def get_patches(inva, wx, ibs, verbose=True):
"""
Loads the patches assigned to a particular word in this stack
>>> inva.wx_to_aids = inva.compute_inverted_list()
>>> verbose=True
"""
config = inva.config
aid_list = inva.wx_to_aids[wx]
X_list = [inva.get_annot(aid) for aid in aid_list]
fxs_groups = [X.fxs(wx) for X in X_list]
all_kpts_list = ibs.depc.d.get_feat_kpts(aid_list, config=config)
sub_kpts_list = vt.ziptake(all_kpts_list, fxs_groups, axis=0)
total_patches = sum(ut.lmap(len, fxs_groups))
chip_list = ibs.depc_annot.d.get_chips_img(aid_list, config=config)
# convert to approprate colorspace
#if colorspace is not None:
# chip_list = vt.convert_image_list_colorspace(chip_list, colorspace)
# ut.print_object_size(chip_list, 'chip_list')
patch_size = 64
shape = (total_patches, patch_size, patch_size, 3)
_prog = ut.ProgPartial(enabled=verbose, lbl='warping patches', bs=True)
_patchiter = ut.iflatten([
vt.get_warped_patches(chip, kpts, patch_size=patch_size)[0]
#vt.get_warped_patches(chip, kpts, patch_size=patch_size, use_cpp=True)[0]
for chip, kpts in _prog(zip(chip_list, sub_kpts_list),
length=len(aid_list))
])
word_patches = vt.fromiter_nd(_patchiter, shape, dtype=np.uint8)
return word_patches
def ensure_names_are_connected(graph, aids_list):
aug_graph = graph.copy().to_undirected()
orig_edges = aug_graph.edges()
unflat_edges = [list(itertools.product(aids, aids)) for aids in aids_list]
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
new_edges = ut.setdiff_ordered(aid_pairs, aug_graph.edges())
preweighted_edges = nx.get_edge_attributes(aug_graph, 'weight')
if preweighted_edges:
orig_edges = ut.setdiff(orig_edges, list(preweighted_edges.keys()))
aug_graph.add_edges_from(new_edges)
# Ensure the largest possible set of original edges is in the MST
nx.set_edge_attributes(aug_graph,
name='weight',
values=dict([(edge, 1.0) for edge in new_edges]))
nx.set_edge_attributes(aug_graph,
name='weight',
values=dict([(edge, 0.1) for edge in orig_edges]))
for cc_sub_graph in nx.connected_component_subgraphs(aug_graph):
mst_sub_graph = nx.minimum_spanning_tree(cc_sub_graph)
for edge in mst_sub_graph.edges():
redge = edge[::-1]
if not (graph.has_edge(*edge) or graph.has_edge(*redge)):
graph.add_edge(*redge, attr_dict={})
def get_name_rowid_edges_from_nids(ibs, nids):
aids_list = ibs.get_name_aids(nids)
import itertools
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
return aids1, aids2
def get_name_rowid_edges_from_nids(ibs, nids):
aids_list = ibs.get_name_aids(nids)
import itertools
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
return aids1, aids2
def get_name_rowid_edges_from_aids2(ibs, aids_list):
# grouped version
import itertools
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
#if full:
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
return aids1, aids2
def compute_word_weights(inva, method='idf'):
"""
Compute a per-word weight like idf
Example:
>>> # xdoctest: +REQUIRES(--slow)
>>> # ENABLE_DOCTEST
>>> from wbia.algo.smk.inverted_index import * # NOQA
>>> qreq_, inva = testdata_inva()
>>> wx_to_weight = inva.compute_word_weights()
>>> print('wx_to_weight = %r' % (wx_to_weight,))
"""
wx_list = sorted(inva.wx_to_aids.keys())
with ut.Timer('Computing %s weights' % (method, )):
if method == 'idf':
ndocs_total = len(inva.aids)
# Unweighted documents
ndocs_per_word = np.array(
[len(set(inva.wx_to_aids[wx])) for wx in wx_list])
weight_per_word = smk_funcs.inv_doc_freq(
ndocs_total, ndocs_per_word)
elif method == 'idf-maw':
# idf denom (the num of docs containing a word for each word)
# The max(maws) denote the prob that this word indexes an annot
ndocs_total = len(inva.aids)
# Weighted documents
wx_to_ndocs = {wx: 0.0 for wx in wx_list}
for wx, maws in zip(ut.iflatten(inva.wx_lists),
ut.iflatten(inva.maws_lists)):
wx_to_ndocs[wx] += min(1.0, max(maws))
ndocs_per_word = ut.take(wx_to_ndocs, wx_list)
weight_per_word = smk_funcs.inv_doc_freq(
ndocs_total, ndocs_per_word)
elif method == 'uniform':
weight_per_word = np.ones(len(wx_list))
wx_to_weight = dict(zip(wx_list, weight_per_word))
wx_to_weight = ut.DefaultValueDict(0, wx_to_weight)
return wx_to_weight
def translate_all():
""" Translates a all python paths in directory """
dpaths = utool.ls_moduledirs('.')
#print('[cyth] translate_all: %r' % (dpaths,))
globkw = {'recursive': True, 'with_dirs': False, 'with_files': True}
# Find all unique python files in directory
fpaths_iter = [
utool.glob(utool.unixpath(dpath), '*.py', **globkw) for dpath in dpaths
]
fpath_iter = utool.iflatten(fpaths_iter)
abspath_iter = map(utool.unixpath, fpath_iter)
fpath_list = list(set(list(abspath_iter)))
#print('[cyth] translate_all: %s' % ('\n'.join(fpath_list),))
# Try to translate each
translate(*fpath_list)
def translate_all():
""" Translates a all python paths in directory """
dpaths = utool.ls_moduledirs('.')
#print('[cyth] translate_all: %r' % (dpaths,))
globkw = {
'recursive': True,
'with_dirs': False,
'with_files': True
}
# Find all unique python files in directory
fpaths_iter = [utool.glob(utool.unixpath(dpath), '*.py', **globkw)
for dpath in dpaths]
fpath_iter = utool.iflatten(fpaths_iter)
abspath_iter = map(utool.unixpath, fpath_iter)
fpath_list = list(set(list(abspath_iter)))
#print('[cyth] translate_all: %s' % ('\n'.join(fpath_list),))
# Try to translate each
translate(*fpath_list)
def ensure_names_are_connected(graph, aids_list):
aug_graph = graph.copy().to_undirected()
orig_edges = aug_graph.edges()
unflat_edges = [list(itertools.product(aids, aids)) for aids in aids_list]
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
new_edges = ut.setdiff_ordered(aid_pairs, aug_graph.edges())
preweighted_edges = nx.get_edge_attributes(aug_graph, 'weight')
if preweighted_edges:
orig_edges = ut.setdiff(orig_edges, list(preweighted_edges.keys()))
aug_graph.add_edges_from(new_edges)
# Ensure the largest possible set of original edges is in the MST
nx.set_edge_attributes(aug_graph, 'weight', dict([(edge, 1.0) for edge in new_edges]))
nx.set_edge_attributes(aug_graph, 'weight', dict([(edge, 0.1) for edge in orig_edges]))
for cc_sub_graph in nx.connected_component_subgraphs(aug_graph):
mst_sub_graph = nx.minimum_spanning_tree(cc_sub_graph)
for edge in mst_sub_graph.edges():
redge = edge[::-1]
if not (graph.has_edge(*edge) or graph.has_edge(*redge)):
graph.add_edge(*redge, attr_dict={})
def setup_pzmtest_subgraph():
import ibeis
ibs = ibeis.opendb(db='PZ_MTEST')
nids = ibs.get_valid_nids()
aids_list = ibs.get_name_aids(nids)
import itertools
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
rng = np.random.RandomState(0)
flags = rng.rand(len(aids1)) > .878
aids1 = ut.compress(aids1, flags)
aids2 = ut.compress(aids2, flags)
for aid1, aid2 in zip(aids1, aids2):
ibs.set_annot_pair_as_positive_match(aid1, aid2)
ibs.set_annot_pair_as_positive_match(aid2, aid1)
rowids = ibs._get_all_annotmatch_rowids()
aids1 = ibs.get_annotmatch_aid1(rowids)
aids2 = ibs.get_annotmatch_aid2(rowids)
def setup_pzmtest_subgraph():
import ibeis
ibs = ibeis.opendb(db='PZ_MTEST')
nids = ibs.get_valid_nids()
aids_list = ibs.get_name_aids(nids)
import itertools
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
rng = np.random.RandomState(0)
flags = rng.rand(len(aids1)) > .878
aids1 = ut.compress(aids1, flags)
aids2 = ut.compress(aids2, flags)
for aid1, aid2 in zip(aids1, aids2):
ibs.set_annot_pair_as_positive_match(aid1, aid2)
ibs.set_annot_pair_as_positive_match(aid2, aid1)
rowids = ibs._get_all_annotmatch_rowids()
aids1 = ibs.get_annotmatch_aid1(rowids)
aids2 = ibs.get_annotmatch_aid2(rowids)
def build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list):
""" helper
these list comprehensions replace the prevous for loop
they still need to be optimized a little bit (and made clearer)
can probably unnest the list comprehensions as well
"""
"""
IGNORE
Legacy::
def old_build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list):
fm_nestlist_ = []
fs_nestlist_ = []
daid_nestlist_ = []
for scores, qfxs, dfxs, daids in zip(sparse_list, qfxs_list, dfxs_list, daids_list):
for rx, cx, score in zip(scores.row, scores.col, scores.data):
_fm = tuple(product(qfxs[rx], dfxs[cx]))
_fs = [score / len(_fm)] * len(_fm)
_daid = [daids[cx]] * len(_fm)
fm_nestlist_.append(_fm)
fs_nestlist_.append(_fs)
daid_nestlist_.append(_daid)
return fm_nestlist_, fs_nestlist_, daid_nestlist_
oldtup_ = old_build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list)
fm_nestlist_, fs_nestlist_, daid_nestlist_ = oldtup_
newtup_ = build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list)
fm_nestlist, fs_nestlist, daid_nestlist = newtup_
assert fm_nestlist == fm_nestlist_
assert fs_nestlist == fs_nestlist_
assert daid_nestlist == daid_nestlist_
47ms
%timeit build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list)
59ms
%timeit old_build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list)
IGNORE
"""
# FIXME: rewrite double comprehension as a flat comprehension
# Build nested feature matches (a single match might have many members)
fm_nestlist = [
tuple(product(qfxs[rx], dfxs[cx]))
for scores, qfxs, dfxs in zip(sparse_list, qfxs_list, dfxs_list)
for rx, cx in zip(scores.row, scores.col)
]
nFm_list = [len(fm) for fm in fm_nestlist]
#fs_unsplit = (score
# for scores in sparse_list
# for score in scores.data)
#daid_unsplit = (daids[cx]
# for scores, daids in zip(sparse_list, daids_list)
# for cx in scores.col)
# Build nested feature scores
fs_unsplit = utool.iflatten((scores.data for scores in sparse_list))
# Build nested feature matches (a single match might have many members)
daid_unsplit = utool.iflatten(
(daids.take(scores.col)
for scores, daids in zip(sparse_list, daids_list)))
# Expand feature scores and daids splitting scores amongst match members
fs_nestlist = [[score / nFm] * nFm
for score, nFm in zip(fs_unsplit, nFm_list)]
daid_nestlist = [[daid] * nFm for daid, nFm in zip(daid_unsplit, nFm_list)]
if DEBUG_SMK:
assert len(fm_nestlist) == len(fs_nestlist), 'inconsistent len'
assert len(fm_nestlist) == len(nFm_list), 'inconsistent len'
assert len(daid_nestlist) == len(fs_nestlist), 'inconsistent len'
min_ = min(2, len(nFm_list))
max_ = min(15, len(nFm_list))
print('nFm_list[_min:_max] = ' +
utool.list_str(nFm_list[min_:max_]))
print('fm_nestlist[_min:_max] = ' +
utool.list_str(fm_nestlist[min_:max_]))
print('fs_nestlist[_min:_max] = ' +
utool.list_str(fs_nestlist[min_:max_]))
print('daid_nestlist[_min:_max] = ' +
utool.list_str(daid_nestlist[min_:max_]))
for fm_, fs_, daid_ in zip(fm_nestlist, fs_nestlist, daid_nestlist):
assert len(fm_) == len(fs_), 'inconsistent len'
assert len(fm_) == len(daid_), 'inconsistent len'
print('[smk_core] checked build_chipmatch correspondence ...ok')
return fm_nestlist, fs_nestlist, daid_nestlist
def invertible_stack(vecs_list, label_list):
"""
Stacks descriptors into a flat structure and returns inverse mapping from
flat database descriptor indexes (dx) to annotation ids (label) and feature
indexes (fx). Feature indexes are w.r.t. annotation indexes.
