def mult_vec_upper_symmetric_by_axpy(self, vec):
"""
This function implements a matrix vector mult based on AXPY using an upper simmetrix matric
:vec: Vector, the vector to work on
:returns: Vector
"""
result = Vector([0]* self.rows())
it = self.full_iterator()
vit = result.vector_iterator()
for i,data in enumerate(it):
#slicing the vector
tl, value, bl = vit.next()
#computing the diagonal value
value = data["a11"] * vec[i] + value
#we perform two different axpy and we make up for the missing part
#by using the equivalent transposed of the column below the diagonal
#aka the row after the diagonal
tl = data["a01"].axpy(tl,vec[i])
bl = data["at12"].axpy(bl,vec[i])
#merging the result back in
vit.merge(tl,value,bl)
return result
def mult_vec_lower_symmetric_by_axpy(self, vec):
"""
This function implements a matrix vector mult based on AXPY using an upper simmetrix matric
:vec: Vector, the vector to work on
:returns: Vector
"""
result = Vector([0]* self.rows())
it = self.full_iterator()
vit = result.vector_iterator()
for i,data in enumerate(it):
#slicing the vector
tl, value, bl = vit.next()
#computing the diagonal value
value = data["a11"] * vec[i] + value
tl = data["at10"].axpy(tl,vec[i])
bl = data["a21"].axpy(bl,vec[i])
#merging the result back in
vit.merge(tl,value,bl)
return result
def mult_vec_lower_triangular_by_axpy(self, vec):
"""
This function implements a lower triangular matrix vector mult based on AXPY
:vec: Vector, the vector to work on
:returns: Vector
"""
result = Vector([0]* self.rows())
it = self.full_iterator()
vit = result.vector_iterator()
for i,data in enumerate(it):
#slicing the vector
tl, value, bl = vit.next()
#computing the diagonal value
value = data["a11"] * vec[i] + value
#computing the upper diagonal value, using an axpy
#and accumulating in the up vector
bl = data["a21"].axpy(bl,vec[i])
#merging the result back in
vit.merge(tl,value,bl)
return result
def mult_vec_upper_symmetric_by_axpy(self, vec):
"""
This function implements a matrix vector mult based on AXPY using an upper simmetrix matric
:vec: Vector, the vector to work on
:returns: Vector
"""
result = Vector([0] * self.rows())
it = self.full_iterator()
vit = result.vector_iterator()
for i, data in enumerate(it):
#slicing the vector
tl, value, bl = vit.next()
#computing the diagonal value
value = data["a11"] * vec[i] + value
#we perform two different axpy and we make up for the missing part
#by using the equivalent transposed of the column below the diagonal
#aka the row after the diagonal
tl = data["a01"].axpy(tl, vec[i])
bl = data["at12"].axpy(bl, vec[i])
#merging the result back in
vit.merge(tl, value, bl)
return result
def mult_vec_lower_symmetric_by_axpy(self, vec):
"""
This function implements a matrix vector mult based on AXPY using an upper simmetrix matric
:vec: Vector, the vector to work on
:returns: Vector
"""
result = Vector([0] * self.rows())
it = self.full_iterator()
vit = result.vector_iterator()
for i, data in enumerate(it):
#slicing the vector
tl, value, bl = vit.next()
#computing the diagonal value
value = data["a11"] * vec[i] + value
tl = data["at10"].axpy(tl, vec[i])
bl = data["a21"].axpy(bl, vec[i])
#merging the result back in
vit.merge(tl, value, bl)
return result
def mult_vec_lower_triangular_by_axpy(self, vec):
"""
This function implements a lower triangular matrix vector mult based on AXPY
:vec: Vector, the vector to work on
:returns: Vector
"""
result = Vector([0] * self.rows())
it = self.full_iterator()
vit = result.vector_iterator()
for i, data in enumerate(it):
#slicing the vector
tl, value, bl = vit.next()
#computing the diagonal value
value = data["a11"] * vec[i] + value
#computing the upper diagonal value, using an axpy
#and accumulating in the up vector
bl = data["a21"].axpy(bl, vec[i])
#merging the result back in
vit.merge(tl, value, bl)
return result
def next(self):
"""
This is the function used to step the iterator forward
:returns: Vector, float, Vector, which are in order the column vector above
the diagonal, the diagonal value, and the vector below the diagonal
"""
if self.counter < self.columns:
#if we did not extinguish the iterator we check if the counter
#is 0, meaning the first iteration.
