def get_yelp_blend():
cnettensor = get_picklecached_thing('cnet.pickle',
conceptnet_with_custom_identities, 'ConceptNet')
# To regenerate these next two, run yelp_matrix_creation.py
yelpwordstensor = get_picklecached_thing('yelp-tfidf.pickle', None, 'Yelp TFIDF')
yelpcatstensor = get_picklecached_thing('yelp_cats_and_price.pickle', None, 'Yelp Cats')
print yelpwordstensor
print yelpcatstensor
return run_3blend_simple(cnettensor, yelpwordstensor, yelpcatstensor, None,
0.03)
def get_yelp_blend():
cnettensor = get_picklecached_thing('cnet.pickle',
conceptnet_with_custom_identities,
'ConceptNet')
# To regenerate these next two, run yelp_matrix_creation.py
yelpwordstensor = get_picklecached_thing('yelp-tfidf.pickle', None,
'Yelp TFIDF')
yelpcatstensor = get_picklecached_thing('yelp_cats_and_price.pickle', None,
'Yelp Cats')
print yelpwordstensor
print yelpcatstensor
return run_3blend_simple(cnettensor, yelpwordstensor, yelpcatstensor, None,
0.03)
def example():
# In the context of this corpus of healthcare-related messages (not
# publicly provided), what do people want?
health = get_picklecached_thing('health.pickle.gz', None)
svd = run_cnet_blend(health, 'health_blend')
svd.summarize(10)
# Make a labeled vector of similarities to the feature person\Desires.
vec = predict_concepts(svd, ('left', 'Desires', 'person'))
# Get the list of concepts in ConceptNet.
concepts = cnet.label_list(0)
# Look for concepts that *aren't* in ConceptNet and are ranked highly
# in the vector of results.
for item in vec.top_items(5000):
# skip over stopwords
contentful = False
for word in item[0].split(' '):
if not en.nl.is_stopword(word):
contentful = True
break
if contentful and item[0] not in concepts:
# found one. print it along with its score.
print item[0], item[1]
cnettensor = get_picklecached_thing('cnet.pickle',
conceptnet_with_custom_identities,
'ConceptNet')
# To regenerate these next two, run yelp_matrix_creation.py
yelpwordstensor = get_picklecached_thing('yelp-tfidf.pickle', None,
'Yelp TFIDF')
yelpcatstensor = get_picklecached_thing('yelp_cats_and_price.pickle', None,
'Yelp Cats')
print yelpwordstensor
print yelpcatstensor
return run_3blend_simple(cnettensor, yelpwordstensor, yelpcatstensor, None,
0.03)
if __name__ == '__main__':
svd = get_picklecached_thing('yelp-cnet-blend.pickle', get_yelp_blend)
export_svdview(svd, "/csc/svdview/data/yelp-tweek.tsv")
if False:
###
### Old blending stuff
###
def find_factor_from_SVD(t1, t2):
sigma1 = t1.svals[0:10]
a = sigma1[0]
sigma2 = t2.svals[0:10]
b = sigma2[0]
return float(a / (a + b))
def normalize_and_copy_three(tensor):
def get_thing(fname, func=None):
return get_picklecached_thing(local_file(fname), func)
def _get_size_blend():
sizes = get_picklecached_thing(FILEPATH+os.sep+'sizematrix.pickle.gz', _make_size_matrix)
cnet = get_picklecached_thing(FILEPATH+os.sep+'cnet.pickle.gz', lambda: conceptnet_2d_from_db('en'))
size_blend = Blend([sizes, cnet]).normalized(mode=[0,1]).bake()
return size_blend
def _make_size_matrix():
matrixlist = []
sizefile = open("size_similarities.physnet")
for line in sizefile:
l = eval(line)
matrixlist.append(((l[0],l[1][1],l[1][2]),40))
return ConceptByFeatureMatrix.from_triples(matrixlist)
def _get_size_blend():
sizes = get_picklecached_thing(FILEPATH+os.sep+'sizematrix.pickle.gz', _make_size_matrix)
cnet = get_picklecached_thing(FILEPATH+os.sep+'cnet.pickle.gz', lambda: conceptnet_2d_from_db('en'))
size_blend = Blend([sizes, cnet]).normalized(mode=[0,1]).bake()
return size_blend
sizeblend = get_picklecached_thing(FILEPATH+os.sep+'sizeblend.pickle.gz', _get_size_blend)
sizesvd = sizeblend.svd(k=100)
#cnet_norm = conceptnet_2d_from_db('en').normalized()
#rawsvd = cnet_norm.svd()
def get_similar_size_examples(concept):
if not concept in sizeblend:
return None
concept_vector = sizesvd.weighted_u[concept,:].hat()
like_concepts = sizesvd.u_angles_to(concept_vector)
return like_concepts
def get_similar_raw_examples(concept):
concept_vector = rawsvd.weighted_u[concept,:].hat()
like_concepts = rawsvd.u_angles_to(concept_vector)