Output:
idx2_desc - flat descriptor stack
idx2_label - inverted index into annotations
idx2_fx - inverted index into features
# Example with 2D Descriptors
Example:
>>> # DISABLE_DOCTEST
>>> from vtool.nearest_neighbors import * # NOQA
>>> DESC_TYPE = np.uint8
>>> label_list = [1, 2, 3, 4, 5]
>>> vecs_list = [
... np.array([[0, 0], [0, 1]], dtype=DESC_TYPE),
... np.array([[5, 3], [2, 30], [1, 1]], dtype=DESC_TYPE),
... np.empty((0, 2), dtype=DESC_TYPE),
... np.array([[5, 3], [2, 30], [1, 1]], dtype=DESC_TYPE),
... np.array([[3, 3], [42, 42], [2, 6]], dtype=DESC_TYPE),
... ]
>>> idx2_vec, idx2_label, idx2_fx = invertible_stack(vecs_list, label_list)
>>> print(repr(idx2_vec.T))
array([[ 0, 0, 5, 2, 1, 5, 2, 1, 3, 42, 2],
[ 0, 1, 3, 30, 1, 3, 30, 1, 3, 42, 6]], dtype=uint8)
>>> print(repr(idx2_label))
array([1, 1, 2, 2, 2, 4, 4, 4, 5, 5, 5])
>>> print(repr(idx2_fx))
array([0, 1, 0, 1, 2, 0, 1, 2, 0, 1, 2])
"""
# INFER DTYPE? dtype = vecs_list[0].dtype
# Build inverted index of (label, fx) pairs
nFeats = sum(list(map(len, vecs_list)))
nFeat_iter = map(len, vecs_list)
label_nFeat_iter = zip(label_list, map(len, vecs_list))
# generate featx inverted index for each feature in each annotation
_ax2_fx = [list(range(nFeat)) for nFeat in nFeat_iter]
# generate label inverted index for each feature in each annotation
'''
# this is not a real test the code just happened to be here. syntax is good though
#-ifdef CYTH_TEST_SWAP
_ax2_label = [[label] * nFeat for (label, nFeat) in label_nFeat_iter]
#-else
'''
_ax2_label = [[label] * nFeat for (label, nFeat) in label_nFeat_iter]
# endif is optional. the end of the functionscope counts as an #endif
'#-endif'
# Flatten generators into the inverted index
_flatlabels = utool.iflatten(_ax2_label)
_flatfeatxs = utool.iflatten(_ax2_fx)
idx2_label = np.fromiter(_flatlabels, np.int32, nFeats)
idx2_fx = np.fromiter(_flatfeatxs, np.int32, nFeats)
# Stack vecsriptors into numpy array corresponding to inverted inexed
# This might throw a MemoryError
idx2_vec = np.vstack(vecs_list)
'#pragma cyth_returntup'
return idx2_vec, idx2_label, idx2_fx
def get_annotmatch_rowids_between(ibs, aids1, aids2, method=None):
"""
Example:
>>> # ENABLE_DOCTEST
>>> from wbia.annotmatch_funcs import * # NOQA
>>> import wbia
>>> ibs = wbia.opendb('PZ_MTEST')
>>> aids1 = aids2 = [1, 2, 3, 4, 5, 6]
>>> rowids_between = ibs.get_annotmatch_rowids_between
>>> ams1 = sorted(rowids_between(aids1, aids2, method=1))
>>> ams2 = sorted(rowids_between(aids1, aids2, method=2))
>>> assert len(ub.find_duplicates(ams1)) == 0
>>> assert len(ub.find_duplicates(ams2)) == 0
>>> assert sorted(ams2) == sorted(ams1)
"""
if method is None:
if len(aids1) * len(aids2) > 5000:
method = 1
else:
method = 2
if method == 1:
# Strategy 1: get all existing rows and see what intersects
# This is better when the enumerated set of rows would be larger than
# the database size
unflat_rowids1L = ibs.get_annotmatch_rowids_from_aid1(aids1)
unflat_rowids1R = ibs.get_annotmatch_rowids_from_aid2(aids1)
unflat_rowids2L = ibs.get_annotmatch_rowids_from_aid1(aids2)
unflat_rowids2R = ibs.get_annotmatch_rowids_from_aid2(aids2)
am_rowids1L = {
r
for r in ut.iflatten(unflat_rowids1L) if r is not None
}
am_rowids1R = {
r
for r in ut.iflatten(unflat_rowids1R) if r is not None
}
am_rowids2L = {
r
for r in ut.iflatten(unflat_rowids2L) if r is not None
}
am_rowids2R = {
r
for r in ut.iflatten(unflat_rowids2R) if r is not None
}
ams12 = am_rowids1L.intersection(am_rowids2R)
ams21 = am_rowids2L.intersection(am_rowids1R)
ams = sorted(ams12.union(ams21))
# ams = sorted(am_rowids1.intersection(am_rowids2))
# rowids2 = ibs.get_annotmatch_rowids_from_aid2(aid_list)
# unflat_rowids1 = ibs.get_annotmatch_rowids_from_aid(aids1)
# unflat_rowids2 = ibs.get_annotmatch_rowids_from_aid(aids2)
# am_rowids1 = {r for r in ut.iflatten(unflat_rowids1) if r is not None}
# am_rowids2 = {r for r in ut.iflatten(unflat_rowids2) if r is not None}
# ams = sorted(am_rowids1.intersection(am_rowids2))
# ams = ut.isect(am_rowids1, am_rowids2)
elif method == 2:
# Strategy 2: enumerate what rows could exist and see what does exist
# This is better when the enumerated set of rows would be smaller than
# the database size
edges = list(ut.product_nonsame(aids1, aids2))
if len(edges) == 0:
ams = []
else:
aids1_, aids2_ = ut.listT(edges)
# ams = ibs.get_annotmatch_rowid_from_undirected_superkey(aids1_, aids2_)
ams = ibs.get_annotmatch_rowid_from_superkey(aids1_, aids2_)
if ams is None:
ams = []
ams = ut.filter_Nones(ams)
return ams
def build_daid2_chipmatch2(invindex, common_wxs, wx2_qaids, wx2_qfxs,
scores_list, daids_list, query_sccw):
"""
Builds explicit chipmatches that the rest of the pipeline plays nice with
Notation:
An explicit cmtup_old is a tuple (fm, fs, fk) feature_matches,
feature_scores, and feature_ranks.
Let N be the number of matches
A feature match, fm{shape=(N, 2), dtype=int32}, is an array where the first
column corresponds to query_feature_indexes (qfx) and the second column
corresponds to database_feature_indexes (dfx).
