jeffrey-hokanson / PSDR / 458

pending completion

Build # 458

Build Type

push

travis-ci-com

Committed by

jeffrey-hokanson

Commit Message

Bumped version to 0.3.10

Run Details

1732 of 2298 branches covered (75.37%)

Branch coverage included in aggregate %.

1 of 1 new or added line in 1 file covered. (100.0%)

7 existing lines in 1 file now uncovered.

5693 of 6502 relevant lines covered (87.56%)

0.88 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

85.94

/psdr/sample/stretch.py

r""" Stretched sampling routines
"""
import numpy as np
from .initial import initial_sample
from .maximin_coffeehouse import maximin_coffeehouse 
from ..geometry import voronoi_vertex_sample, cdist
from cvxpy.error import SolverError
from ..exceptions import EmptyDomainException

def _maximin_block_candidate(domain, Xf, L):
        r"""
        """
        X0 = initial_sample(domain, L, Nsamp = 2*len(Xf))
        #X0 = maximin_coffeehouse(domain, 2*len(Xf), L = L, N0 = 1, full = False, verbose = False)
        #X0 = domain.sample(2*len(Xf))
        X = voronoi_vertex_sample(domain, Xf, X0, L = L, randomize = False) 
        # find the point furthest away 
        D = cdist(Xf, X, L)
        d = np.min(D, axis = 0)
        k = np.argmax(d)
        return X[k], d[k]



def stretch_sample_domain(domain, X, Ls, verbose = False):
        r"""


        Parameters
        ----------
        domain: Domain
                domain of dimension m on which to sample
        X: array-like (M, m)
                Existing samples on the domain
        Ls: list of arrays of size (*, m)
                List of metrics in which to perform dimension reduction        
        """
        # Because we'll be popping items out of this list, we create a copy
        Ls = Ls.copy()
        # A copy of the domain we'll keep adding constraints to        
        domain = domain.copy()

        it = 0        
        while len(Ls) > 0:
                # Compute candidate points and distances
                xd = []
                for k, L in enumerate(Ls):
                        try:
                                X, d = _maximin_block_candidate(domain, X, L)
                        except SolverError:
                                pass
                        xd.append( list([X, d, L, k]))
                #xd = [list(_maximin_block_candidate(domain, X, L)) + [L, k] for k, L in enumerate(Ls)]
                if len(xd) == 0:
                        break
                #for x, d, L, k in xd:
                #        print(f'{k:2d} dist {d:8.2e}')
                
                # Find the largest distance
                x, d, L, k = max(xd, key = lambda l:l[1])
                # Remove this L from the list
                Ls.pop(k)


                # Try to add the new constraint
                domain_new = domain.add_constraints(A_eq = L, b_eq = L @ x)

                try:
                        # ensure we'll be able to sample from the new domain
                        domain_new.sample(1)
                        # If we're a point, we can stop
                        if domain_new.is_point:
                                return domain_new 
                except (SolverError, ValueError, EmptyDomainException):
                        # If either of these two error, return the previous working domain
                        return domain
        
                # Update the domain
                domain = domain_new        
                it += 1
                if verbose:
                        print(f'iter {it:2d}, domain size {len(domain)}, eq constraints {len(domain.A_eq)}')

        return domain

def stretch_sample(domain, Xf, Ls, verbose = False):
        domain_stretch = stretch_sample_domain(domain, Xf, Ls, verbose = verbose)
        #print(domain_stretch.A_eq)
        #print(domain_stretch.b_eq)
        X0 = domain_stretch.sample(100)
        X = voronoi_vertex_sample(domain_stretch, Xf, X0)
        D = cdist(Xf, X)
        d = np.min(D, axis = 0)
        k = np.argmax(d)
        return X[k]
        
 

class StretchSample:
        def __init__(self, fun, X = None, fX = None):
                self.fun = fun
                         

