def find_e2(L, tol=1e-4):
'''
find the eigen vector that corresponds to the smallest non-zeros eigen values of the Laplacian matrix.
Input:
L: the Laplacian matrix, a numpy float matrix of shape n by n.
tol: the tolerance threshold for eigen values, a float scalar. A very small positive value. If an eigen value is smaller than tol, then we consider the eigen value as being 0.
Output:
e2: the eigen vector corresponding to the smallest non-zero eigen value, a numpy float vector of length n.
For example, if the eigen values are [0.5, 0.000000000001, 2.], the first eigen vector is the answer.
Here we assume 0.00000001 is close enough to 0 because is smaller than the tolerance level (tol).
For example, if the eigen values are [2., 0.0000000000001, 0.3], the last eigen vector is the answer.
For example, if the eigen values are [0.00003, 0.000001, 0.3], the last eigen vector is the answer.
'''
#########################################
## INSERT YOUR CODE HERE
Ep = p1.sort_eigen_pairs(p1.compute_eigen_pairs(L))
for row in Ep:
if row[0] >= tol:
e2 = row[1]
break
#########################################
return e2
''' TEST: Now you can test the correctness of your code above by typing `nosetests -v test5.py:test_find_e2' in the terminal. '''
def compute_P(C,k):
'''
Compute the projection matrix P by combining the k eigen vectors, that correspond to the top k largest eigen values of matrix C.
Here the projection matrix P includes all the k principle components.
Input:
C: the covariance matrix, a numpy float matrix of shape p by p.
k: the number of dimensions to reduce to (k should be smaller than p), an integer scalar
Output:
P: the projection matrix, a numpy float matrix of shape p by k.
For example, if we sort the eigen pairs of matrix C in descending order, and
the result [(v1,e1),(v2,e2),...., (vp, ep)], here v1 >= v2 >= v3 ...
The projection matrix should be [e1^T, e2^T, ..., ek^T], here e^T represents the transpose of a row vector e (which is a column vector)
Hint: you can re-use the functions in problem 1 to solve this problem.
'''
#########################################
## INSERT YOUR CODE HERE
eigen = p1.compute_eigen_pairs(C)
eigen_sorted = p1.sort_eigen_pairs(eigen, '')
arr = np.zeros((len(eigen_sorted[0][1]), k))
for i, x in enumerate(eigen_sorted):
if i >= k: break
transpose = x[1][np.newaxis].transpose()
for j in range(len(x[1])):
arr[j][i] = transpose[j]
P = np.array(arr)
#########################################
return P
''' TEST: Now you can test the correctness of your code above by typing `nosetests -v test2.py:test_compute_P' in the terminal. '''
def compute_P(C,k):
'''
Compute the projection matrix P by combining the k eigen vectors, that correspond to the top k largest eigen values of matrix C.
Here the projection matrix P includes all the k principle components.
Input:
C: the covariance matrix, a numpy float matrix of shape p by p.
k: the number of dimensions to reduce to (k should be smaller than p), an integer scalar
Output:
P: the projection matrix, a numpy float matrix of shape p by k.
For example, if we sort the eigen pairs of matrix C in descending order, and
the result [(v1,e1),(v2,e2),...., (vp, ep)], here v1 >= v2 >= v3 ...
The projection matrix should be [e1^T, e2^T, ..., ek^T], here e^T represents the transpose of a row vector e (which is a column vector)
Hint: you can re-use the functions in problem 1 to solve this problem.
'''
#########################################
## INSERT YOUR CODE HERE
Ep = p1.compute_eigen_pairs(C)
eigenp = p1.sort_eigen_pairs(Ep, order='descending')
klargest =[]
pairs = []
for i in range(k):
klargest.append(eigenp[i])
pairs.append(klargest[i][1])
P = np.reshape(pairs,(k,C.shape[0]))
P = P.T
#########################################
return P
''' TEST: Now you can test the correctness of your code above by typing `nosetests -v test2.py:test_compute_P' in the terminal. '''
def find_e2(L, tol=1e-4):
'''
find the eigen vector that corresponds to the smallest non-zeros eigen values of the Laplacian matrix.