#if this is the case the TL (top left part) of the matrix
#will be an empty vector
#todo: probably the check is useless the slice operation would retunr an empty
#list anyway
if self.counter == 0:
TL = Vector([])
else:
TL = Vector(self.data[self.counter][:self.counter])
#we extrapolate out the diagonal value , here is where we lose the reference
#to the original value
value = self.data[self.counter][self.counter]
#the BL (bottom left part) is nothing more then the remaining column vector
#below the the diagonal
BL = Vector(self.data[self.counter][self.counter+1:])
self.counter +=1
return [TL, value, BL]
else:
raise StopIteration()
def test_solve_upper_triangular_system(self):
mat = SquareMatrix.from_list(3, [-2, 0, 0, -1, -3, 0, 1, -2, 1])
vec = Vector([6, 9, 3])
it = vec.vector_iterator(True)
res = mat.solve_upper_triangular_system(vec)
self.is_vector_equal_to_list(res, [1, -5, 3])
def test_creation_from_vectors(self):
vec1 = Vector(self.lsF2[:4])
vec2 = Vector.copy(vec1)
vec3 = Vector.copy(vec1)
vec4 = Vector.copy(vec1)
mat = matrix.Matrix.from_vectors([vec1, vec2, vec3, vec4])
self.is_matrix_equal(self.matF2, mat)
def test_creation_from_vectors(self):
vec1 = Vector(self.lsF2[:4])
vec2 = Vector.copy(vec1)
vec3 = Vector.copy(vec1)
vec4 = Vector.copy(vec1)
mat = matrix.Matrix.from_vectors([vec1,vec2,vec3,vec4])
self.is_matrix_equal(self.matF2, mat)
def test_mult_vec_by_axpy_parted(self):
res = self.matF1.mult_vec_by_axpy_parted(Vector([1, 2, 3, 4]))
self.is_vector_equal_to_list(res, [0, 0, 0, 0])
res = self.matF2.mult_vec_by_axpy_parted(Vector([1, 2, 3, 4]))
self.is_vector_equal_to_list(res, [10, 20, 30, 40])
res = self.matF3.mult_vec_by_axpy_parted(Vector([5, 4, 3]))
self.is_vector_equal_to_list(res, [41, 16, 38])
def test_mult_vec_lower_symmetric_by_axpy(self):
vec = Vector([3, 5, 3, 4])
res = self.matF11.mult_vec_lower_symmetric_by_axpy(vec)
self.is_vector_equal_to_list(res, [38, 41, 49, 60])
self.matF11.set_upper_simmetry()
vec = Vector([3, 5, 3, 4])
res = self.matF11.mult_vec_lower_symmetric_by_axpy(vec)
self.is_vector_equal_to_list(res, [38, 41, 49, 60])
def test_solve_upper_triangular_system(self):
mat = SquareMatrix.from_list(3, [-2,0,0,-1,-3,0,1,-2,1])
vec = Vector([6,9,3])
it = vec.vector_iterator(True)
res = mat.solve_upper_triangular_system(vec)
self.is_vector_equal_to_list(res,[1,-5,3])
def test_mult_by_transposed_matrix_axpy(self):
res = self.matF1.mult_vec_by_transposed_matrix_axpy(
Vector([1, 2, 3, 4]))
self.is_vector_equal_to_list(res, [0, 0, 0, 0])
res = self.matF2.mult_vec_by_transposed_matrix_axpy(
Vector([1, 2, 3, 4]))
self.is_vector_equal_to_list(res, [30, 30, 30, 30])
res = self.matF3.mult_vec_by_transposed_matrix_axpy(Vector([5, 4, 3]))
self.is_vector_equal_to_list(res, [31, 42, 20])
def test_mult_vec_by_axpy(self):
res = self.matF1.mult_vec_by_dot(Vector([1, 2, 3, 4]))
self.is_vector_equal_to_list(res, [0, 0, 0, 0])
res = self.matF2.mult_vec_by_dot(Vector([1, 2, 3, 4]))
self.is_vector_equal_to_list(res, [10, 20, 30, 40])
#mat = self.matF2.mult_matrix_update_by_column(self.matF9)
#self.is_matrix_equal_to_matrix(mat, self.matF2)
res = self.matF3.mult_vec_by_dot(Vector([5, 4, 3]))
self.is_vector_equal_to_list(res, [41, 16, 38])