A feature score, fs{shape=(N,), dtype=float64} is an array of scores
A feature rank, fk{shape=(N,), dtype=int16} is an array of ranks
Returns:
daid2_chipmatch (dict) : (daid2_fm, daid2_fs, daid2_fk)
Return Format::
daid2_fm (dict): {daid: fm, ...}
daid2_fs (dict): {daid: fs, ...}
daid2_fk (dict): {daid: fk, ...}
Example:
>>> from ibeis.algo.hots.smk.smk_core import * # NOQA
>>> from ibeis.algo.hots.smk import smk_debug
>>> ibs, invindex, qindex, qparams = smk_debug.testdata_match_kernel_L2()
>>> wx2_qrvecs, wx2_qmaws, wx2_qaids, wx2_qfxs, query_sccw = qindex
>>> smk_alpha = ibs.cfg.query_cfg.smk_cfg.smk_alpha
>>> smk_thresh = ibs.cfg.query_cfg.smk_cfg.smk_thresh
>>> withinfo = True # takes an 11s vs 2s
>>> args = (wx2_qrvecs, wx2_qmaws, wx2_qaids, wx2_qfxs, query_sccw, invindex, withinfo, smk_alpha, smk_thresh)
>>> retL1 = match_kernel_L1(*args)
>>> (daid2_totalscore, common_wxs, scores_list, daids_list, idf_list, daid_agg_keys,) = retL1
>>> daid2_chipmatch_old = build_daid2_chipmatch2(invindex, common_wxs, wx2_qaids, wx2_qfxs, scores_list, daids_list, query_sccw)
>>> daid2_chipmatch_new = build_daid2_chipmatch3(invindex, common_wxs, wx2_qaids, wx2_qfxs, scores_list, daids_list, query_sccw)
>>> print(utool.is_dicteq(daid2_chipmatch_old[0], daid2_chipmatch_new[0]))
>>> print(utool.is_dicteq(daid2_chipmatch_old[2], daid2_chipmatch_new[2]))
>>> print(utool.is_dicteq(daid2_chipmatch_old[1], daid2_chipmatch_new[1]))
%timeit build_daid2_chipmatch2(invindex, common_wxs, wx2_qaids, wx2_qfxs, scores_list, daids_list, query_sccw)
%timeit build_daid2_chipmatch3(invindex, common_wxs, wx2_qaids, wx2_qfxs, scores_list, daids_list, query_sccw)
"""
# FIXME: move groupby to vtool
if utool.VERBOSE:
print('[smk_core] build cmtup_old')
wx2_dfxs = invindex.wx2_fxs
daid2_sccw = invindex.daid2_sccw
qfxs_list = [wx2_qfxs[wx] for wx in common_wxs]
dfxs_list = [wx2_dfxs[wx] for wx in common_wxs]
shapes_list = [scores.shape for scores in scores_list] # 51us
shape_ranges = [(mem_arange(w), mem_arange(h)) for (w, h) in shapes_list] # 230us
ijs_list = [mem_meshgrid(wrange, hrange) for (wrange, hrange) in shape_ranges] # 278us
# Normalize scores for words, nMatches, and query sccw (still need daid sccw)
nscores_iter = (scores * query_sccw for scores in scores_list)
# FIXME: Preflatten all of these lists
out_ijs = [
list(zip(_is.flat, _js.flat))
for (_is, _js) in ijs_list
]
out_qfxs = [
[qfxs[ix] for (ix, jx) in ijs]
for (qfxs, ijs) in zip(qfxs_list, out_ijs)
]
out_dfxs = [
[dfxs[jx] for (ix, jx) in ijs]
for (dfxs, ijs) in zip(dfxs_list, out_ijs)
]
out_daids = (
[daids[jx] for (ix, jx) in ijs]
for (daids, ijs) in zip(daids_list, out_ijs)
)
out_scores = (
[nscores[ijx] for ijx in ijs]
for (nscores, ijs) in zip(nscores_iter, out_ijs)
)
nested_fm_iter = [
[
tuple(product(qfxs_, dfxs_))
for qfxs_, dfxs_ in zip(qfxs, dfxs)
]
for qfxs, dfxs in zip(out_qfxs, out_dfxs)
]
all_fms = np.array(list(utool.iflatten(utool.iflatten(nested_fm_iter))), dtype=hstypes.FM_DTYPE)
nested_nmatch_list = [[len(fm) for fm in fms] for fms in nested_fm_iter]
nested_daid_iter = (
[
[daid] * nMatch
for nMatch, daid in zip(nMatch_list, daids)
]
for nMatch_list, daids in zip(nested_nmatch_list, out_daids)
)
nested_score_iter = (
[
[score / nMatch] * nMatch
for nMatch, score in zip(nMatch_list, scores)
]
for nMatch_list, scores in zip(nested_nmatch_list, out_scores)
)
all_daids_ = np.array(list(utool.iflatten(utool.iflatten(nested_daid_iter))), dtype=hstypes.INDEX_TYPE)
all_fss = np.array(list(utool.iflatten(utool.iflatten(nested_score_iter))), dtype=hstypes.FS_DTYPE)
# Filter out 0 scores
keep_xs = np.where(all_fss > 0)[0]
all_fss = all_fss.take(keep_xs)
all_fms = all_fms.take(keep_xs, axis=0)
all_daids_ = all_daids_.take(keep_xs)
daid_keys, groupxs = clustertool.group_indices(all_daids_)
fs_list = clustertool.apply_grouping(all_fss, groupxs)
fm_list = clustertool.apply_grouping(all_fms, groupxs)
daid2_fm = {daid: fm for daid, fm in zip(daid_keys, fm_list)}
daid2_fs = {daid: fs * daid2_sccw[daid] for daid, fs in zip(daid_keys, fs_list)}
# FIXME: generalize to when nAssign > 1
daid2_fk = {daid: np.ones(fs.size, dtype=hstypes.FK_DTYPE) for daid, fs in zip(daid_keys, fs_list)}
daid2_chipmatch = (daid2_fm, daid2_fs, daid2_fk)
return daid2_chipmatch
def stagger_group(list_):
return ut.filter_Nones(ut.iflatten(zip_longest(*list_)))
aug_graph = graph.copy()
# remove cut edges from augmented graph
edge_to_iscut = nx.get_edge_attributes(aug_graph, 'is_cut')
cut_edges = [
(u, v)
for (u, v, d) in aug_graph.edges(data=True)
if not (d.get('is_cut') or d.get('decision', 'unreviewed') in ['nomatch'])
]
cut_edges = [edge for edge, flag in edge_to_iscut.items() if flag]
aug_graph.remove_edges_from(cut_edges)
# Enumerate cliques inside labels
unflat_edges = [list(ut.itertwo(nodes)) for nodes in label_to_nodes.values()]
node_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
# Remove candidate MST edges that exist in the original graph
orig_edges = list(aug_graph.edges())
candidate_mst_edges = [edge for edge in node_pairs if not aug_graph.has_edge(*edge)]
# randomness prevents chains and visually looks better
rng = np.random.RandomState(42)
def _randint():
return 0
return rng.randint(0, 100)
aug_graph.add_edges_from(candidate_mst_edges)
# Weight edges in aug_graph such that existing edges are chosen
def build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list):
""" helper
these list comprehensions replace the prevous for loop
they still need to be optimized a little bit (and made clearer)
can probably unnest the list comprehensions as well
"""
"""
IGNORE
Legacy::
def old_build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list):
fm_nestlist_ = []
fs_nestlist_ = []
daid_nestlist_ = []
for scores, qfxs, dfxs, daids in zip(sparse_list, qfxs_list, dfxs_list, daids_list):
for rx, cx, score in zip(scores.row, scores.col, scores.data):
_fm = tuple(product(qfxs[rx], dfxs[cx]))
_fs = [score / len(_fm)] * len(_fm)
_daid = [daids[cx]] * len(_fm)
fm_nestlist_.append(_fm)
fs_nestlist_.append(_fs)
daid_nestlist_.append(_daid)
return fm_nestlist_, fs_nestlist_, daid_nestlist_
oldtup_ = old_build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list)
fm_nestlist_, fs_nestlist_, daid_nestlist_ = oldtup_
newtup_ = build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list)
fm_nestlist, fs_nestlist, daid_nestlist = newtup_
assert fm_nestlist == fm_nestlist_
assert fs_nestlist == fs_nestlist_
assert daid_nestlist == daid_nestlist_
47ms
%timeit build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list)
59ms
%timeit old_build_correspondences(sparse_list, qfxs_list, dfxs_list, daids_list)
IGNORE
"""
# FIXME: rewrite double comprehension as a flat comprehension
# Build nested feature matches (a single match might have many members)
fm_nestlist = [
tuple(product(qfxs[rx], dfxs[cx]))
for scores, qfxs, dfxs in zip(sparse_list, qfxs_list, dfxs_list)
for rx, cx in zip(scores.row, scores.col)
]
nFm_list = [len(fm) for fm in fm_nestlist]
#fs_unsplit = (score
# for scores in sparse_list
# for score in scores.data)
#daid_unsplit = (daids[cx]
# for scores, daids in zip(sparse_list, daids_list)
# for cx in scores.col)
# Build nested feature scores
fs_unsplit = utool.iflatten(
(scores.data for scores in sparse_list))
# Build nested feature matches (a single match might have many members)
daid_unsplit = utool.iflatten(
(daids.take(scores.col)
for scores, daids in zip(sparse_list, daids_list)))
# Expand feature scores and daids splitting scores amongst match members
fs_nestlist = [
[score / nFm] * nFm
for score, nFm in zip(fs_unsplit, nFm_list)
]
daid_nestlist = [
[daid] * nFm
for daid, nFm in zip(daid_unsplit, nFm_list)
]
if DEBUG_SMK:
assert len(fm_nestlist) == len(fs_nestlist), 'inconsistent len'
assert len(fm_nestlist) == len(nFm_list), 'inconsistent len'
assert len(daid_nestlist) == len(fs_nestlist), 'inconsistent len'
min_ = min(2, len(nFm_list))
max_ = min(15, len(nFm_list))
print('nFm_list[_min:_max] = ' + utool.list_str(nFm_list[min_:max_]))
print('fm_nestlist[_min:_max] = ' + utool.list_str(fm_nestlist[min_:max_]))
print('fs_nestlist[_min:_max] = ' + utool.list_str(fs_nestlist[min_:max_]))
print('daid_nestlist[_min:_max] = ' + utool.list_str(daid_nestlist[min_:max_]))
for fm_, fs_, daid_ in zip(fm_nestlist, fs_nestlist, daid_nestlist):
assert len(fm_) == len(fs_), 'inconsistent len'
assert len(fm_) == len(daid_), 'inconsistent len'
print('[smk_core] checked build_chipmatch correspondence ...ok')
return fm_nestlist, fs_nestlist, daid_nestlist
def compute_data_sccw_(idx2_daid, wx2_drvecs, wx2_dflags, wx2_aids, wx2_idf,
wx2_dmaws, smk_alpha, smk_thresh, verbose=False):
"""
Computes sccw normalization scalar for the database annotations.
