about summary refs log tree commit diff
diff options
context:
space:
mode:
authorNguyễn Gia Phong <vn.mcsinyx@gmail.com>2018-10-17 22:14:52 +0700
committerNguyễn Gia Phong <vn.mcsinyx@gmail.com>2018-10-17 22:14:52 +0700
commitcec34bee27b184769f81e473e95a49421e9eb0a4 (patch)
tree06e2a7879088d49f960eac03995d7aa594f3fc1a
parentea6dede6ac3766a4d35d763bb0cf3f49f2bf0581 (diff)
downloadcp-cec34bee27b184769f81e473e95a49421e9eb0a4.tar.gz
MATH1.1
-rwxr-xr-xtoys/gauss.py83
1 files changed, 83 insertions, 0 deletions
diff --git a/toys/gauss.py b/toys/gauss.py
new file mode 100755
index 0000000..793be06
--- /dev/null
+++ b/toys/gauss.py
@@ -0,0 +1,83 @@
+#!/usr/bin/env python3
+"""This module provide Gaussian and Gauss-Jordan eliminations on numpy
+array of floats since they are not provided by numpy.
+
+Note that matrix operations are straightforward with numpy.array, namely
+add (+ -), scalar ops (+ - * / **) and mul (@), so try to make use of
+the library while doing your matrix homework ;-P
+"""
+
+from itertools import islice
+from math import inf, isclose
+from numpy import array
+
+
+def non0(row):
+    """Return index of first nonzero entry in row."""
+    for index, entry in enumerate(row):
+        if entry: return index
+    else:
+        return inf
+
+
+def gauss(mat):
+    """Return a matrix in row-echelon form, which is row-equivalent to mat,
+    along with shorthand method of notation of row operations performed.
+    """
+    if mat.size == 0: return mat, []
+    m, matrix, operations = len(mat), mat.__copy__(), []
+    # Fuck optimization: we won't have 1000x1000 matrices for homework anyway.
+    for i in range(m):
+        while i + 1 < m:
+            eye, ar = min(islice(enumerate(matrix), i + 1, m),
+                          default=inf, key=(lambda t: non0(t[1])))
+            if non0(ar) >= non0(matrix[i]): break
+            operations.append('R{} <-> R{}'.format(i + 1, eye + 1))
+            matrix[i], matrix[eye] = ar, matrix[i]
+
+        row = matrix[i]
+        j = non0(row)
+        if j == inf: return matrix, operations
+        if row[j] != 1:     # floating point accuracy is a total disaster tho
+            operations.append('({})R{} -> R{}'.format(1 / row[j], i+1, i+1))
+            row /= row[j]
+
+        for l in range(i + 1, m):
+            k = -matrix[l][j]
+            if k:   # OMG I want Python 3.7 so much
+                operations.append('R{} + {}R{} -> R{}'.format(
+                    l + 1, '' if k == 1 else '({})'.format(k), i + 1, l + 1))
+                matrix[l] += k * row
+    return matrix, operations
+
+
+def gauss_jordan(mat):
+    """Return a matrix in reduced row-echelon form, which is
+    row-equivalent to mat, along with shorthand method of notation of
+    row operations performed.
+    """
+    if mat.size == 0: return mat, []
+    matrix, operations = gauss(mat)
+    m = len(mat)
+    for i in range(1, m):
+        j = non0(matrix[i])
+        if j == inf: return matrix, operations
+        k = -matrix[i - 1][j]
+        if k:
+            operations.append('R{} + {}R{} -> R{}'.format(
+                i, '' if k == 1 else '({})'.format(k), i + 1, i))
+            matrix[i - 1] += k * matrix[i]
+    return matrix, operations
+
+
+if __name__ == '__main__':
+    A = array([[1, -2, 3, 9],
+               [-1, 3, 0, -4],
+               [2, -5, 5, 17]], dtype=float)
+    print(*A, sep='\n', end='\n\n')
+    mat, ops = gauss(A)
+    print(*ops, sep='\n')
+    print(*mat, sep='\n', end='\n\n')
+    mat, ops = gauss_jordan(A)
+    print(*ops, sep='\n')
+    print(*mat, sep='\n', end='\n\n')