def group_daids_for_indexing_by_name(ibs, daid_list, num_indexers=8,
verbose=True):
"""
returns groups with only one annotation per name in each group
"""
tup = ibs.group_annots_by_known_names(daid_list)
aidgroup_list, invalid_aids = tup
largest_groupsize = max(map(len, aidgroup_list))
num_bins = min(largest_groupsize, num_indexers)
if verbose or ut.VERYVERBOSE:
print('[mindex] num_indexers = %d ' % (num_indexers,))
print('[mindex] largest_groupsize = %d ' % (largest_groupsize,))
print('[mindex] num_bins = %d ' % (num_bins,))
# Group annotations for indexing according to the split criteria
aids_list, overflow_aids = ut.sample_zip(
aidgroup_list, num_bins, allow_overflow=True, per_bin=1)
if __debug__:
# All groups have the same name
nidgroup_list = ibs.unflat_map(ibs.get_annot_name_rowids, aidgroup_list)
for nidgroup in nidgroup_list:
assert ut.allsame(nidgroup), 'bad name grouping'
if __debug__:
# All subsiquent indexer are subsets (in name/identity space)
# of the previous
nids_list = ibs.unflat_map(ibs.get_annot_name_rowids, aids_list)
prev_ = None
for nids in nids_list:
if prev_ is None:
prev_ = set(nids)
else:
assert prev_.issuperset(nids), 'bad indexer grouping'
return aids_list, overflow_aids, num_bins
def group_daids_for_indexing_by_name(ibs, daid_list, num_indexers=8,
verbose=True):
"""
returns groups with only one annotation per name in each group
"""
tup = ibs.group_annots_by_known_names(daid_list)
aidgroup_list, invalid_aids = tup
largest_groupsize = max(map(len, aidgroup_list))
num_bins = min(largest_groupsize, num_indexers)
if verbose or ut.VERYVERBOSE:
print('[mindex] num_indexers = %d ' % (num_indexers,))
print('[mindex] largest_groupsize = %d ' % (largest_groupsize,))
print('[mindex] num_bins = %d ' % (num_bins,))
# Group annotations for indexing according to the split criteria
aids_list, overflow_aids = ut.sample_zip(
aidgroup_list, num_bins, allow_overflow=True, per_bin=1)
if __debug__:
# All groups have the same name
nidgroup_list = ibs.unflat_map(ibs.get_annot_name_rowids, aidgroup_list)
for nidgroup in nidgroup_list:
assert ut.allsame(nidgroup), 'bad name grouping'
if __debug__:
# All subsiquent indexer are subsets (in name/identity space)
# of the previous
nids_list = ibs.unflat_map(ibs.get_annot_name_rowids, aids_list)
prev_ = None
for nids in nids_list:
if prev_ is None:
prev_ = set(nids)
else:
assert prev_.issuperset(nids), 'bad indexer grouping'
return aids_list, overflow_aids, num_bins
def __init__(split_index, ibs, daid_list, num_forests=8):
print('[nnsindex] make HOTSMultiIndex over %d annots' % (len(daid_list),))
# Remove unknown names
aid_list = daid_list
known_aids_list, unknown_aids = ibsfuncs.group_annots_by_known_names(ibs, aid_list)
num_bins = min(max(map(len, known_aids_list)), num_forests)
# Put one name per forest
forest_aids, overflow_aids = utool.sample_zip(
known_aids_list, num_bins, allow_overflow=True, per_bin=1)
forest_indexes = []
extra_indexes = []
for tx, aids in enumerate(forest_aids):
print('[nnsindex] building forest %d/%d with %d aids' %
(tx + 1, num_bins, len(aids)))
if len(aids) > 0:
hsindex = HOTSIndex(ibs, aids)
forest_indexes.append(hsindex)
if len(overflow_aids) > 0:
print('[nnsindex] building overflow forest')
overflow_index = HOTSIndex(ibs, overflow_aids)
extra_indexes.append(overflow_index)
if len(unknown_aids) > 0:
print('[nnsindex] building unknown forest')
unknown_index = HOTSIndex(ibs, unknown_aids)
extra_indexes.append(unknown_index)
#print('[nnsindex] building normalizer forest') # TODO
split_index.forest_indexes = forest_indexes
split_index.extra_indexes = extra_indexes
def __init__(split_index, ibs, daid_list, num_forests=8):
print('[nnsindex] make NNSplitIndex over %d annots' % (len(daid_list),))
aid_list = daid_list
nid_list = ibs.get_annot_nids(aid_list)
#flag_list = ibs.get_annot_exemplar_flag(aid_list)
nid2_aids = utool.group_items(aid_list, nid_list)
key_list = nid2_aids.keys()
aids_list = nid2_aids.values()
isunknown_list = ibs.is_nid_unknown(key_list)
known_aids = utool.filterfalse_items(aids_list, isunknown_list)
uknown_aids = utool.flatten(utool.filter_items(aids_list, isunknown_list))
num_forests_ = min(max(map(len, aids_list)), num_forests)
# Put one name per forest
forest_aids, overflow_aids = utool.sample_zip(known_aids, num_forests_,
allow_overflow=True,
per_bin=1)
forest_indexes = []
extra_indexes = []
for tx, aids in enumerate(forest_aids):
print('[nnsindex] building forest %d/%d with %d aids' % (tx + 1, num_forests_, len(aids)))
if len(aids) > 0:
nn_index = NNIndex(ibs, aids)
forest_indexes.append(nn_index)
if len(overflow_aids) > 0:
print('[nnsindex] building overflow forest')
overflow_index = NNIndex(ibs, overflow_aids)
extra_indexes.append(overflow_index)
if len(uknown_aids) > 0:
print('[nnsindex] building unknown forest')
unknown_index = NNIndex(ibs, uknown_aids)
extra_indexes.append(unknown_index)
#print('[nnsindex] building normalizer forest') # TODO
split_index.forest_indexes = forest_indexes
split_index.extra_indexes = extra_indexes