This is gamma from the SMK paper.
sccw is a self consistency critiron weight --- a scalar which ensures
the score of K(X, X) = 1
Args:
idx2_daid ():
wx2_drvecs ():
wx2_aids ():
wx2_idf ():
wx2_dmaws ():
smk_alpha ():
smk_thresh ():
Returns:
daid2_sccw
Example:
>>> # SLOW_DOCTEST
>>> from ibeis.algo.hots.smk.smk_index import * # NOQA
>>> from ibeis.algo.hots.smk import smk_index
>>> from ibeis.algo.hots.smk import smk_debug
>>> #tup = smk_debug.testdata_compute_data_sccw(db='testdb1')
>>> tup = smk_debug.testdata_compute_data_sccw(db='PZ_MTEST')
>>> ibs, annots_df, invindex, wx2_idxs, wx2_idf, wx2_drvecs, wx2_aids, qparams = tup
>>> wx2_dflags = invindex.wx2_dflags
>>> ws2_idxs = invindex.wx2_idxs
>>> wx2_dmaws = invindex.wx2_dmaws
>>> idx2_daid = invindex.idx2_daid
>>> daids = invindex.daids
>>> smk_alpha = qparams.smk_alpha
>>> smk_thresh = qparams.smk_thresh
>>> wx2_idf = wx2_idf
>>> verbose = True
>>> invindex.invindex_dbgstr()
>>> invindex.report_memory()
>>> invindex.report_memsize()
>>> daid2_sccw = smk_index.compute_data_sccw_(idx2_daid, wx2_drvecs, wx2_dflags, wx2_aids, wx2_idf, wx2_dmaws, smk_alpha, smk_thresh, verbose)
"""
#for wx in wx_sublist:
# print(len(wx2_dmaws
verbose_ = ut.VERBOSE or verbose
if ut.DEBUG2:
from ibeis.algo.hots.smk import smk_debug
smk_debug.check_wx2(wx2_rvecs=wx2_drvecs, wx2_aids=wx2_aids)
if not ut.QUIET:
print('\n[smk_index.sccw] +--- Start Compute Data Self Consistency Weight')
if verbose_:
print('[smk_index.sccw] Compute SCCW smk_alpha=%r, smk_thresh=%r: ' % (smk_alpha, smk_thresh))
mark1, end1_ = ut.log_progress(
'[smk_index.sccw] SCCW group (by present words): ', len(wx2_drvecs),
freq=100, with_time=WITH_TOTALTIME)
# Group by daids first and then by word index
# Get list of aids and rvecs w.r.t. words (ie one item per word)
wx_sublist = np.array(list(wx2_drvecs.keys()))
aids_perword = [wx2_aids[wx] for wx in wx_sublist]
# wx_list1: Lays out word indexes for each annotation
# tx_list1: Temporary within annotation subindex + wx uniquely identifies
# item in wx2_drvecs, wx2_dflags, and wx2_dmaws
# Flatten out indexes to perform grouping
flat_aids = np.hstack(aids_perword)
count = len(flat_aids)
txs_perword = [np.arange(aids.size) for aids in aids_perword]
flat_txs = np.hstack(txs_perword)
# fromiter is faster for flat_wxs because is not a list of numpy arrays
wxs_perword = ([wx] * len(aids) for wx, aids in zip(wx_sublist, aids_perword))
flat_wxs = np.fromiter(ut.iflatten(wxs_perword), hstypes.INDEX_TYPE, count)
# Group flat indexes by annotation id
unique_aids, annot_groupxs = clustertool.group_indices(flat_aids)
# Wxs and Txs grouped by annotation id
wxs_perannot = clustertool.apply_grouping_iter(flat_wxs, annot_groupxs)
txs_perannot = clustertool.apply_grouping_iter(flat_txs, annot_groupxs)
# Group by word inside each annotation group
wxsubgrouping_perannot = [clustertool.group_indices(wxs)
for wxs in wxs_perannot]
word_groupxs_perannot = (groupxs for wxs, groupxs in wxsubgrouping_perannot)
txs_perword_perannot = [clustertool.apply_grouping(txs, groupxs)
for txs, groupxs in
zip(txs_perannot, word_groupxs_perannot)]
wxs_perword_perannot = [wxs for wxs, groupxs in wxsubgrouping_perannot]
# Group relavent data for sccw measure by word for each annotation grouping
def _vector_subgroup_by_wx(wx2_arr, wxs_perword_perannot, txs_perword_perannot):
return [[wx2_arr[wx].take(txs, axis=0)
for wx, txs in zip(wx_perword_, txs_perword_)]
for wx_perword_, txs_perword_ in
zip(wxs_perword_perannot, txs_perword_perannot)]
def _scalar_subgroup_by_wx(wx2_scalar, wxs_perword_perannot):
return [[wx2_scalar[wx] for wx in wxs] for wxs in wxs_perword_perannot]
subgrouped_drvecs = _vector_subgroup_by_wx(wx2_drvecs, wxs_perword_perannot, txs_perword_perannot)
subgrouped_dmaws = _vector_subgroup_by_wx(wx2_dmaws, wxs_perword_perannot, txs_perword_perannot)
# If we aren't using dmaws replace it with an infinite None iterator
#subgrouped_dmaws = iter(lambda: None, 1)
subgrouped_dflags = _vector_subgroup_by_wx(wx2_dflags, wxs_perword_perannot, txs_perword_perannot)
#subgrouped_dflags = iter(lambda: None, 1)
subgrouped_idfs = _scalar_subgroup_by_wx(wx2_idf, wxs_perword_perannot)
if verbose_:
end1_()
mark2, end2_ = ut.log_progress(lbl='[smk_index.sccw] SCCW Sum (over daid): ',
total=len(unique_aids), freq=100, with_time=WITH_TOTALTIME)
progiter = ut.ProgressIter(lbl='[smk_index.sccw] SCCW Sum (over daid): ',
total=len(unique_aids), freq=10, with_time=WITH_TOTALTIME)
else:
progiter = ut.identity
if ut.DEBUG2:
from ibeis.algo.hots.smk import smk_debug
smk_debug.check_data_smksumm(subgrouped_idfs, subgrouped_drvecs)
sccw_list = [
smk_scoring.sccw_summation(rvecs_list, flags_list, idf_list, maws_list, smk_alpha, smk_thresh)
for rvecs_list, flags_list, maws_list, idf_list in
progiter(zip(subgrouped_drvecs, subgrouped_dflags, subgrouped_dmaws, subgrouped_idfs))
]
daid2_sccw = dict(zip(unique_aids, sccw_list))
if verbose_:
end2_()
print('[smk_index.sccw] L___ End Compute Data SCCW\n')
return daid2_sccw
def make_netx_graph_from_aid_groups(ibs, aids_list, only_reviewed_matches=True,
invis_edges=None, ensure_edges=None,
temp_nids=None, allow_directed=False):
r"""
Args:
ibs (ibeis.IBEISController): image analysis api
aids_list (list):
Example:
>>> # DISABLE_DOCTEST
>>> from ibeis.viz.viz_graph import * # NOQA
>>> import ibeis
>>> ibs = ibeis.opendb(defaultdb='testdb1')
>>> aids_list = [[1, 2, 3, 4], [5, 6, 7]]
>>> invis_edges = [(1, 5)]
>>> only_reviewed_matches = True
>>> graph = make_netx_graph_from_aid_groups(ibs, aids_list,
>>> only_reviewed_matches,
>>> invis_edges)
>>> list(nx.connected_components(graph.to_undirected()))
"""
#aids_list, nid_list = ibs.group_annots_by_name(aid_list)
unique_aids = list(ut.flatten(aids_list))
# grouped version
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
if only_reviewed_matches:
annotmatch_rowids = ibs.get_annotmatch_rowid_from_superkey(aids1, aids2)
annotmatch_rowids = ut.filter_Nones(annotmatch_rowids)
aids1 = ibs.get_annotmatch_aid1(annotmatch_rowids)
aids2 = ibs.get_annotmatch_aid2(annotmatch_rowids)
graph = make_netx_graph_from_aidpairs(ibs, aids1, aids2, unique_aids=unique_aids)
if ensure_edges is not None:
if ensure_edges == 'all':
ensure_edges = list(ut.upper_diag_self_prodx(list(graph.nodes())))
ensure_edges_ = []
for edge in ensure_edges:
edge = tuple(edge)
redge = tuple(edge[::-1]) # HACK
if graph.has_edge(*edge):
ensure_edges_.append(edge)
pass
#nx.set_edge_attributes(graph, 'weight', {edge: .001})
elif (not allow_directed) and graph.has_edge(*redge):
ensure_edges_.append(redge)
#nx.set_edge_attributes(graph, 'weight', {redge: .001})
pass
else:
ensure_edges_.append(edge)
#graph.add_edge(*edge, weight=.001)
graph.add_edge(*edge)
if temp_nids is None:
unique_nids = ibs.get_annot_nids(list(graph.nodes()))
else:
# HACK
unique_nids = [1] * len(list(graph.nodes()))
#unique_nids = temp_nids
nx.set_node_attributes(graph, 'nid', dict(zip(graph.nodes(), unique_nids)))
import plottool as pt
ensure_names_are_connected(graph, aids_list)
# Color edges by nid
color_by_nids(graph, unique_nids=unique_nids)
if invis_edges:
for edge in invis_edges:
if graph.has_edge(*edge):
nx.set_edge_attributes(graph, 'style', {edge: 'invis'})
nx.set_edge_attributes(graph, 'invisible', {edge: True})
else:
graph.add_edge(*edge, style='invis', invisible=True)
# Hack color images orange
if ensure_edges:
nx.set_edge_attributes(graph, 'color',
{tuple(edge): pt.ORANGE for edge in ensure_edges_})
return graph
def compute_data_sccw_(idx2_daid, wx2_drvecs, wx2_dflags, wx2_aids, wx2_idf,
wx2_dmaws, smk_alpha, smk_thresh, verbose=False):
"""
Computes sccw normalization scalar for the database annotations.