def _get_color_blend():
colors = get_picklecached_thing(FILEPATH+os.sep+'colormatrix.pickle.gz', _make_color_matrix)
cnet = get_picklecached_thing(FILEPATH+os.sep+'cnet.pickle.gz', lambda: conceptnet_2d_from_db('en'))
colorblend = Blend([colors, cnet]).normalized(mode=[0,1]).bake()
return colorblend
if word == '': continue
print color, word
matrixlist.append(((word, 'HasColor', color), 10))
matrixlist.append(((word, 'HasProperty', 'colorful'), 10))
matrixlist.append(((word, 'HasProperty', color), 10))
matrixlist.append(((color, 'HasColor', color), 50))
matrixlist.append(((color, 'HasProperty', color), 50))
return ConceptByFeatureMatrix.from_triples(matrixlist)
def _get_color_blend():
colors = get_picklecached_thing(FILEPATH+os.sep+'colormatrix.pickle.gz', _make_color_matrix)
cnet = get_picklecached_thing(FILEPATH+os.sep+'cnet.pickle.gz', lambda: conceptnet_2d_from_db('en'))
colorblend = Blend([colors, cnet]).normalized(mode=[0,1]).bake()
return colorblend
colorblend = get_picklecached_thing(FILEPATH+os.sep+'colorblend.pickle.gz', _get_color_blend)
thesvd = colorblend.svd(k=100)
colorful_concepts = thesvd.u.label_list(0)
#print thesvd.summarize(10)
colorful_vec = thesvd.v[('right', u'HasProperty', u'colorful'), :]
colorlist = ['blue', 'black', 'brown', 'green', 'grey', 'orange', 'pink', 'purple', 'red', 'white', 'yellow']
rgb = {'blue': (0,0,255), 'black': (0,0,0), 'brown': (139, 69, 19), 'green': (0, 255, 0), 'grey': (100,100,100), 'orange': (255, 165,0), 'pink': (255,105,180), 'purple': (160, 32, 240), 'red': (255,0,0), 'white': (255, 255, 255), 'yellow': (255,255,0)}
#colorvecs = [(x, thesvd.weighted_u[x,:]) for x in colorlist]
colorvecs = [(x, thesvd.weighted_u[x,:]*.1 + thesvd.weighted_v[('right', 'HasColor', x),:]) for x in colorlist]
def how_colorful(word, thesvd):
wordvc = thesvd.weighted_u[word,:]
return wordvc.hat() * make_category(thesvd, concepts=rgb.keys())
def _get_color_mix(adhoc, thesvd):
vec_dict = {}
cnet[key] = 0
return cnet
def get_yelp_blend():
cnettensor = get_picklecached_thing('cnet.pickle',
conceptnet_with_custom_identities, 'ConceptNet')
# To regenerate these next two, run yelp_matrix_creation.py
yelpwordstensor = get_picklecached_thing('yelp-tfidf.pickle', None, 'Yelp TFIDF')
yelpcatstensor = get_picklecached_thing('yelp_cats_and_price.pickle', None, 'Yelp Cats')
print yelpwordstensor
print yelpcatstensor
return run_3blend_simple(cnettensor, yelpwordstensor, yelpcatstensor, None,
0.03)
if __name__ == '__main__':
svd = get_picklecached_thing('yelp-cnet-blend.pickle', get_yelp_blend)
export_svdview(svd, "/csc/svdview/data/yelp-tweek.tsv")
if False:
###
### Old blending stuff
###
def find_factor_from_SVD(t1, t2):
sigma1 = t1.svals[0:10]
a = sigma1[0]
sigma2 = t2.svals[0:10]
b = sigma2[0]
def get_thing(fname, func=None):
return get_picklecached_thing(local_file(fname), func)
def run_cnet_blend(other, FILENAME='blend'):
blend = get_picklecached_thing(FILENAME+'.pickle.gz',
lambda: make_blend(other))
svd = blend.svd()
write_packed(svd.u, FILENAME, unstem=lambda x: x)
return svd
from csc.conceptnet4.models import *
from csc.conceptnet4.analogyspace import *
from csc.divisi.util import get_picklecached_thing
from csc.divisi.blend import Blend
from csc.divisi.export_svdview import write_packed
cnet = get_picklecached_thing('cnet.pickle.gz', lambda: conceptnet_2d_from_db('en'))
def make_blend(other):
return Blend([cnet, other])
def run_cnet_blend(other, FILENAME='blend'):
blend = get_picklecached_thing(FILENAME+'.pickle.gz',
lambda: make_blend(other))
svd = blend.svd()
write_packed(svd.u, FILENAME, unstem=lambda x: x)
return svd
def example():
# In the context of this corpus of healthcare-related messages (not
# publicly provided), what do people want?
health = get_picklecached_thing('health.pickle.gz', None)
svd = run_cnet_blend(health, 'health_blend')
svd.summarize(10)
# Make a labeled vector of similarities to the feature person\Desires.
vec = predict_concepts(svd, ('left', 'Desires', 'person'))
# Get the list of concepts in ConceptNet.
concepts = cnet.label_list(0)