def rbd(X, dmax, Er=0.01):
"""Decompose a matrix X using reduced basis decomposition. X = YT. Y basis vectors are constructed using mod_gramm_schmidt until an maximum basis number threshold (dmax) or an error requirement (Er) are satisfied."""
#Call start_state from the setup.py module to instantiate the parameters for the RBD
X, Y, T, d, e_cur, i, used_i, complete = start_state(X)
#Set the current_state to the initial parameters; subject to iterative change.
current_state = [X, Y, T, d, e_cur, Er, i, used_i, complete]
#While the requirements for the RBD have not been met...
while d <= dmax and e_cur > Er and complete == False:
#Orthogonally project the X[:,i] vector into Y, calling the mod_gramm_schmidt.py module.
orthonormalization = mod_gramm_schmidt(*current_state)
#mod_gramm_schmidt has a clause that signals complete to equal True.
complete = orthonormalization[-1]
#Break the while loop if mod_gramm_schmidt tells us the decomposition is complete.
if complete == True:
current_state = orthonormalization
break
else:
#Prepare for the next round of orthonormalization by calling the find_error.py and check_error.py modules.
current_state = check_error(*find_error(*orthonormalization))
#Update the active parameters, some of which govern the while loop. check_error() has a clause that signals completion.
X, Y, T, d, e_cur, Er, i, used_i, complete = current_state
#Final output of the function is a list of all the parameters.
return current_state
def test_find_new_vector():
"""Test that find_error returns new e_cur and i values. """
#Setup parameters
test_image = mpimg.imread('images/DrawingHands.jpg')
X, Y, T, d, e_cur, i, used_i, complete = start_state(test_image)
Er = 0.0000000001
#Complete a full iteration of RBD
d1_parameters = check_error(*find_error(
*mod_gramm_schmidt(X, Y, T, d, e_cur, Er, i, used_i, complete)))
#Take a snapshot of e_cur and i before running the find_error function
e_cur_before_find_error = d1_parameters[4]
i_before_find_error = d1_parameters[6]
#Capture the results of find_error
find_error_results = find_error(*mod_gramm_schmidt(*d1_parameters))
e_cur_after_find_error = find_error_results[4]
i_after_find_error = find_error_results[6]
used_i = find_error_results[7]
#Assert that the new e_cur and i values returned by find_error are distinct.
assert e_cur_before_find_error != e_cur_after_find_error
assert i_before_find_error != i_after_find_error
assert i_after_find_error not in used_i
def test_d1_not_done():
"""Test mod_gramm_schmidt where d!=0 and there is a need to continue building the basis."""
#Setup parameters
X, Y, T, d, e_cur, i, used_i, complete = start_state(test_image)
Er = 0.0000000001
#Complete 1 round of the RBD decomposition algorithm
d1_parameters = check_error(*find_error(
*mod_gramm_schmidt(X, Y, T, d, e_cur, Er, i, used_i, complete)))
#Apply gramm_schmidt at the beginning of stage d = 1
gramm_schmidt_results = mod_gramm_schmidt(*d1_parameters)
#Update testing parameters
X, Y, T, d, e_cur, Er, i, used_i, complete = gramm_schmidt_results
#Test datatypes and values
assert type(X) == np.ndarray
assert type(Y) == np.ndarray
assert type(T) == np.ndarray
assert type(d) == int
assert type(Er) == float
assert type(used_i) == list
assert complete == False
#Assert that an orthogonal projection has been added to Y, and that a vector has been added to T. The dimensions of the dot product of Y and T should equal X.
assert Y.shape[1] == T.shape[0] == 2
compression = np.dot(Y, T)
assert compression.shape == X.shape
#Definition of orthogonal matrix
print("Y.T.dot(Y)", Y.T.dot(Y))
assert np.isclose(np.linalg.det(Y.T.dot(Y).astype(float)), 1)
def test_rbd_setup():
'''Test harness for the setup.py module'''
test_image = mpimg.imread('images/DrawingHands.jpg')
dmax = 40
X, Y, T, d, e_cur, i, used_i, complete = start_state(test_image)
#Test datatypes and values
assert type(X) == np.ndarray
assert type(Y) == np.ndarray
assert type(T) == np.ndarray
assert type(d) == int
assert type(e_cur) == float
assert type(i) == int
assert type(used_i) == list
assert complete == False
assert d < dmax
#Test shapes
assert Y.shape[0] == X.shape[0]
assert Y.shape[1] == 1
assert T.shape[0] == 1
assert T.shape[1] == X.shape[1]
assert len(used_i) == 0
assert d < dmax
def test_decomposition_complete():
"""Test case for when e_cur <= Er, signaling that the decomposition is done."""