This is gamma from the SMK paper.
sccw is a self consistency critiron weight --- a scalar which ensures
the score of K(X, X) = 1
Args:
idx2_daid ():
wx2_drvecs ():
wx2_aids ():
wx2_idf ():
wx2_dmaws ():
smk_alpha ():
smk_thresh ():
Returns:
daid2_sccw
Example:
>>> # SLOW_DOCTEST
>>> from ibeis.algo.hots.smk.smk_index import * # NOQA
>>> from ibeis.algo.hots.smk import smk_index
>>> from ibeis.algo.hots.smk import smk_debug
>>> #tup = smk_debug.testdata_compute_data_sccw(db='testdb1')
>>> tup = smk_debug.testdata_compute_data_sccw(db='PZ_MTEST')
>>> ibs, annots_df, invindex, wx2_idxs, wx2_idf, wx2_drvecs, wx2_aids, qparams = tup
>>> wx2_dflags = invindex.wx2_dflags
>>> ws2_idxs = invindex.wx2_idxs
>>> wx2_dmaws = invindex.wx2_dmaws
>>> idx2_daid = invindex.idx2_daid
>>> daids = invindex.daids
>>> smk_alpha = qparams.smk_alpha
>>> smk_thresh = qparams.smk_thresh
>>> wx2_idf = wx2_idf
>>> verbose = True
>>> invindex.invindex_dbgstr()
>>> invindex.report_memory()
>>> invindex.report_memsize()
>>> daid2_sccw = smk_index.compute_data_sccw_(idx2_daid, wx2_drvecs, wx2_dflags, wx2_aids, wx2_idf, wx2_dmaws, smk_alpha, smk_thresh, verbose)
"""
#for wx in wx_sublist:
# print(len(wx2_dmaws
verbose_ = ut.VERBOSE or verbose
if ut.DEBUG2:
from ibeis.algo.hots.smk import smk_debug
smk_debug.check_wx2(wx2_rvecs=wx2_drvecs, wx2_aids=wx2_aids)
if not ut.QUIET:
print('\n[smk_index.sccw] +--- Start Compute Data Self Consistency Weight')
if verbose_:
print('[smk_index.sccw] Compute SCCW smk_alpha=%r, smk_thresh=%r: ' % (smk_alpha, smk_thresh))
# Group by daids first and then by word index
# Get list of aids and rvecs w.r.t. words (ie one item per word)
wx_sublist = np.array(list(wx2_drvecs.keys()))
aids_perword = [wx2_aids[wx] for wx in wx_sublist]
# wx_list1: Lays out word indexes for each annotation
# tx_list1: Temporary within annotation subindex + wx uniquely identifies
# item in wx2_drvecs, wx2_dflags, and wx2_dmaws
# Flatten out indexes to perform grouping
flat_aids = np.hstack(aids_perword)
count = len(flat_aids)
txs_perword = [np.arange(aids.size) for aids in aids_perword]
flat_txs = np.hstack(txs_perword)
# fromiter is faster for flat_wxs because is not a list of numpy arrays
wxs_perword = ([wx] * len(aids) for wx, aids in zip(wx_sublist, aids_perword))
flat_wxs = np.fromiter(ut.iflatten(wxs_perword), hstypes.INDEX_TYPE, count)
# Group flat indexes by annotation id
unique_aids, annot_groupxs = clustertool.group_indices(flat_aids)
# Wxs and Txs grouped by annotation id
wxs_perannot = clustertool.apply_grouping_iter(flat_wxs, annot_groupxs)
txs_perannot = clustertool.apply_grouping_iter(flat_txs, annot_groupxs)
# Group by word inside each annotation group
wxsubgrouping_perannot = [clustertool.group_indices(wxs)
for wxs in wxs_perannot]
word_groupxs_perannot = (groupxs for wxs, groupxs in wxsubgrouping_perannot)
txs_perword_perannot = [clustertool.apply_grouping(txs, groupxs)
for txs, groupxs in
zip(txs_perannot, word_groupxs_perannot)]
wxs_perword_perannot = [wxs for wxs, groupxs in wxsubgrouping_perannot]
# Group relavent data for sccw measure by word for each annotation grouping
def _vector_subgroup_by_wx(wx2_arr, wxs_perword_perannot, txs_perword_perannot):
return [[wx2_arr[wx].take(txs, axis=0)
for wx, txs in zip(wx_perword_, txs_perword_)]
for wx_perword_, txs_perword_ in
zip(wxs_perword_perannot, txs_perword_perannot)]
def _scalar_subgroup_by_wx(wx2_scalar, wxs_perword_perannot):
return [[wx2_scalar[wx] for wx in wxs] for wxs in wxs_perword_perannot]
subgrouped_drvecs = _vector_subgroup_by_wx(wx2_drvecs, wxs_perword_perannot, txs_perword_perannot)
subgrouped_dmaws = _vector_subgroup_by_wx(wx2_dmaws, wxs_perword_perannot, txs_perword_perannot)
# If we aren't using dmaws replace it with an infinite None iterator
#subgrouped_dmaws = iter(lambda: None, 1)
subgrouped_dflags = _vector_subgroup_by_wx(wx2_dflags, wxs_perword_perannot, txs_perword_perannot)
#subgrouped_dflags = iter(lambda: None, 1)
subgrouped_idfs = _scalar_subgroup_by_wx(wx2_idf, wxs_perword_perannot)
if verbose_:
progiter = ut.ProgressIter(lbl='[smk_index.sccw] SCCW Sum (over daid): ',
total=len(unique_aids), freq=10, with_time=WITH_TOTALTIME)
else:
progiter = ut.identity
if ut.DEBUG2:
from ibeis.algo.hots.smk import smk_debug
smk_debug.check_data_smksumm(subgrouped_idfs, subgrouped_drvecs)
sccw_list = [
smk_scoring.sccw_summation(rvecs_list, flags_list, idf_list, maws_list, smk_alpha, smk_thresh)
for rvecs_list, flags_list, maws_list, idf_list in
progiter(zip(subgrouped_drvecs, subgrouped_dflags, subgrouped_dmaws, subgrouped_idfs))
]
daid2_sccw = dict(zip(unique_aids, sccw_list))
if verbose_:
print('[smk_index.sccw] L___ End Compute Data SCCW\n')
return daid2_sccw
def sccw_summation(rvecs_list, flags_list, idf_list, maws_list, smk_alpha, smk_thresh):
r"""
Computes gamma from "To Aggregate or not to aggregate". Every component in
each list is with repsect to a different word.
scc = self consistency criterion
It is a scalar which ensure K(X, X) = 1
Args:
rvecs_list (list of ndarrays): residual vectors for every word
idf_list (list of floats): idf weight for each word
maws_list (list of ndarrays): multi-assign weights for each word for each residual vector
smk_alpha (float): selectivity power
smk_thresh (float): selectivity threshold
Returns:
float: sccw self-consistency-criterion weight
Math:
\begin{equation}
\gamma(X) = (\sum_{c \in \C} w_c M(X_c, X_c))^{-.5}
\end{equation}
Example:
>>> from ibeis.algo.hots.smk.smk_scoring import * # NOQA
>>> from ibeis.algo.hots.smk import smk_scoring
>>> from ibeis.algo.hots.smk import smk_debug
>>> #idf_list, rvecs_list, maws_list, smk_alpha, smk_thresh, wx2_flags = smk_debug.testdata_sccw_sum(db='testdb1')
>>> tup = smk_debug.testdata_sccw_sum(db='PZ_MTEST', nWords=128000)
>>> idf_list, rvecs_list, flags_list, maws_list, smk_alpha, smk_thresh = tup
>>> sccw = smk_scoring.sccw_summation(rvecs_list, flags_list, idf_list, maws_list, smk_alpha, smk_thresh)
>>> print(sccw)
0.0201041835751
CommandLine:
python smk_match.py --db PZ_MOTHERS --nWords 128
Ignore:
0.0384477314197
qmaws_list = dmaws_list = maws_list
drvecs_list = qrvecs_list = rvecs_list
dflags_list = qflags_list = flags_list
flags_list = flags_list[7:10]
maws_list = maws_list[7:10]
idf_list = idf_list[7:10]
rvecs_list = rvecs_list[7:10]
"""
num_rvecs = len(rvecs_list)
if DEBUG_SMK:
assert maws_list is None or len(maws_list) == num_rvecs, 'inconsistent lengths'
assert num_rvecs == len(idf_list), 'inconsistent lengths'
assert maws_list is None or list(map(len, maws_list)) == list(map(len, rvecs_list)), 'inconsistent per word lengths'
assert flags_list is None or list(map(len, maws_list)) == list(map(len, flags_list)), 'inconsistent per word lengths'
assert flags_list is None or len(flags_list) == num_rvecs, 'inconsistent lengths'
# Indexing with asymetric multi-assignment might get you a non 1 self score?
# List of scores for every word.
scores_list = score_matches(rvecs_list, rvecs_list, flags_list, flags_list,
maws_list, maws_list, smk_alpha, smk_thresh,
idf_list)
if DEBUG_SMK:
assert len(scores_list) == num_rvecs, 'bad rvec and score'
assert len(idf_list) == len(scores_list), 'bad weight and score'
# Summation over all residual vector scores
_count = sum((scores.size for scores in scores_list))
_iter = utool.iflatten(scores.ravel() for scores in scores_list)
self_rawscore = np.fromiter(_iter, np.float64, _count).sum()
# Square root inverse to enforce normalized self-score is 1.0
sccw = np.reciprocal(np.sqrt(self_rawscore))
try:
assert not np.isinf(sccw), 'sccw cannot be infinite'
assert not np.isnan(sccw), 'sccw cannot be nan'
except AssertionError as ex:
utool.printex(ex, 'problem computing self consistency criterion weight',
keys=['num_rvecs'], iswarning=True)
if num_rvecs > 0:
raise
else:
sccw = 1
return sccw
def stagger_group(list_):
return ut.filter_Nones(ut.iflatten(zip_longest(*list_)))
def build_daid2_chipmatch2(invindex, common_wxs, wx2_qaids, wx2_qfxs,
scores_list, daids_list, query_sccw):
"""
Builds explicit chipmatches that the rest of the pipeline plays nice with
Notation:
An explicit cmtup_old is a tuple (fm, fs, fk) feature_matches,
feature_scores, and feature_ranks.
Let N be the number of matches
A feature match, fm{shape=(N, 2), dtype=int32}, is an array where the first
column corresponds to query_feature_indexes (qfx) and the second column
corresponds to database_feature_indexes (dfx).
A feature score, fs{shape=(N,), dtype=float64} is an array of scores
A feature rank, fk{shape=(N,), dtype=int16} is an array of ranks
Returns:
daid2_chipmatch (dict) : (daid2_fm, daid2_fs, daid2_fk)
Return Format::
daid2_fm (dict): {daid: fm, ...}
daid2_fs (dict): {daid: fs, ...}
daid2_fk (dict): {daid: fk, ...}
Example:
>>> from ibeis.algo.hots.smk.smk_core import * # NOQA
>>> from ibeis.algo.hots.smk import smk_debug
>>> ibs, invindex, qindex, qparams = smk_debug.testdata_match_kernel_L2()
>>> wx2_qrvecs, wx2_qmaws, wx2_qaids, wx2_qfxs, query_sccw = qindex
>>> smk_alpha = ibs.cfg.query_cfg.smk_cfg.smk_alpha
>>> smk_thresh = ibs.cfg.query_cfg.smk_cfg.smk_thresh
>>> withinfo = True # takes an 11s vs 2s
>>> args = (wx2_qrvecs, wx2_qmaws, wx2_qaids, wx2_qfxs, query_sccw, invindex, withinfo, smk_alpha, smk_thresh)
>>> retL1 = match_kernel_L1(*args)
>>> (daid2_totalscore, common_wxs, scores_list, daids_list, idf_list, daid_agg_keys,) = retL1
>>> daid2_chipmatch_old = build_daid2_chipmatch2(invindex, common_wxs, wx2_qaids, wx2_qfxs, scores_list, daids_list, query_sccw)
>>> daid2_chipmatch_new = build_daid2_chipmatch3(invindex, common_wxs, wx2_qaids, wx2_qfxs, scores_list, daids_list, query_sccw)
>>> print(utool.is_dicteq(daid2_chipmatch_old[0], daid2_chipmatch_new[0]))
>>> print(utool.is_dicteq(daid2_chipmatch_old[2], daid2_chipmatch_new[2]))
>>> print(utool.is_dicteq(daid2_chipmatch_old[1], daid2_chipmatch_new[1]))
%timeit build_daid2_chipmatch2(invindex, common_wxs, wx2_qaids, wx2_qfxs, scores_list, daids_list, query_sccw)
%timeit build_daid2_chipmatch3(invindex, common_wxs, wx2_qaids, wx2_qfxs, scores_list, daids_list, query_sccw)
"""
# FIXME: move groupby to vtool
if utool.VERBOSE:
print('[smk_core] build cmtup_old')
wx2_dfxs = invindex.wx2_fxs
daid2_sccw = invindex.daid2_sccw
qfxs_list = [wx2_qfxs[wx] for wx in common_wxs]
dfxs_list = [wx2_dfxs[wx] for wx in common_wxs]
shapes_list = [scores.shape for scores in scores_list] # 51us
shape_ranges = [(mem_arange(w), mem_arange(h))
for (w, h) in shapes_list] # 230us
ijs_list = [
mem_meshgrid(wrange, hrange) for (wrange, hrange) in shape_ranges
] # 278us
# Normalize scores for words, nMatches, and query sccw (still need daid sccw)
nscores_iter = (scores * query_sccw for scores in scores_list)
# FIXME: Preflatten all of these lists
out_ijs = [list(zip(_is.flat, _js.flat)) for (_is, _js) in ijs_list]
out_qfxs = [[qfxs[ix] for (ix, jx) in ijs]
for (qfxs, ijs) in zip(qfxs_list, out_ijs)]
out_dfxs = [[dfxs[jx] for (ix, jx) in ijs]
for (dfxs, ijs) in zip(dfxs_list, out_ijs)]
out_daids = ([daids[jx] for (ix, jx) in ijs]
for (daids, ijs) in zip(daids_list, out_ijs))
out_scores = ([nscores[ijx] for ijx in ijs]
for (nscores, ijs) in zip(nscores_iter, out_ijs))
nested_fm_iter = [[
tuple(product(qfxs_, dfxs_)) for qfxs_, dfxs_ in zip(qfxs, dfxs)
] for qfxs, dfxs in zip(out_qfxs, out_dfxs)]
all_fms = np.array(list(utool.iflatten(utool.iflatten(nested_fm_iter))),
dtype=hstypes.FM_DTYPE)
nested_nmatch_list = [[len(fm) for fm in fms] for fms in nested_fm_iter]
nested_daid_iter = ([
[daid] * nMatch for nMatch, daid in zip(nMatch_list, daids)
] for nMatch_list, daids in zip(nested_nmatch_list, out_daids))
nested_score_iter = ([
[score / nMatch] * nMatch
for nMatch, score in zip(nMatch_list, scores)
] for nMatch_list, scores in zip(nested_nmatch_list, out_scores))
all_daids_ = np.array(list(utool.iflatten(
utool.iflatten(nested_daid_iter))),
dtype=hstypes.INDEX_TYPE)
all_fss = np.array(list(utool.iflatten(utool.iflatten(nested_score_iter))),
dtype=hstypes.FS_DTYPE)
# Filter out 0 scores
keep_xs = np.where(all_fss > 0)[0]
all_fss = all_fss.take(keep_xs)
all_fms = all_fms.take(keep_xs, axis=0)
all_daids_ = all_daids_.take(keep_xs)
daid_keys, groupxs = clustertool.group_indices(all_daids_)
fs_list = clustertool.apply_grouping(all_fss, groupxs)
fm_list = clustertool.apply_grouping(all_fms, groupxs)
daid2_fm = {daid: fm for daid, fm in zip(daid_keys, fm_list)}
daid2_fs = {
daid: fs * daid2_sccw[daid]
for daid, fs in zip(daid_keys, fs_list)
}
# FIXME: generalize to when nAssign > 1
daid2_fk = {
daid: np.ones(fs.size, dtype=hstypes.FK_DTYPE)
for daid, fs in zip(daid_keys, fs_list)
}
daid2_chipmatch = (daid2_fm, daid2_fs, daid2_fk)
return daid2_chipmatch
def get_annotmatch_subgraph(ibs):
r"""
http://bokeh.pydata.org/en/latest/
https://github.com/jsexauer/networkx_viewer
TODO: Need a special visualization
In the web I need:
* graph of annotations matches.
* can move them around.
* edit lines between them.
* http://stackoverflow.com/questions/15373530/web-graph-visualization-tool
This should share functionality with a name view.
Args:
ibs (IBEISController): ibeis controller object
CommandLine:
python -m ibeis.annotmatch_funcs --exec-get_annotmatch_subgraph --show
# Networkx example
python -m ibeis.viz.viz_graph --test-show_chipmatch_graph:0 --show
Ignore:
from ibeis import viz
Example:
>>> # ENABLE_DOCTEST
>>> from ibeis.annotmatch_funcs import * # NOQA
>>> import ibeis
>>> ibs = ibeis.opendb(defaultdb='PZ_MTEST')
>>> result = get_annotmatch_subgraph(ibs)
>>> ut.show_if_requested()
"""
#import ibeis
#ibs = ibeis.opendb(db='PZ_MTEST')
#rowids = ibs._get_all_annotmatch_rowids()
#aids1 = ibs.get_annotmatch_aid1(rowids)
#aids2 = ibs.get_annotmatch_aid2(rowids)
#
#
nids = ibs.get_valid_nids()
nids = nids[0:5]
aids_list = ibs.get_name_aids(nids)
import itertools
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
# Enumerate annotmatch properties
rng = np.random.RandomState(0)
edge_props = {
'weight': rng.rand(len(aids1)),
'reviewer_confidence': rng.rand(len(aids1)),
'algo_confidence': rng.rand(len(aids1)),
}
# Remove data that does not need to be visualized
# (dont show all the aids if you dont have to)
thresh = .5
flags = edge_props['weight'] > thresh
aids1 = ut.compress(aids1, flags)
aids2 = ut.compress(aids2, flags)
edge_props = {key: ut.compress(val, flags) for key, val in edge_props.items()}
edge_keys = list(edge_props.keys())
edge_vals = ut.dict_take(edge_props, edge_keys)
unique_aids = list(set(aids1 + aids2))
# Make a graph between the chips
nodes = list(zip(unique_aids))
edges = list(zip(aids1, aids2, *edge_vals))
node_lbls = [('aid', 'int')]
edge_lbls = [('weight', 'float')]
from ibeis.viz import viz_graph
netx_graph = viz_graph.make_netx_graph(nodes, edges, node_lbls, edge_lbls)
fnum = None
#zoom = kwargs.get('zoom', .4)
zoom = .4
viz_graph.viz_netx_chipgraph(ibs, netx_graph, fnum=fnum, with_images=True, zoom=zoom)
def get_annotmatch_subgraph(ibs):
r"""
http://bokeh.pydata.org/en/latest/
https://github.com/jsexauer/networkx_viewer
TODO: Need a special visualization
In the web I need:
* graph of annotations matches.
* can move them around.
* edit lines between them.
* http://stackoverflow.com/questions/15373530/web-graph-visualization-tool
This should share functionality with a name view.
