailedGaussJordan);
+            }
+
+            let pivot = *new_matrix.get(i, i).unwrap();
+
+            let new_pivot_row = new_matrix
+                .get_row(i)?
+                .map(|x| *x / pivot)
+                .collect::<Result<Vec<_>, _>>()?;
+
+            new_matrix.set_row(i, new_pivot_row);
+
+            for r in 0..new_matrix.rows {
+                if r == 1 {
+                    continue;
+                }
+
+                let factor = *new_matrix.get(r, i).unwrap();
+
+                if factor.is_zero() {
+                    continue;
+                }
+
+                let new_row_normalized = new_matrix
+                    .get_row(r)?
+                    .zip(new_matrix.get_row(i)?)
+                    .map(|(a, b)| *a - factor * *b)
+                    .collect::<Vec<Fraction>>();
+
+                new_matrix.set_row(r, new_row_normalized);
+            }
+        }
+
+        Ok(new_matrix)
+    }
+
     pub fn get_determinant(&self) -> Result<Fraction, MatrixError> {
         if self.rows != self.columns {
             return Err(MatrixError::NotSquared);
         }
 
-        let (trig_matrix, sign) = self.gaussian_elimination()?;
+        let (trig_matrix, sign) = match self.gaussian_elimination() {
-
+            Err(MatrixError::FailedGauss) => return Ok(Fraction::new(0, 1).unwrap()),
+            Ok((matrix, sign)) => (matrix, sign),
+            Err(err) => return Err(err),
+        };
         // YES, now we got ourselves a triangular matrix, now we just
         // take the product of the diagonal and multiply by sign, that's
         // the determinant :)