def uscsdist(proto, cs='', usermvis=False):
"""
Return the relative cosine distance to ensure and quinine given a vector of activity for a particular event
:param proto: dict of prototypical vectors from protovectors (based on GLM)
:param event: max or mean around replay or stimulus, to be compared to protovector
:param cs: required if usermvis is set to true
:param usermvis: use the visual-subtracted glm vector if available if true
:return: ensure-quinine distance where values > 0 are closer to ensure, values < 0 closer to quinine
"""
if not usermvis:
if cs == 'plus':
return cosinedist(proto['ensure'], proto['plus'])
elif cs == 'minus':
return cosinedist(proto['quinine'], proto['minus'])
else:
fitfun = lambda vs, x: vs[0] * x
errfun = lambda vs, x, y: fitfun(vs, x) - y
if cs == 'plus':
[vscalc,
success] = optimize.leastsq(errfun, [0.5],
args=(proto['ensure'], proto['plus']))
return abs(vscalc[0])
elif cs == 'minus':
[vscalc, success] = optimize.leastsq(errfun, [0.5],
args=(proto['quinine'],
proto['minus']))
return abs(vscalc[0])
return 1
def vectorcosineworker(headwords, workpiledict, resultdict):
"""
itemfromworkpile = (avalues, lemmavalues, lemmalength)
:return:
"""
while headwords:
try:
headword = headwords.pop()
except IndexError:
headword = None
if headword:
item = workpiledict[headword]
avalues = item[0]
lemmavalues = item[1]
lemmalength = item[2]
try:
cv = cosinedist(avalues, lemmavalues)
except NameError:
# scipy not available
cv = findcosinedist(avalues, lemmavalues, lemmalength)
resultdict[headword] = cv
return resultdict
def qdist(proto, event, cs='', usermvis=True):
"""
Return the cosine distance to quinine given a vector of activity for a particular event
:param proto: dict of prototypical vectors from protovectors (based on GLM)
:param event: max or mean around replay or stimulus, to be compared to protovector
:param cs: required if usermvis is set to true
:param usermvis: use the visual-subtracted glm vector if available if true
:return: ensure-quinine distance where values > 0 are closer to ensure, values < 0 closer to quinine
"""
unitvec = event / np.nansum(np.abs(event))
ensdist = np.nan
if usermvis and '%s-quinine' % cs in proto:
ensdist = cosinedist(proto['%s-quinine' % cs], unitvec)
elif not usermvis and 'quinine' in proto:
ensdist = cosinedist(proto['quinine'], unitvec)
return ensdist
def caclulatecosinevalues(focusword, vectorspace, headwords):
"""
cos(α,β) = α · β / ||α|| ||β||
||α|| = sqrt(α · α)
α = sum([a1, a2, ... ai])
returns a dict:
{'facultas': 0.7418011102528389, 'frater': 0.7418011102528389, 'superus': 0.6535898384862247, ...}
:param focusword:
:param vectorspace:
:param headwords:
:return:
"""
# lengths = dict()
# for w in headwords:
# vals = list()
# for key in vectorspace.keys():
# vals.append(vectorspace[key][w])
#
# lengths[w] = findvectorlength(vals)
numberedsentences = vectorspace.keys()
lemmavalues = list()
for num in numberedsentences:
try:
lemmavalues.append(vectorspace[num][focusword])
except KeyError:
# print('KeyError in caclulatecosinevalues()')
# we know that the word appears, but it might have been there more than once...
# we just lost that information when the KeyError bit
lemmavalues.append(1)
try:
cosinedist
lvl = None
except NameError:
# we did not import numpy/scipy
lvl = findvectorlength(lemmavalues)
cosinevals = dict()
for w in headwords:
avalues = list()
for num in numberedsentences:
avalues.append(vectorspace[num][w])
# scipy will choke if you send it div-by-zero data
# RuntimeWarning: invalid value encountered in true_divide dist = 1.0 - np.dot(u, v) / (norm(u) * norm(v))
# be careful not to end up with 0 in lemmavalues with 'voluptas' becuase of the 'v-for-u' issue
# print('w/av/bv', sum(avalues), sum(lemmavalues), '({w})'.format(w=w))
try:
cosinevals[w] = cosinedist(avalues, lemmavalues)
except NameError:
# we did not import numpy/scipy
cosinevals[w] = findcosinedist(avalues, lemmavalues, lvl)
return cosinevals