Args:
ibs (IBEISController): ibeis controller object
CommandLine:
python -m ibeis.annotmatch_funcs --exec-get_annotmatch_subgraph --show
# Networkx example
python -m ibeis.viz.viz_graph --test-show_chipmatch_graph:0 --show
Ignore:
from ibeis import viz
Example:
>>> # ENABLE_DOCTEST
>>> from ibeis.annotmatch_funcs import * # NOQA
>>> import ibeis
>>> ibs = ibeis.opendb(defaultdb='PZ_MTEST')
>>> result = get_annotmatch_subgraph(ibs)
>>> ut.show_if_requested()
"""
#import ibeis
#ibs = ibeis.opendb(db='PZ_MTEST')
#rowids = ibs._get_all_annotmatch_rowids()
#aids1 = ibs.get_annotmatch_aid1(rowids)
#aids2 = ibs.get_annotmatch_aid2(rowids)
#
#
nids = ibs.get_valid_nids()
nids = nids[0:5]
aids_list = ibs.get_name_aids(nids)
import itertools
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
# Enumerate annotmatch properties
rng = np.random.RandomState(0)
edge_props = {
'weight': rng.rand(len(aids1)),
'reviewer_confidence': rng.rand(len(aids1)),
'algo_confidence': rng.rand(len(aids1)),
}
# Remove data that does not need to be visualized
# (dont show all the aids if you dont have to)
thresh = .5
flags = edge_props['weight'] > thresh
aids1_ = ut.compress(aids1, flags)
aids2_ = ut.compress(aids2, flags)
chosen_props = ut.dict_subset(edge_props, ['weight'])
edge_props = ut.map_dict_vals(ut.partial(ut.compress, flag_list=flags),
chosen_props)
edge_keys = list(edge_props.keys())
edge_vals = ut.dict_take(edge_props, edge_keys)
edge_attr_list = [dict(zip(edge_keys, vals_)) for vals_ in zip(*edge_vals)]
unique_aids = list(set(aids1_ + aids2_))
# Make a graph between the chips
nodes = unique_aids
edges = list(zip(aids1_, aids2_, edge_attr_list))
import networkx as nx
graph = nx.DiGraph()
graph.add_nodes_from(nodes)
graph.add_edges_from(edges)
from ibeis.viz import viz_graph
fnum = None
#zoom = kwargs.get('zoom', .4)
viz_graph.viz_netx_chipgraph(ibs,
graph,
fnum=fnum,
with_images=True,
augment_graph=False)
def invert_index(vecs_list, fgws_list, ax_list, fxs_list, verbose=ut.NOT_QUIET):
r"""
Aggregates descriptors of input annotations and returns inverted information
Args:
vecs_list (list):
fgws_list (list):
ax_list (list):
fxs_list (list):
verbose (bool): verbosity flag(default = True)
Returns:
tuple: (idx2_vec, idx2_fgw, idx2_ax, idx2_fx)
CommandLine:
python -m ibeis.algo.hots.neighbor_index invert_index
Example:
>>> # ENABLE_DOCTEST
>>> from ibeis.algo.hots.neighbor_index import * # NOQA
>>> rng = np.random.RandomState(42)
>>> DIM_SIZE = 16
>>> nFeat_list = [3, 0, 4, 1]
>>> vecs_list = [rng.randn(nFeat, DIM_SIZE) for nFeat in nFeat_list]
>>> fgws_list = [rng.randn(nFeat) for nFeat in nFeat_list]
>>> fxs_list = [np.arange(nFeat) for nFeat in nFeat_list]
>>> ax_list = np.arange(len(vecs_list))
>>> fgws_list = None
>>> verbose = True
>>> tup = invert_index(vecs_list, fgws_list, ax_list, fxs_list)
>>> (idx2_vec, idx2_fgw, idx2_ax, idx2_fx) = tup
>>> result = 'output depth_profile = %s' % (ut.depth_profile(tup),)
>>> print(result)
output depth_profile = [(8, 16), 1, 8, 8]
Example:
>>> # ENABLE_DOCTEST
>>> from ibeis.algo.hots.neighbor_index import * # NOQA
>>> import ibeis
>>> qreq_ = ibeis.testdata_qreq_(defaultdb='testdb1', a='default:species=zebra_plains', p='default:fgw_thresh=.999')
>>> vecs_list, fgws_list, fxs_list = get_support_data(qreq_, qreq_.daids)
>>> ax_list = np.arange(len(vecs_list))
>>> input_ = vecs_list, fgws_list, ax_list, fxs_list
>>> print('input depth_profile = %s' % (ut.depth_profile(input_),))
>>> tup = invert_index(*input_)
>>> (idx2_vec, idx2_fgw, idx2_ax, idx2_fx) = tup
>>> result = 'output depth_profile = %s' % (ut.depth_profile(tup),)
>>> print(result)
output depth_profile = [(1912, 128), 1912, 1912, 1912]
"""
if ut.VERYVERBOSE:
print('[nnindex] stacking descriptors from %d annotations' % len(ax_list))
try:
nFeat_list = np.array(list(map(len, vecs_list)))
# Remove input without any features
is_valid = nFeat_list > 0
nFeat_list = nFeat_list.compress(is_valid)
vecs_list = ut.compress(vecs_list, is_valid)
if fgws_list is not None:
fgws_list = ut.compress(fgws_list, is_valid)
ax_list = ut.compress(ax_list, is_valid)
fxs_list = ut.compress(fxs_list, is_valid)
# Flatten into inverted index
axs_list = [[ax] * nFeat for (ax, nFeat) in zip(ax_list, nFeat_list)]
nFeats = sum(nFeat_list)
idx2_ax = np.fromiter(ut.iflatten(axs_list), np.int32, nFeats)
idx2_fx = np.fromiter(ut.iflatten(fxs_list), np.int32, nFeats)
idx2_vec = np.vstack(vecs_list)
if fgws_list is None:
idx2_fgw = None
else:
idx2_fgw = np.hstack(fgws_list)
try:
assert len(idx2_fgw) == len(idx2_vec), 'error. weights and vecs do not correspond'
except Exception as ex:
ut.printex(ex, keys=[(len, 'idx2_fgw'), (len, 'idx2_vec')])
raise
assert idx2_vec.shape[0] == idx2_ax.shape[0]
assert idx2_vec.shape[0] == idx2_fx.shape[0]
except MemoryError as ex:
ut.printex(ex, 'cannot build inverted index', '[!memerror]')
raise
if ut.VERYVERBOSE or verbose:
print('[nnindex] stacked nVecs={nVecs} from nAnnots={nAnnots}'.format(
nVecs=len(idx2_vec), nAnnots=len(ax_list)))
print('[nnindex] idx2_vecs dtype={}, memory={}'.format(
idx2_vec.dtype,
ut.byte_str2(idx2_vec.size * idx2_vec.dtype.itemsize)))
return idx2_vec, idx2_fgw, idx2_ax, idx2_fx
def make_netx_graph_from_aid_groups(ibs,
aids_list,
only_reviewed_matches=True,
invis_edges=None,
ensure_edges=None,
temp_nids=None,
allow_directed=False):
r"""
Args:
ibs (ibeis.IBEISController): image analysis api
aids_list (list):
Example:
>>> # DISABLE_DOCTEST
>>> from ibeis.viz.viz_graph import * # NOQA
>>> import ibeis
>>> ibs = ibeis.opendb(defaultdb='testdb1')
>>> aids_list = [[1, 2, 3, 4], [5, 6, 7]]
>>> invis_edges = [(1, 5)]
>>> only_reviewed_matches = True
>>> graph = make_netx_graph_from_aid_groups(ibs, aids_list,
>>> only_reviewed_matches,
>>> invis_edges)
>>> list(nx.connected_components(graph.to_undirected()))
"""
#aids_list, nid_list = ibs.group_annots_by_name(aid_list)
unique_aids = list(ut.flatten(aids_list))
# grouped version
unflat_edges = (list(itertools.product(aids, aids)) for aids in aids_list)
aid_pairs = [tup for tup in ut.iflatten(unflat_edges) if tup[0] != tup[1]]
aids1 = ut.get_list_column(aid_pairs, 0)
aids2 = ut.get_list_column(aid_pairs, 1)
if only_reviewed_matches:
annotmatch_rowids = ibs.get_annotmatch_rowid_from_superkey(
aids1, aids2)
annotmatch_rowids = ut.filter_Nones(annotmatch_rowids)
aids1 = ibs.get_annotmatch_aid1(annotmatch_rowids)
aids2 = ibs.get_annotmatch_aid2(annotmatch_rowids)
graph = make_netx_graph_from_aidpairs(ibs,
aids1,
aids2,
unique_aids=unique_aids)
if ensure_edges is not None:
if ensure_edges == 'all':
ensure_edges = list(ut.upper_diag_self_prodx(list(graph.nodes())))
ensure_edges_ = []
for edge in ensure_edges:
edge = tuple(edge)
redge = tuple(edge[::-1]) # HACK
if graph.has_edge(*edge):
ensure_edges_.append(edge)
pass
#nx.set_edge_attributes(graph, 'weight', {edge: .001})
elif (not allow_directed) and graph.has_edge(*redge):
ensure_edges_.append(redge)
#nx.set_edge_attributes(graph, 'weight', {redge: .001})
pass
else:
ensure_edges_.append(edge)
#graph.add_edge(*edge, weight=.001)
graph.add_edge(*edge)
if temp_nids is None:
unique_nids = ibs.get_annot_nids(list(graph.nodes()))
else:
# HACK
unique_nids = [1] * len(list(graph.nodes()))
#unique_nids = temp_nids
nx.set_node_attributes(graph, 'nid', dict(zip(graph.nodes(), unique_nids)))
import plottool as pt
ensure_names_are_connected(graph, aids_list)
# Color edges by nid
color_by_nids(graph, unique_nids=unique_nids)
if invis_edges:
for edge in invis_edges:
if graph.has_edge(*edge):
nx.set_edge_attributes(graph, 'style', {edge: 'invis'})
nx.set_edge_attributes(graph, 'invisible', {edge: True})
else:
graph.add_edge(*edge, style='invis', invisible=True)
# Hack color images orange
if ensure_edges:
nx.set_edge_attributes(
graph, 'color', {tuple(edge): pt.ORANGE
for edge in ensure_edges_})
return graph
def invertible_stack(vecs_list, label_list):
"""
Stacks descriptors into a flat structure and returns inverse mapping from
flat database descriptor indexes (dx) to annotation ids (label) and feature
indexes (fx). Feature indexes are w.r.t. annotation indexes.