1	r""" Stretched sampling routines
2	"""
3	import numpy as np	1✔
4	from .initial import initial_sample	1✔
5	from .maximin_coffeehouse import maximin_coffeehouse	1✔
6	from ..geometry import voronoi_vertex_sample, cdist	1✔
7	from cvxpy.error import SolverError	1✔
8	from ..exceptions import EmptyDomainException	1✔
9
10	def _maximin_block_candidate(domain, Xf, L):	1✔
11	r"""
12	"""
13	X0 = initial_sample(domain, L, Nsamp = 2*len(Xf))	1✔
14	#X0 = maximin_coffeehouse(domain, 2*len(Xf), L = L, N0 = 1, full = False, verbose = False)
15	#X0 = domain.sample(2*len(Xf))
16	X = voronoi_vertex_sample(domain, Xf, X0, L = L, randomize = False)	1✔
17	# find the point furthest away
18	D = cdist(Xf, X, L)	1✔
19	d = np.min(D, axis = 0)	1✔
20	k = np.argmax(d)	1✔
21	return X[k], d[k]	1✔
22
23
24
25	def stretch_sample_domain(domain, X, Ls, verbose = False):	1✔
26	r"""
27
28
29	Parameters
30	----------
31	domain: Domain
32	domain of dimension m on which to sample
33	X: array-like (M, m)
34	Existing samples on the domain
35	Ls: list of arrays of size (*, m)
36	List of metrics in which to perform dimension reduction
37	"""
38	# Because we'll be popping items out of this list, we create a copy
39	Ls = Ls.copy()	1✔
40	# A copy of the domain we'll keep adding constraints to
41	domain = domain.copy()	1✔
42
43	it = 0	1✔
44	while len(Ls) > 0:	1✔
45	# Compute candidate points and distances
46	xd = []	1✔
47	for k, L in enumerate(Ls):	1✔
48	try:	1✔
49	X, d = _maximin_block_candidate(domain, X, L)	1✔
UNCOV 50	except SolverError:	×
UNCOV 51	pass	×
52	xd.append( list([X, d, L, k]))	1✔
53	#xd = [list(_maximin_block_candidate(domain, X, L)) + [L, k] for k, L in enumerate(Ls)]
54	if len(xd) == 0:	1!
UNCOV 55	break	×
56	#for x, d, L, k in xd:
57	# print(f'{k:2d} dist {d:8.2e}')
58
59	# Find the largest distance
60	x, d, L, k = max(xd, key = lambda l:l[1])	1✔
61	# Remove this L from the list
62	Ls.pop(k)	1✔
63
64
65	# Try to add the new constraint
66	domain_new = domain.add_constraints(A_eq = L, b_eq = L @ x)	1✔
67
68	try:	1✔
69	# ensure we'll be able to sample from the new domain
70	domain_new.sample(1)	1✔
71	# If we're a point, we can stop
72	if domain_new.is_point:	1✔
73	return domain_new	1✔
UNCOV 74	except (SolverError, ValueError, EmptyDomainException):	×
75	# If either of these two error, return the previous working domain
UNCOV 76	return domain	×
77
78	# Update the domain
79	domain = domain_new	1✔
80	it += 1	1✔
81	if verbose:	1!
UNCOV 82	print(f'iter {it:2d}, domain size {len(domain)}, eq constraints {len(domain.A_eq)}')	×
83
84	return domain	1✔
85
86	def stretch_sample(domain, Xf, Ls, verbose = False):	1✔
87	domain_stretch = stretch_sample_domain(domain, Xf, Ls, verbose = verbose)	1✔
88	#print(domain_stretch.A_eq)
89	#print(domain_stretch.b_eq)
90	X0 = domain_stretch.sample(100)	1✔
91	X = voronoi_vertex_sample(domain_stretch, Xf, X0)	1✔
92	D = cdist(Xf, X)	1✔
93	d = np.min(D, axis = 0)	1✔
94	k = np.argmax(d)	1✔
95	return X[k]	1✔
96
97
98
99	class StretchSample:	1✔
100	def __init__(self, fun, X = None, fX = None):	1✔
UNCOV 101	self.fun = fun	×
102

jeffrey-hokanson / PSDR / 458

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous