def index_data_annots(annots_df,
daids,
words,
with_internals=True,
aggregate=False,
alpha=3,
thresh=0,
with_pandas=WITH_PANDAS):
"""
Builds the initial inverted index from a dataframe, daids, and words.
Optionally builds the internals of the inverted structure
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs, annots_df, daids, qaids, words = smk_debug.testdata_words()
>>> with_internals = False
>>> invindex = index_data_annots(annots_df, daids, words, with_internals)
#>>> print(utool.hashstr(repr(list(invindex.__dict__.values()))))
#v8+i5i8+55j0swio
"""
if utool.VERBOSE:
print('[smk_index] index_data_annots')
flann_params = {}
_words = pdh.ensure_values(words)
wordflann = nntool.flann_cache(_words,
flann_params=flann_params,
appname='smk')
_daids = pdh.ensure_values(daids)
_vecs_list = pdh.ensure_2d_values(annots_df['vecs'][_daids])
_idx2_dvec, _idx2_daid, _idx2_dfx = nntool.invertable_stack(
_vecs_list, _daids)
# Pandasify
if with_pandas:
idx_series = pdh.IntIndex(np.arange(len(_idx2_daid)), name='idx')
idx2_dfx = pdh.IntSeries(_idx2_dfx, index=idx_series, name='fx')
idx2_daid = pdh.IntSeries(_idx2_daid, index=idx_series, name='aid')
idx2_dvec = pd.DataFrame(_idx2_dvec,
index=idx_series,
columns=VEC_COLUMNS)
else:
idx2_dfx = _idx2_dfx
idx2_daid = _idx2_daid
idx2_dvec = _idx2_dvec
pass
invindex = InvertedIndex(words, wordflann, idx2_dvec, idx2_daid, idx2_dfx,
daids)
if with_internals:
compute_data_internals_(invindex, aggregate, alpha, thresh) # 99%
return invindex
def make_annot_df(ibs):
"""
Creates a panda dataframe using an ibeis controller
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs = smk_debug.testdata_ibeis()
>>> annots_df = make_annot_df(ibs)
>>> print(utool.hashstr(repr(annots_df.values)))
j12n+x93m4c!4un3
#>>> from ibeis.model.hots.smk import smk_debug
#>>> smk_debug.rrr()
#>>> smk_debug.check_dtype(annots_df)
"""
aid_list = ibs.get_valid_aids() # 80us
kpts_list = ibs.get_annot_kpts(aid_list) # 40ms
vecs_list = ibs.get_annot_desc(aid_list) # 50ms
aid_series = pdh.IntSeries(np.array(aid_list, dtype=INTEGER_TYPE),
name='aid')
kpts_df = pdh.pandasify_list2d(kpts_list, aid_series, KPT_COLUMNS, 'fx',
'kpts') # 6.7ms
vecs_df = pdh.pandasify_list2d(vecs_list, aid_series, VEC_COLUMNS, 'fx',
'vecs') # 7.1ms
# Pandas Annotation Dataframe
annots_df = pd.concat([kpts_df, vecs_df], axis=1) # 845 us
return annots_df
def compute_word_idf_(wx_series,
wx2_idxs,
idx2_aid,
daids,
with_pandas=WITH_PANDAS):
"""
Returns the inverse-document-frequency weighting for each word
internals step 2
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs, annots_df, daids, qaids, invindex, wx2_idxs = smk_debug.testdata_raw_internals1()
>>> wx_series = invindex.words.index
>>> idx2_aid = invindex.idx2_daid
>>> wx2_idf = compute_word_idf_(wx_series, wx2_idxs, idx2_aid, daids)
>>> print(wx2_idf.shape)
(8000,)
#>>> wx2_idxs = invindex.wx2_idxs
"""
if utool.VERBOSE:
mark, end_ = utool.log_progress('[smk_index] Word IDFs: ',
len(wx_series),
flushfreq=500,
writefreq=50)
mark(0)
wx_series_values = pdh.ensure_values(wx_series)
idx2_aid_values = pdh.ensure_values(idx2_aid)
wx2_idxs_values = pdh.ensure_values_subset(wx2_idxs, wx_series_values)
#with utool.Timer('method 1'): # 0.16s
idxs_list = [
pdh.ensure_values(idxs).astype(INDEX_TYPE) for idxs in wx2_idxs_values
] # 11%
aids_list = [
idx2_aid_values.take(idxs) if len(idxs) > 0 else []
for idxs in idxs_list
]
nTotalDocs = len(daids)
nDocsWithWord_list = [len(set(aids)) for aids in aids_list] # 68%
# compute idf half of tf-idf weighting
idf_list = [
np.log(nTotalDocs /
nDocsWithWord).astype(FLOAT_TYPE) if nDocsWithWord > 0 else 0.0
for nDocsWithWord in nDocsWithWord_list
] # 17.8 ms # 13%
if utool.VERBOSE:
end_()
if with_pandas:
wx2_idf = pdh.IntSeries(idf_list, index=wx_series, name='idf')
else:
wx2_idf = dict(zip(wx_series_values, idf_list))
return wx2_idf
def compute_data_gamma_(idx2_daid,
wx2_rvecs,
wx2_aids,
wx2_weight,
alpha=3,
thresh=0):
"""
Internals step4
Computes gamma normalization scalar for the database annotations
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs, annots_df, invindex, wx2_idxs, wx2_idf, wx2_rvecs, wx2_aids = smk_debug.testdata_raw_internals2()
>>> alpha = ibs.cfg.query_cfg.smk_cfg.alpha
>>> thresh = ibs.cfg.query_cfg.smk_cfg.thresh
>>> idx2_daid = invindex.idx2_daid
>>> wx2_weight = wx2_idf
>>> daids = invindex.daids
>>> use_cache = USE_CACHE_GAMMA and False
>>> daid2_gamma = compute_data_gamma_(idx2_daid, wx2_rvecs, wx2_aids, wx2_weight, daids, use_cache=use_cache)
"""
# Gropuing by aid and words
wx_sublist = pdh.ensure_values(pdh.ensure_index(wx2_rvecs))
if utool.VERBOSE:
print('[smk_index] Compute Gamma alpha=%r, thresh=%r: ' %
(alpha, thresh))
mark1, end1_ = utool.log_progress('[smk_index] Gamma Group: ',
len(wx_sublist),
flushfreq=100,
writefreq=50)
rvecs_list1 = pdh.ensure_values_subset(wx2_rvecs, wx_sublist)
aids_list = pdh.ensure_values_subset(wx2_aids, wx_sublist)
daid2_wx2_drvecs = utool.ddict(lambda: utool.ddict(list))
# Group by daids first and then by word index
for wx, aids, rvecs in zip(wx_sublist, aids_list, rvecs_list1):
group_aids, groupxs = smk_speed.group_indicies(aids)
rvecs_group = smk_speed.apply_grouping(rvecs, groupxs) # 2.9 ms
for aid, rvecs_ in zip(group_aids, rvecs_group):
daid2_wx2_drvecs[aid][wx] = rvecs_
if utool.VERBOSE:
end1_()
# For every daid, compute its gamma using pregrouped rvecs
# Summation over words for each aid
if utool.VERBOSE:
mark2, end2_ = utool.log_progress('[smk_index] Gamma Sum: ',
len(daid2_wx2_drvecs),
flushfreq=100,
writefreq=25)
aid_list = list(daid2_wx2_drvecs.keys())
wx2_aidrvecs_list = list(daid2_wx2_drvecs.values())
aidwxs_list = [
list(wx2_aidrvecs.keys()) for wx2_aidrvecs in wx2_aidrvecs_list
]
aidrvecs_list = [
list(wx2_aidrvecs.values()) for wx2_aidrvecs in wx2_aidrvecs_list
]
aidweight_list = [[wx2_weight[wx] for wx in aidwxs]
for aidwxs in aidwxs_list]
#gamma_list = []
#for weight_list, rvecs_list in zip(aidweight_list, aidrvecs_list):
# assert len(weight_list) == len(rvecs_list), 'one list for each word'
# gamma = smk_core.gamma_summation2(rvecs_list, weight_list, alpha, thresh) # 66.8 %
# #weight_list = np.ones(weight_list.size)
# gamma_list.append(gamma)
gamma_list = [
smk_core.gamma_summation2(rvecs_list, weight_list, alpha, thresh)
for weight_list, rvecs_list in zip(aidweight_list, aidrvecs_list)
]
daid2_gamma = pdh.IntSeries(gamma_list, index=aid_list, name='gamma')
if utool.VERBOSE:
end2_()
return daid2_gamma
def assign_to_words_(wordflann,
words,
idx2_vec,
idx_name='idx',
dense=True,
nAssign=1,
with_pandas=WITH_PANDAS):
"""
Assigns descriptor-vectors to nearest word. Returns forward and inverted index.
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs, annots_df, daids, qaids, invindex = smk_debug.testdata_raw_internals0()
>>> words = invindex.words
>>> wordflann = invindex.wordflann
>>> idx2_vec = invindex.idx2_dvec
>>> dense = True
>>> nAssign = ibs.cfg.query_cfg.smk_cfg.nAssign
>>> idx_name, series_name = 'idx', 'wx2_idxs'
>>> _dbargs = (wordflann, words, idx2_vec, idx_name, dense, nAssign)
>>> wx2_idxs, idx2_wx = assign_to_words_(*_dbargs)
"""
idx2_vec_values = pdh.ensure_values(idx2_vec)
# Find each vectors nearest word
#TODO: multiple assignment
_idx2_wx, _idx2_wdist = wordflann.nn_index(idx2_vec_values, nAssign)
if nAssign > 1:
#((words[_idx2_wx[:,0]].astype(np.float64) - idx2_vec_values.astype(np.float64)) ** 2).sum(axis=0)
#_idx2_wdist[:,0]
#np.sqrt(((words[_idx2_wx[:,0]].astype(np.float64) - idx2_vec_values.astype(np.float64)) ** 2).sum(axis=0))
# mutli assignment filtering as in
# http://lear.inrialpes.fr/pubs/2010/JDS10a/jegou_improvingbof_preprint.pdf
alpha = 1.2
thresh = alpha * _idx2_wdist.T[0:1].T
invalid = _idx2_wdist >= thresh
# Weighting as in Lost in Quantization
sigma = 80
unnorm_weight = np.exp(
np.divide(-_idx2_wdist.astype(np.float64), 2 * (sigma**2)))
masked_weight = np.ma.masked_array(unnorm_weight, mask=invalid)
weight = masked_weight / masked_weight.sum(axis=1)[:, np.newaxis]
masked_wxs = np.ma.masked_array(_idx2_wx, mask=invalid)
idx2_wxs = map(utool.filter_Nones, masked_wxs.tolist())
idx2_wx_weights = map(utool.filter_Nones, weight.tolist())
#masked_weight1 = np.ma.masked_array(_idx2_wdist, mask=invalid)
#weight1 = masked_weight1 / masked_weight1.sum(axis=1)[:, np.newaxis]
# multiple assignment weight: expt(-(d ** 2) / (2 * sigma ** 2))
# The distance d_0 is used to filter asignments with distance less than
# alpha * d_0 where alpha = 1.2
PANDAS_GROUP = True or with_pandas
# Compute inverted index
if PANDAS_GROUP:
# Pandas grouping seems to be faster in this instance
word_assignments = pd.DataFrame(_idx2_wx, columns=['wx']) # 141 us
word_group = word_assignments.groupby('wx') # 34.5 us
_wx2_idxs = word_group['wx'].indices # 8.6 us
else:
idx2_idx = np.arange(len(idx2_vec))
wx_list, groupxs = smk_speed.group_indicies(_idx2_wx) # 5.52 ms
idxs_list = smk_speed.apply_grouping(idx2_idx, groupxs) # 2.9 ms
_wx2_idxs = dict(zip(wx_list, idxs_list)) # 753 us
#
if with_pandas:
idx_series = pdh.ensure_index(idx2_vec)
wx_series = pdh.ensure_index(words)
wx2_idxs = pdh.pandasify_dict1d(_wx2_idxs,
wx_series,
idx_name, ('wx2_' + idx_name + 's'),
dense=dense) # 274 ms 97.4 %
idx2_wx = pdh.IntSeries(_idx2_wx, index=idx_series, name='wx')
else:
if dense:
wx2_idxs = {
wx: _wx2_idxs[wx].astype(INDEX_TYPE)
if wx in _wx2_idxs else np.empty(0, dtype=INDEX_TYPE)
for wx in range(len(words))
}
#wx2_idxs = _wx2_idxs
#for wx in range(len(words)):
# if wx not in wx2_idxs:
# wx2_idxs[wx] = np.empty(0, dtype=INDEX_TYPE)
else:
wx2_idxs = _wx2_idxs
idx2_wx = _idx2_wx
return wx2_idxs, idx2_wx
def compute_data_gamma_(idx2_daid,
wx2_rvecs,
wx2_aids,
wx2_idf,
alpha=3,
thresh=0):
"""