def test_creation(self):
mat = SquareMatrix(4, None, True)
self.is_matrix_equal(self.matF8, mat)
mat = SquareMatrix(4, [
Vector([1, 0, 0, 0]),
Vector([0, 1, 0, 0]),
Vector([0, 0, 1, 0]),
Vector([0, 0, 0, 1])
], False)
self.is_matrix_equal(self.matF8, mat)
def next(self):
if (self.end_condition()):
#top matrix
#column above diagonal
self.a01 = Vector(self.data[self.counter][:self.counter])
#diagonal value
self.a11= self.data[self.counter][self.counter]
#columb below diagonal
self.a21= Vector(self.data[self.counter][self.counter+1:])
#row before diagonal
self.at10= Vector(Vector([self.data[x][self.counter] for x in range(self.counter)]))
#row after diagonal
self.at12 = Vector([self.data[x][self.counter] for x in range(self.counter+1,self.columns)])
##matirces
if self.counter == 0:
self.A00 = matrix.Matrix(0,0)
self.A20 = matrix.Matrix(0,0)
else:
self.A00 = matrix.Matrix.from_vectors([Vector(self.data[c][:self.counter] ) for c in range(self.counter)])
#here we check first if there is no zero column or row in the matrix we would create, if so we create an empty
#matrix
if (self.counter+1 < (self.rows) and self.counter <(self.columns)):
self.A20 = matrix.Matrix.from_vectors([Vector(self.data[c][self.counter+1:] ) for c in range(self.counter)])
else:
self.A20 = matrix.Matrix(0,0)
#here we need to check if we are not going out of bound
if (self.counter+1 <self.columns) and self.counter > 0 :
self.A02 = matrix.Matrix.from_vectors([Vector(self.data[c][:self.counter] ) for c in range(self.counter+1,self.columns)])
else:
self.A02= matrix.Matrix(0,0)
#here we check first if there is no zero column or row in the matrix we would create, if so we create an empty
#matrix
if (self.counter+1 < (self.rows) and self.counter+1 <(self.columns)):
self.A22 = matrix.Matrix.from_vectors([Vector(self.data[c][self.counter+1:] ) for c in range(self.counter+1,self.columns)])
else:
self.A22 = matrix.Matrix(0,0)
self.counter += self.counter_speed
to_return = {"A00": self.A00,
"a01": self.a01,
"A02": self.A02,
"at10":self.at10,
"a11": self.a11,
"at12":self.at12,
"A20": self.A20,
"a21": self.a21,
"A22": self.A22}
return to_return
else:
raise StopIteration()
def test_solve_lower_triangular_system(self):
mat = SquareMatrix.from_list(3, [2, 4, 6, 4, -2, -4, -2, 6, 2])
L, U = mat.lu_factorization()
vec = Vector([-10, 20, 18])
res = L.solve_lower_triangular_system(vec)
self.is_vector_equal_to_list(res, [-10, 40, -16])
def test_gaussian_appended_reduction_upper_triangular(self):
mat = SquareMatrix.from_list(3, [2, 4, 6, 4, -2, -4, -2, 6, 2])
res = mat.gaussian_reduction_upper_triangular(True)
vec = Vector([-10, 20, 18])
res = res.gaussian_appended_reduction_upper_triangular(vec)
self.is_vector_equal_to_list(res, [-10, 40, -16])
def test_solve_linear_system(self):
mat = SquareMatrix.from_list(
4, [2, -2, 4, -4, 0, -1, -1, 1, 1, 1, 5, -3, 2, -1, 4, -8])
vec = Vector([2, 2, 11, -3])
res = mat.solve_linear_system(vec)
self.is_vector_equal_to_list(res, [1, -1, 2, -1])
def next(self):
"""
Function to setp forward the iterator
"""
#first we check if finised the iteration
if self.counter < self.rows:
#we slice the data and upgrade the counter
to_return = Vector([self.data[c][self.counter] for c in xrange(self.columns)])
self.counter +=1
return to_return
else:
#if we are done we rise a stop iteration exception
raise StopIteration()
def from_list(cls, size=4, data=None):
"""
Alternative constuctor used to build a square matrix from a list of elements
Args:
:size: integer, the size of the NxN matrix
:data: float[], the data used for thea matrix columns
"""
vectors = []
for c in range(size):
vec = Vector([data[c * size + r] for r in range(size)])
vectors.append(vec)
inst = cls(size, vectors, False)
return inst
def test_gaussian_reduction_upper_triangular(self):
mat = SquareMatrix.from_list(3, [2, 4, 6, 4, -2, -4, -2, 6, 2])
res = mat.gaussian_reduction_upper_triangular()
self.is_matrix_equal_to_list(res, [2, 0, 0, 4, -10, 0, -2, 10, -8])
res = mat.gaussian_reduction_upper_triangular(True)
self.is_matrix_equal_to_list(res, [2, 2, 3, 4, -10, 1.6, -2, 10, -8])
vec = Vector([-10, 20, 18])
res = mat.gaussian_reduction_upper_triangular(True, vec)
self.is_matrix_equal_to_list(res, [2, 2, 3, 4, -10, 1.6, -2, 10, -8])
self.is_vector_equal_to_list(vec, [-10, 40, -16])
mat = SquareMatrix.from_list(3, [1, -3, 1, 1, 0, 13, 2, -3, 20])
res = mat.gaussian_reduction_upper_triangular()
self.is_matrix_equal_to_list(res, [1, 0, 0, 1, 3, 0, 2, 3, 6])
def mult_vec_lower_symmetric_by_dot(self, vec):
"""
This funtion transforms a vector using a lower simmetric matrix, means
we don't neet to use the upper traingular matrix which might old other
data, we use dot product under the hood to perform the computation
:vec: Vector, the vector to be transformed
"""
result = Vector([0] * self.rows())
it = self.full_iterator()
vit = vec.vector_iterator()
for i, data in enumerate(it):
#we slice the vector calling manually the iterator next
tl, value, bl = vit.next()
#here instead of the upper column before the diagonal we use the
#row before the diagonal
result[i] = (data["at10"].dot(tl) + data["a11"] * value +
data["a21"].dot(bl))
return result
def mult_vec_upper_symmetric_by_dot(self, vec):
"""
This function multiplies the given vector by an upper simmetric matrix, this will
allow to perform less computation by exploiting the simmetry
:vec: Vector, the vector to be transformed
:returns: Vector
"""
result = Vector([0] * self.rows())
it = self.full_iterator()
vit = vec.vector_iterator()
for i, data in enumerate(it):
#we slice the vector calling manually the iterator next
tl, value, bl = vit.next()
#to nothe here instead the first part of the row (before the diagonal
#we use the column before the diagona, exploiting the simmetry
result[i] = (data["a01"].dot(tl) + data["a11"] * value +
data["at12"].dot(bl))
return result
def solve_lower_triangular_system(self, b):
"""
This function is going to solve for
L*z = b
where we solve for z, b is given and L is a lower unit triangular matrix
(most likely the result of a LU operation)
:b: Vector, the vector needed to solve the system
:returns: Vector
"""
res = Vector.copy(b)
bit = res.vector_iterator()
it = self.full_iterator()
for TL, value, BL in bit:
data = it.next()
BL = BL + (-(value * data["a21"]))
bit.merge(TL, value, BL)
return res
def solve_lower_triangular_system(self, b):
"""
This function is going to solve for
L*z = b
where we solve for z, b is given and L is a lower unit triangular matrix