Input:
L: the Laplacian matrix, a numpy float matrix of shape n by n.
tol: the tolerance threshold for eigen values, a float scalar. A very small positive value. If an eigen value is smaller than tol, then we consider the eigen value as being 0.
Output:
e2: the eigen vector corresponding to the smallest non-zero eigen value, a numpy float vector of length n.
For example, if the eigen values are [0.5, 0.000000000001, 2.], the first eigen vector is the answer.
Here we assume 0.00000001 is close enough to 0 because is smaller than the tolerance level (tol).
For example, if the eigen values are [2., 0.0000000000001, 0.3], the last eigen vector is the answer.
For example, if the eigen values are [0.00003, 0.000001, 0.3], the last eigen vector is the answer.
Hint: you could solve this problem using 6 lines of code.
'''
#########################################
## INSERT YOUR CODE HERE
pairs = p1.compute_eigen_pairs(L)
vals = np.argsort([i[0] for i in pairs])
e2 = []
for i in vals:
if pairs[i][0] < tol:
continue
else:
e2 = pairs[i][1]
break
#########################################
return e2
''' TEST: Now you can test the correctness of your code above by typing `nosetests -v test5.py:test_find_e2' in the terminal. '''
def compute_P(C, k):
'''
Compute the projection matrix P by combining the k eigen vectors, that correspond to the top k largest eigen values of matrix C.
Here the projection matrix P includes all the k principle components.
Input:
C: the covariance matrix, a numpy float matrix of shape p by p.
k: the number of dimensions to reduce to (k should be smaller than p), an integer scalar
Output:
P: the projection matrix, a numpy float matrix of shape p by k.
For example, if we sort the eigen pairs of matrix C in descending order, and
the result [(v1,e1),(v2,e2),...., (vp, ep)], here v1 >= v2 >= v3 ...
The projection matrix should be [e1^T, e2^T, ..., ek^T], here e^T represents the transpose of a row vector e (which is a column vector)
Hint: you can re-use the functions in problem 1 to solve this problem.
'''
#########################################
## INSERT YOUR CODE HERE
eigenPairs = p1.compute_eigen_pairs(C)
sortedE = p1.sort_eigen_pairs(eigenPairs, False)
P = (np.vstack([x[1] for x in sortedE]).T)[:, :k]
#########################################
return P
''' TEST: Now you can test the correctness of your code above by typing `nosetests s-v test2.py:test_compute_P' in the terminal. '''
def find_e2(L, tol=1e-4):
'''
find the eigen vector that corresponds to the smallest non-zeros eigen values of the Laplacian matrix.
Input:
L: the Laplacian matrix, a numpy float matrix of shape n by n.
tol: the tolerance threshold for eigen values, a float scalar. A very small positive value. If an eigen value is smaller than tol, then we consider the eigen value as being 0.
Output:
e2: the eigen vector corresponding to the smallest non-zero eigen value, a numpy float vector of length n.
For example, if the eigen values are [0.5, 0.000000000001, 2.], the first eigen vector is the answer.
Here we assume 0.00000001 is close enough to 0 because is smaller than the tolerance level (tol).
For example, if the eigen values are [2., 0.0000000000001, 0.3], the last eigen vector is the answer.
For example, if the eigen values are [0.00003, 0.000001, 0.3], the last eigen vector is the answer.
Hint: you could solve this problem using 6 lines of code.
'''
#########################################
## INSERT YOUR CODE HERE
ep = p1.sort_eigen_pairs(p1.compute_eigen_pairs(L))
E = np.asarray([x[1] for x in ep])
v = np.asarray([x[0] for x in ep])
val = 0
while tol < 1:
tol *= 10
val += 1
v = np.around(v, val)
minimum = float('inf')
index = float('nan')
for i in range(len(v)):
if (0 < v[i] < minimum):
minimum = v[i]
index = i
e2 = E[:, index]
#########################################
return e2
''' TEST: Now you can test the correctness of your code above by typing `nosetests -v test5.py:test_find_e2' in the terminal. '''