Computes gamma normalization scalar for the database annotations
Internals step4
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs, annots_df, invindex, wx2_idxs, wx2_idf, wx2_rvecs, wx2_aids = smk_debug.testdata_raw_internals2()
>>> alpha = ibs.cfg.query_cfg.smk_cfg.alpha
>>> thresh = ibs.cfg.query_cfg.smk_cfg.thresh
>>> idx2_daid = invindex.idx2_daid
>>> wx2_idf = wx2_idf
>>> daids = invindex.daids
>>> use_cache = USE_CACHE_GAMMA and False
>>> daid2_gamma = compute_data_gamma_(idx2_daid, wx2_rvecs, wx2_aids, wx2_idf, daids, use_cache=use_cache)
"""
if utool.DEBUG2:
from ibeis.model.hots.smk import smk_debug
smk_debug.rrr()
smk_debug.check_wx2(wx2_rvecs=wx2_rvecs, wx2_aids=wx2_aids)
wx_sublist = pdh.ensure_values(pdh.ensure_index(wx2_rvecs))
if utool.VERBOSE:
print('[smk_index] Compute Gamma alpha=%r, thresh=%r: ' %
(alpha, thresh))
mark1, end1_ = utool.log_progress(
'[smk_index] Gamma group (by word): ',
len(wx_sublist),
flushfreq=100,
writefreq=50,
with_totaltime=True)
# Get list of aids and rvecs w.r.t. words
aids_list = pdh.ensure_values_subset(wx2_aids, wx_sublist)
rvecs_list1 = pdh.ensure_values_subset(wx2_rvecs, wx_sublist)
# Group by daids first and then by word index
daid2_wx2_drvecs = utool.ddict(lambda: utool.ddict(list))
for wx, aids, rvecs in zip(wx_sublist, aids_list, rvecs_list1):
group_aids, groupxs = clustertool.group_indicies(aids)
rvecs_group = clustertool.apply_grouping(rvecs, groupxs) # 2.9 ms
for aid, rvecs_ in zip(group_aids, rvecs_group):
daid2_wx2_drvecs[aid][wx] = rvecs_
if utool.VERBOSE:
end1_()
# For every daid, compute its gamma using pregrouped rvecs
# Summation over words for each aid
if utool.VERBOSE:
mark2, end2_ = utool.log_progress(
'[smk_index] Gamma Sum (over daid): ',
len(daid2_wx2_drvecs),
flushfreq=100,
writefreq=25,
with_totaltime=True)
# Get lists w.r.t daids
aid_list = list(daid2_wx2_drvecs.keys())
# list of mappings from words to rvecs foreach daid
# [wx2_aidrvecs_1, ..., wx2_aidrvecs_nDaids,]
_wx2_aidrvecs_list = list(daid2_wx2_drvecs.values())
_aidwxs_iter = (list(wx2_aidrvecs.keys())
for wx2_aidrvecs in _wx2_aidrvecs_list)
aidrvecs_list = [
list(wx2_aidrvecs.values()) for wx2_aidrvecs in _wx2_aidrvecs_list
]
aididf_list = [[wx2_idf[wx] for wx in aidwxs] for aidwxs in _aidwxs_iter]
#gamma_list = []
if utool.DEBUG2:
try:
for count, (idf_list, rvecs_list) in enumerate(
zip(aididf_list, aidrvecs_list)):
assert len(idf_list) == len(
rvecs_list), 'one list for each word'
#gamma = smk_core.gamma_summation2(rvecs_list, idf_list, alpha, thresh)
except Exception as ex:
utool.printex(ex)
utool.embed()
raise
gamma_list = [
smk_core.gamma_summation2(rvecs_list, idf_list, alpha, thresh)
for idf_list, rvecs_list in zip(aididf_list, aidrvecs_list)
]
if WITH_PANDAS:
daid2_gamma = pdh.IntSeries(gamma_list, index=aid_list, name='gamma')
else:
daid2_gamma = dict(zip(aid_list, gamma_list))
if utool.VERBOSE:
end2_()
return daid2_gamma
def assign_to_words_(wordflann,
words,
idx2_vec,
idx_name='idx',
dense=True,
nAssign=1,
massign_alpha=1.2,
massign_sigma=80):
"""
Assigns descriptor-vectors to nearest word.