#Setup parameters
X, Y, T, d, e_cur, i, used_i, complete = start_state(test_image)
Er = np.inf
#Capture results of RBD at the point before check_error is run.
params_before_check_error = find_error(
*mod_gramm_schmidt(X, Y, T, d, e_cur, Er, i, used_i, complete))
#Capture results after running check_error in a case that should return complete == True
params_after_check_error = check_error(*params_before_check_error)
#Test that check_error modifies complete to equal True.
assert params_after_check_error[-1] == True != params_before_check_error[-1]
def test_decomposition_incomplete():
"""Test case for check_error when the error requirement is not met. d increases to 1 for the next iteration of RBD."""
#Setup parameters
X, Y, T, d, e_cur, i, used_i, complete = start_state(test_image)
Er = 0.000000000000001
#Capture results of RBD at the point before check_error is run.
params_before_check_error = find_error(
*mod_gramm_schmidt(X, Y, T, d, e_cur, Er, i, used_i, complete))
#Capture results after running check_error in a case that should return complete == False
params_after_check_error = check_error(*params_before_check_error)
#Test that d is increased by 1 after check_error.
d_before_check_error = params_before_check_error[3]
d_after_check_error = params_after_check_error[3]
assert d_before_check_error < d_after_check_error
assert d_after_check_error - d_before_check_error == 1
def test_d1_done():
"""Test mod_gramm_schmidt where d!=0 and the error requirement is met, ending the RBD algorithm."""
#Setup parameters
X, Y, T, d, e_cur, i, used_i, complete = start_state(test_image)
Er = 0.00001
#Complete 1 round of the RBD decomposition algorithm and update parameters.
d1_parameters = check_error(*find_error(
*mod_gramm_schmidt(X, Y, T, d, e_cur, Er, i, used_i, complete)))
X, Y, T, d, e_cur, Er, i, used_i, complete = d1_parameters
#print("d: ",d)
#print("Y: ", Y)
#Ensure that the algorithm will think the decomposition is complete.
Er = 50000
#Apply gramm_schmidt at the beginning of stage d = 1
gramm_schmidt_results = mod_gramm_schmidt(X, Y, T, d, e_cur, Er, i, used_i,
complete)
assert gramm_schmidt_results[-1] == True
def test_d0():
"""Test the first iteration of mod_gramm_schmidt where d == 0"""
#Setup parameters
X, Y, T, d, e_cur, i, used_i, complete = start_state(test_image)
Er = 0.0000000001
#Apply gramm_schmidt
gramm_schmidt_results = mod_gramm_schmidt(X, Y, T, d, e_cur, Er, i, used_i,
complete)
#Update testing parameters
X, Y, T, d, e_cur, Er, i, used_i, complete = gramm_schmidt_results
#Test datatypes and values
assert type(X) == np.ndarray
assert type(Y) == np.ndarray
assert type(T) == np.ndarray
assert type(d) == int
assert type(Er) == float
assert type(i) == int
assert type(used_i) == list
assert complete == False
#Test that Y and T have non-zero basis
assert np.count_nonzero(Y) > 0
assert np.count_nonzero(T) > 0
#Test that Y and T have the right shapes.
assert Y.shape[1] == 1
assert Y.shape[0] == X.shape[0]
assert T.shape[1] == X.shape[1]
assert T.shape[0] == 1
#Assert that an index has been added to used_i"""
assert len(used_i) == 1
return decomposition
#If the error requirement has not been met, normalize u_d and add it to Y as a column vector. Add the dot product of u_d.transpose and X to T
else:
basis_vector = np.divide(u_d, np.linalg.norm(u_d))
Y = np.column_stack((Y, basis_vector))
T = np.row_stack((T, np.dot(Y[:, d].T, X)))
used_i.append(i)
#Return updated parameters, leading to the find_error process of the algorithm.
update_state = (X, Y, T, d, e_cur, Er, i, used_i, complete)
return update_state
if __name__ == '__main__':
from setup import start_state
alist = np.array([[1, 1, 0, 1, 0], [1, 0, 1, 1, 0], [0, 1, 1, 1, 0]])
X, Y, T, d, e_cur, i, used_i, complete = start_state(alist)
Er = 0.000001
test_results = mod_gramm_schmidt(X, Y, T, d, e_cur, Er, i, used_i,
complete)
X, Y, T, d, e_cur, Er, i, used_i, complete = test_results
def display_state(attributes):
names = ['X', 'Y', 'U', 'd', 'e_cur', 'Er', 'i', 'used_i', 'complete']
for i in range(len(names)):
print("\n\n", names[i], " :\n", attributes[i])