Output:
idx2_desc - flat descriptor stack
idx2_label - inverted index into annotations
idx2_fx - inverted index into features
# Example with 2D Descriptors
Example:
>>> # DISABLE_DOCTEST
>>> from vtool.nearest_neighbors import * # NOQA
>>> DESC_TYPE = np.uint8
>>> label_list = [1, 2, 3, 4, 5]
>>> vecs_list = [
... np.array([[0, 0], [0, 1]], dtype=DESC_TYPE),
... np.array([[5, 3], [2, 30], [1, 1]], dtype=DESC_TYPE),
... np.empty((0, 2), dtype=DESC_TYPE),
... np.array([[5, 3], [2, 30], [1, 1]], dtype=DESC_TYPE),
... np.array([[3, 3], [42, 42], [2, 6]], dtype=DESC_TYPE),
... ]
>>> idx2_vec, idx2_label, idx2_fx = invertible_stack(vecs_list, label_list)
>>> print(repr(idx2_vec.T))
array([[ 0, 0, 5, 2, 1, 5, 2, 1, 3, 42, 2],
[ 0, 1, 3, 30, 1, 3, 30, 1, 3, 42, 6]], dtype=uint8)
>>> print(repr(idx2_label))
array([1, 1, 2, 2, 2, 4, 4, 4, 5, 5, 5])
>>> print(repr(idx2_fx))
array([0, 1, 0, 1, 2, 0, 1, 2, 0, 1, 2])
"""
# INFER DTYPE? dtype = vecs_list[0].dtype
# Build inverted index of (label, fx) pairs
nFeats = sum(list(map(len, vecs_list)))
nFeat_iter = map(len, vecs_list)
label_nFeat_iter = zip(label_list, map(len, vecs_list))
# generate featx inverted index for each feature in each annotation
_ax2_fx = [list(range(nFeat)) for nFeat in nFeat_iter]
# generate label inverted index for each feature in each annotation
'''
# this is not a real test the code just happened to be here. syntax is good though
#-ifdef CYTH_TEST_SWAP
_ax2_label = [[label] * nFeat for (label, nFeat) in label_nFeat_iter]
#-else
'''
_ax2_label = [[label] * nFeat for (label, nFeat) in label_nFeat_iter]
# endif is optional. the end of the functionscope counts as an #endif
'#-endif'
# Flatten generators into the inverted index
_flatlabels = utool.iflatten(_ax2_label)
_flatfeatxs = utool.iflatten(_ax2_fx)
idx2_label = np.fromiter(_flatlabels, np.int32, nFeats)
idx2_fx = np.fromiter(_flatfeatxs, np.int32, nFeats)
# Stack vecsriptors into numpy array corresponding to inverted inexed
# This might throw a MemoryError
idx2_vec = np.vstack(vecs_list)
'#pragma cyth_returntup'
return idx2_vec, idx2_label, idx2_fx
def match_single(smk, qaid, daids, qreq_, verbose=True):
"""
CommandLine:
python -m wbia.algo.smk.smk_pipeline SMK.match_single --profile
python -m wbia.algo.smk.smk_pipeline SMK.match_single --show
python -m wbia SMK.match_single -a ctrl:qmingt=2 --profile --db PZ_Master1
python -m wbia SMK.match_single -a ctrl --profile --db GZ_ALL
Example:
>>> # FUTURE_ENABLE
>>> from wbia.algo.smk.smk_pipeline import * # NOQA
>>> import wbia
>>> qreq_ = wbia.testdata_qreq_(defaultdb='PZ_MTEST')
>>> ibs = qreq_.ibs
>>> daids = qreq_.daids
>>> #ibs, daids = wbia.testdata_aids(defaultdb='PZ_MTEST', default_set='dcfg')
>>> qreq_ = SMKRequest(ibs, daids[0:1], daids, {'agg': True,
>>> 'num_words': 1000,
>>> 'sv_on': True})
>>> qreq_.ensure_data()
>>> qaid = qreq_.qaids[0]
>>> daids = qreq_.daids
>>> daid = daids[1]
>>> verbose = True
>>> cm = qreq_.smk.match_single(qaid, daids, qreq_)
>>> ut.quit_if_noshow()
>>> ut.qtensure()
>>> cm.ishow_analysis(qreq_)
>>> ut.show_if_requested()
"""
from wbia.algo.hots import chip_match
from wbia.algo.hots import pipeline
alpha = qreq_.qparams['smk_alpha']
thresh = qreq_.qparams['smk_thresh']
agg = qreq_.qparams['agg']
# nAnnotPerName = qreq_.qparams.nAnnotPerNameSVER
sv_on = qreq_.qparams.sv_on
if sv_on:
nNameShortList = qreq_.qparams.nNameShortlistSVER
shortsize = nNameShortList
else:
shortsize = None
X = qreq_.qinva.get_annot(qaid)
# Determine which database annotations need to be checked
# with ut.Timer('searching qaid=%r' % (qaid,), verbose=verbose):
hit_inva_wxs = ut.take(qreq_.dinva.wx_to_aids, X.wx_list)
hit_daids = np.array(list(set(ut.iflatten(hit_inva_wxs))))
# Mark impossible daids
# with ut.Timer('checking impossible daids=%r' % (qaid,), verbose=verbose):
valid_flags = check_can_match(qaid, hit_daids, qreq_)
valid_daids = hit_daids.compress(valid_flags)
shortlist = ut.Shortlist(shortsize)
# gammaX = smk.gamma(X, wx_to_weight, agg, alpha, thresh)
_prog = ut.ProgPartial(lbl='smk scoring qaid=%r' % (qaid, ),
enabled=verbose,
bs=True,
adjust=True)
wx_to_weight = qreq_.dinva.wx_to_weight
debug = False
if debug:
qnid = qreq_.get_qreq_annot_nids([qaid])[0]
daids = np.array(qreq_.daids)
dnids = qreq_.get_qreq_annot_nids(daids)
correct_aids = daids[np.where(dnids == qnid)[0]]
daid = correct_aids[0]
if agg:
for daid in _prog(valid_daids):
Y = qreq_.dinva.get_annot(daid)
item = match_kernel_agg(X, Y, wx_to_weight, alpha, thresh)
shortlist.insert(item)
else:
for daid in _prog(valid_daids):
Y = qreq_.dinva.get_annot(daid)
item = match_kernel_sep(X, Y, wx_to_weight, alpha, thresh)
shortlist.insert(item)
# Build chipmatches for the shortlist results
# with ut.Timer('build cms', verbose=verbose):
cm = chip_match.ChipMatch(qaid=qaid, fsv_col_lbls=['smk'])
cm.daid_list = []
cm.fm_list = []
cm.fsv_list = []
_prog = ut.ProgPartial(lbl='smk build cm qaid=%r' % (qaid, ),
enabled=verbose,
bs=True,
adjust=True)
for item in _prog(shortlist):
(score, score_list, Y, X_idx, Y_idx) = item
X_fxs = ut.take(X.fxs_list, X_idx)
Y_fxs = ut.take(Y.fxs_list, Y_idx)
# Only build matches for those that sver will use
if agg:
X_maws = ut.take(X.maws_list, X_idx)
Y_maws = ut.take(Y.maws_list, Y_idx)
fm, fs = smk_funcs.build_matches_agg(X_fxs, Y_fxs, X_maws,
Y_maws, score_list)
else:
fm, fs = smk_funcs.build_matches_sep(X_fxs, Y_fxs, score_list)
if len(fm) > 0:
# assert not np.any(np.isnan(fs))
daid = Y.aid
fsv = fs[:, None]
cm.daid_list.append(daid)
cm.fm_list.append(fm)
cm.fsv_list.append(fsv)
cm._update_daid_index()
cm.arraycast_self()
cm.score_name_maxcsum(qreq_)
# if False:
# cm.assert_self(qreq_=qreq_, verbose=True)
if sv_on:
cm = pipeline.sver_single_chipmatch(qreq_, cm, verbose=verbose)
cm.score_name_maxcsum(qreq_)
return cm
def sccw_summation(rvecs_list, flags_list, idf_list, maws_list, smk_alpha,
smk_thresh):
r"""
Computes gamma from "To Aggregate or not to aggregate". Every component in
each list is with repsect to a different word.
scc = self consistency criterion
It is a scalar which ensure K(X, X) = 1
Args:
rvecs_list (list of ndarrays): residual vectors for every word
idf_list (list of floats): idf weight for each word
maws_list (list of ndarrays): multi-assign weights for each word for each residual vector
smk_alpha (float): selectivity power
smk_thresh (float): selectivity threshold
Returns:
float: sccw self-consistency-criterion weight
Math:
\begin{equation}
\gamma(X) = (\sum_{c \in \C} w_c M(X_c, X_c))^{-.5}
\end{equation}
Example:
>>> from ibeis.algo.hots.smk.smk_scoring import * # NOQA
>>> from ibeis.algo.hots.smk import smk_scoring
>>> from ibeis.algo.hots.smk import smk_debug
>>> #idf_list, rvecs_list, maws_list, smk_alpha, smk_thresh, wx2_flags = smk_debug.testdata_sccw_sum(db='testdb1')
>>> tup = smk_debug.testdata_sccw_sum(db='PZ_MTEST', nWords=128000)
>>> idf_list, rvecs_list, flags_list, maws_list, smk_alpha, smk_thresh = tup
>>> sccw = smk_scoring.sccw_summation(rvecs_list, flags_list, idf_list, maws_list, smk_alpha, smk_thresh)
>>> print(sccw)
0.0201041835751
CommandLine:
python smk_match.py --db PZ_MOTHERS --nWords 128
Ignore:
0.0384477314197
qmaws_list = dmaws_list = maws_list
drvecs_list = qrvecs_list = rvecs_list
dflags_list = qflags_list = flags_list
flags_list = flags_list[7:10]
maws_list = maws_list[7:10]
idf_list = idf_list[7:10]
rvecs_list = rvecs_list[7:10]
"""
num_rvecs = len(rvecs_list)
if DEBUG_SMK:
assert maws_list is None or len(
maws_list) == num_rvecs, 'inconsistent lengths'
assert num_rvecs == len(idf_list), 'inconsistent lengths'
assert maws_list is None or list(map(len, maws_list)) == list(
map(len, rvecs_list)), 'inconsistent per word lengths'
assert flags_list is None or list(map(len, maws_list)) == list(
map(len, flags_list)), 'inconsistent per word lengths'
assert flags_list is None or len(
flags_list) == num_rvecs, 'inconsistent lengths'
# Indexing with asymetric multi-assignment might get you a non 1 self score?
# List of scores for every word.
scores_list = score_matches(rvecs_list, rvecs_list, flags_list, flags_list,
maws_list, maws_list, smk_alpha, smk_thresh,
idf_list)
if DEBUG_SMK:
assert len(scores_list) == num_rvecs, 'bad rvec and score'
assert len(idf_list) == len(scores_list), 'bad weight and score'
# Summation over all residual vector scores
_count = sum((scores.size for scores in scores_list))
_iter = utool.iflatten(scores.ravel() for scores in scores_list)
self_rawscore = np.fromiter(_iter, np.float64, _count).sum()
# Square root inverse to enforce normalized self-score is 1.0
sccw = np.reciprocal(np.sqrt(self_rawscore))
try:
assert not np.isinf(sccw), 'sccw cannot be infinite'
assert not np.isnan(sccw), 'sccw cannot be nan'
except AssertionError as ex:
utool.printex(ex,
'problem computing self consistency criterion weight',
keys=['num_rvecs'],
iswarning=True)
if num_rvecs > 0:
raise
else:
sccw = 1
return sccw