Returns inverted index, multi-assigned weights, and forward index
wx2_idxs - word index -> vector indexes
wx2_maws - word index -> multi-assignment weights
idf2_wxs - vector index -> assigned word indexes
>>> from ibeis.model.hots.smk.smk_index import * # NOQA
>>> from ibeis.model.hots.smk import smk_debug
>>> ibs, annots_df, daids, qaids, invindex = smk_debug.testdata_raw_internals0()
>>> words = invindex.words
>>> wordflann = invindex.wordflann
>>> idx2_vec = invindex.idx2_dvec
>>> dense = True
>>> nAssign = ibs.cfg.query_cfg.smk_cfg.nAssign
>>> _dbargs = (wordflann, words, idx2_vec, idx_name, dense, nAssign)
>>> wx2_idxs, wx2_maws, idx2_wxs = assign_to_words_(*_dbargs)
"""
idx2_vec_values = pdh.ensure_values(idx2_vec)
# Assign each vector to the nearest visual words
_idx2_wx, _idx2_wdist = wordflann.nn_index(idx2_vec_values, nAssign)
if nAssign > 1:
# MultiAssignment Filtering from Improving Bag of Features
# http://lear.inrialpes.fr/pubs/2010/JDS10a/jegou_improvingbof_preprint.pdf
thresh = np.multiply(massign_alpha, _idx2_wdist.T[0:1].T)
invalid = np.greater_equal(_idx2_wdist, thresh)
# Weighting as in Lost in Quantization
gauss_numer = -_idx2_wdist.astype(np.float64)
gauss_denom = 2 * (massign_sigma**2)
gauss_exp = np.divide(gauss_numer, gauss_denom)
unnorm_maw = np.exp(gauss_exp)
# Mask invalid multiassignment weights
masked_unorm_maw = np.ma.masked_array(unnorm_maw, mask=invalid)
# Normalize multiassignment weights from 0 to 1
masked_norm = masked_unorm_maw.sum(axis=1)[:, np.newaxis]
masked_maw = np.divide(masked_unorm_maw, masked_norm)
masked_wxs = np.ma.masked_array(_idx2_wx, mask=invalid)
# Remove masked weights and word indexes
idx2_wxs = list(map(utool.filter_Nones, masked_wxs.tolist()))
idx2_maws = list(map(utool.filter_Nones, masked_maw.tolist()))
else:
idx2_wxs = _idx2_wx.tolist()
idx2_maws = [1.0] * len(idx2_wxs)
# Invert mapping -- Group by word indexes
jagged_idxs = ([idx] * len(wxs) for idx, wxs in enumerate(idx2_wxs))
wx_keys, groupxs = clustertool.jagged_group(idx2_wxs)
idxs_list = clustertool.apply_jagged_grouping(jagged_idxs, groupxs)
maws_list = clustertool.apply_jagged_grouping(idx2_maws, groupxs)
wx2_idxs = dict(zip(wx_keys, idxs_list))
wx2_maws = dict(zip(wx_keys, maws_list))
if WITH_PANDAS:
idx_series = pdh.ensure_index(idx2_vec)
wx_series = pdh.ensure_index(words)
wx2_idxs = pdh.pandasify_dict1d(wx2_idxs,
wx_series,
idx_name, ('wx2_' + idx_name + 's'),
dense=dense)
idx2_wxs = pdh.IntSeries(idx2_wxs, index=idx_series, name='wx')
return wx2_idxs, wx2_maws, idx2_wxs