(most likely the result of a LU operation)
:b: Vector, the vector needed to solve the system
:returns: Vector
"""
res = Vector.copy(b)
bit = res.vector_iterator()
it = self.full_iterator()
for TL,value,BL in bit:
data = it.next()
BL = BL + (-(value * data["a21"]))
bit.merge(TL,value,BL)
return res
def mult_vec_lower_triangular_by_dot(self, vec):
"""
This function is multiplying a vector by an upper triangular matrix
using a parted matrix and the dot product approach, the optimmization comes
from the fact we don't need to perform the computation of the row after the
current diagonal index
:vec: Vector, the value to multiply
:returns: Vector
"""
result = Vector([0] * self.rows())
it = self.full_iterator()
vit = vec.vector_iterator()
for i, data in enumerate(it):
#we slice the vector calling manually the iterator next
tl, value, bl = vit.next()
#we accumualte the result in the proper index of the result vector
#the result is composed of the result of diagonal value being multiplied
#plus the row before the diagonal
result[i] = (data["a11"] * value + tl.dot(data["at10"]))
return result
def solve_upper_triangular_system(self,b):
"""
This function is the complementary one for solving a liner equation system
based on LU,
with solve_lower_triangular_system we solved L* z = b
where now we can solve U* x = b where z = U*x and U is the upper triangular
matrix from the factoriazition, basically we are performing a backward sostiuition
:b: Vector, the vector needed to solve the system
:returns: Vector
"""
res = Vector.copy(b)
bit = res.vector_iterator(True)
it = self.full_iterator(True)
for TL,value,BL in bit:
data = it.next()
value -= data["at12"].dot(BL)
value /= data["a11"]
bit.merge(TL,value,BL)
return res
def solve_upper_triangular_system(self, b):
"""
This function is the complementary one for solving a liner equation system
based on LU,
with solve_lower_triangular_system we solved L* z = b
where now we can solve U* x = b where z = U*x and U is the upper triangular
matrix from the factoriazition, basically we are performing a backward sostiuition
:b: Vector, the vector needed to solve the system
:returns: Vector
"""
res = Vector.copy(b)
bit = res.vector_iterator(True)
it = self.full_iterator(True)
for TL, value, BL in bit:
data = it.next()
value -= data["at12"].dot(BL)
value /= data["a11"]
bit.merge(TL, value, BL)
return res
def copy(cls, mat):
"""
Alternative constructor to generate a matrix identical to the given one
"""
return cls(mat.columns(),
[Vector.copy(v) for v in mat.column_iterator()], False)
def test_mult_vec_upper_triangular_by_axpy(self):
vec = Vector([1, 2, 3, 4])
res = self.matF9.mult_vec_upper_triangular_by_axpy(vec)
self.is_vector_equal_to_list(res, [10, 18, 21, 16])
def test_mult_vec_lower_triangular_by_axpy(self):
vec = Vector([1, 2, 3, 4])
res = self.matF11.mult_vec_lower_triangular_by_axpy(vec)
self.is_vector_equal_to_list(res, [1, 6, 18, 40])
def merge(self, TL, value, BL):
self.data[self.counter -1] = Vector(TL.values + [value] + BL.values)
def copy(cls, mat):
"""
Alternative constructor to generate a matrix identical to the given one
"""
return cls(mat.columns(),[Vector.copy(v) for v in mat.column_iterator()],False)
