def trait_variance(ancestor_pno, db, trait='w2v', n_gens = 2, enforce_func = lambda x: True):
"""
Computes the trait mean and variance*. For now the variance is the norm of the vector of component-wise variances.
"""
# TODO: Need to unfold tfidfs into sparse vector to compute variance.
if trait == 'lda':
raise RuntimeError('Trait variance not currently supported for {}'.format(trait))
stats = {}
# mean_field_name = str(n_gens)+'_gen_trait_mean_' + trait # JM comment out 7/10
var_field_name = str(n_gens)+'_gen_trait_variance_' + trait
trait_field, _, densify_func, _ = _trait_info[trait]
parent = db.traits.find_one({'_id': ancestor_pno}, {'_id': 1, 'citedby': 1, trait_field:1})
lineage = crawl_lineage(db, ancestor_pno, n_gens, fields = ['_id', 'citedby', trait_field], flatten=True, enforce_func = enforce_func)
if lineage is None:
# stats[mean_field_name] = -1 # JM Comment out 7/10
stats[var_field_name] = -1
return stats
traits = [doc.get(trait_field, None) for doc in lineage]
traits = np.array([densify_func(t) for t in traits if t is not None], dtype=np.float64)
if len(traits) > 1:
pass
# stats[mean_field_name] = list(np.mean(traits, axis=0)) # JM comment out 7/10
elif len(traits) == 1:
pass
# stats[mean_field_name] = traits # JM comment out 7/10
elif len(traits) == 0:
# stats[mean_field_name] = -1 # JM comment out 7/10
stats[var_field_name] = -1
return stats
else:
raise RuntimeError("Less than 0 traits???")
stats[var_field_name] = np.linalg.norm(np.var(traits, axis=0)) # For now, get a real number by computing norm.
return stats
def main():
db = MongoClient().patents
family_names = [
"stents",
"zeolites",
"bubblejet",
"cellphone",
"pcr",
"microarrays",
"semiconductors",
"nonwovenwebs",
"rsa",
"browser",
]
family_pnos = [4655771, 4061724, 4723129, 5103459, 4683202, 5143854, 4064521, 4340563, 4405829, 5572643]
family_thresholds = [350, 60, 75, 225, 150, 175, 125, 100, 400, 250]
lilfriend_names = [
"skate",
"murphybed",
"hummingbirdfeeder",
"telescopicumbrella",
"hybridengine",
"minesweeper",
"humanoidrobot",
"recumbentbike",
"hangglider",
"ziplock",
]
lilfriend_pnos = [6000704, 4766623, 5454348, 4880023, 5191766, 3938459, 6377014, 5284351, 4417707, 6004032]
lilfriend_thresholds = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
bigfriend_names = [
"dentallaser",
"ballisticvest",
"hungryhippos",
"sharkprod",
"gatlinggun",
"nuclearwastetreatment",
"gfp",
"roughterrainchasis",
"bowflex",
"radaraltimeter",
]
bigfriend_pnos = [5616141, 4287607, 4119312, 4667431, 4154143, 4274976, 5491084, 4061199, 4725057, 4945360]
bigfriend_thresholds = [25, 25, 10, 12, 8, 9, 25, 30, 15, 10]
names = family_names + lilfriend_names + bigfriend_names
pnos = family_pnos + lilfriend_pnos + bigfriend_pnos
thresholds = family_thresholds + lilfriend_thresholds + bigfriend_thresholds
for pno, threshold, name in zip(pnos, thresholds, names):
print "getting lineage for patent {} ({}), with threhold {}.".format(pno, name, threshold)
lineage = crawl_lineage(
db, pno, n_generations=5, enforce_func=lambda x: len(x.get("citedby", [])) >= threshold, flatten=True
)
adj = subnet_adj_dict(lineage)
dict_fn = "/Users/jmenick/Desktop/sandbox/jacobs_pca_dicts/{}_pca_dict.p".format(name)
colordict = load_obj(dict_fn)
savefn = "{}_{}_force_pca_test.pdf".format(pno, name)
network_plot(pno, adj, colordict, False, savefn)
print "done with {}".format(name)
def test2():
db = MongoClient().patents
family_names = [
'stents', 'zeolites', 'bubblejet', 'cellphone', 'pcr', 'microarrays',
'semiconductors', 'nonwovenwebs', 'rsa', 'browser'
]
family_pnos = [
4655771, 4061724, 4723129, 5103459, 4683202, 5143854, 4064521, 4340563,
4405829, 5572643
]
family_thresholds = [350, 60, 75, 225, 150, 175, 125, 100, 400, 250]
lilfriend_names = [
'skate', 'murphybed', 'hummingbirdfeeder', 'telescopicumbrella',
'hybridengine', 'minesweeper', 'humanoidrobot', 'recumbentbike',
'hangglider', 'ziplock'
]
lilfriend_pnos = [
6000704, 4766623, 5454348, 4880023, 5191766, 3938459, 6377014, 5284351,
4417707, 6004032
]
lilfriend_thresholds = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
bigfriend_names = [
'dentallaser', 'ballisticvest', 'hungryhippos', 'sharkprod',
'gatlinggun', 'nuclearwastetreatment', 'gfp', 'roughterrainchasis',
'bowflex', 'radaraltimeter'
]
bigfriend_pnos = [
5616141, 4287607, 4119312, 4667431, 4154143, 4274976, 5491084, 4061199,
4725057, 4945360
]
bigfriend_thresholds = [25, 25, 10, 12, 8, 9, 25, 30, 15, 10]
names = family_names + lilfriend_names + bigfriend_names
pnos = family_pnos + lilfriend_pnos + bigfriend_pnos
thresholds = family_thresholds + lilfriend_thresholds + bigfriend_thresholds
names = names[:1]
pnos = pnos[:1]
thresholds = thresholds[:1]
for pno, threshold, name in zip(pnos, thresholds, names):
print "getting lineage for patent {} ({}), with threhold {}.".format(
pno, name, threshold)
lineage = crawl_lineage(
db,
pno,
n_generations=5,
enforce_func=lambda x: len(x.get('citedby', [])) >= threshold,
flatten=True)
adj = subnet_adj_dict(lineage)
dict_fn = '/Users/jmenick/Desktop/sandbox/jacobs_pca_dicts/{}_pca_dict.p'.format(
name)
colordict = load_obj(dict_fn)
savefn = '{}_{}_force_pca_test.pdf'.format(pno, name)
print "getting plot..."
network_plot(pno, adj, colordict, False, savefn)
print "done with {}".format(name)
def test():
db = MongoClient().patents
pno = 4723129
threshold = 75
print "getting lineage..."
lineage = crawl_lineage(
db, pno, n_generations=5, enforce_func=lambda x: len(x.get("citedby", [])) >= threshold, flatten=True
)
adj = subnet_adj_dict(lineage)
bubblejet_color_dict_fn = "/Users/jmenick/Desktop/sandbox/jacobs_pca_dicts/bubblejet_pca_dict.p"
bubblejet_colors = load_obj(bubblejet_color_dict_fn)
savefn = "{}_force_pca_test.pdf".format(pno)
print "making plot..."
network_plot(pno, adj, bubblejet_colors, True, savefn)
return adj, bubblejet_colors
def community_colors(db, pno, threshold, show_vis = False, savefn=None):
# Get the patents in the lineage and the adjacency dictionary.
lineage = crawl_lineage(
db, pno, n_generations = 5,
enforce_func = lambda x: len(x.get('citedby', [])) >= threshold,
flatten=True
)
adj = subnet_adj_dict(lineage)
# detect communities
detector = detect.CommunityDetector(adj)
communities = detector.run()
n_communities = len(communities)
community_lookup = util.get_community_lookup(communities)
# assign each patent a color, and provide a lookup dictionary
colors = visualize.discrete_color_scheme(n_communities+1)
node_color_lookup = {node: colors[community_lookup[node]] for node in adj.keys()}
# make the visualization.
G = visualize.get_graph(adj)
G.graph['ancestor'] = pno
ancestor_idx = G.nodes().index(G.graph['ancestor'])
node_colors = [colors[community_lookup[node]] for node in G.nodes()]
node_colors[G.nodes().index(G.graph['ancestor'])] = colors[n_communities]
default_node_size = 60
node_sizes = [default_node_size for node in G.nodes()]
node_sizes[ancestor_idx] = 6*default_node_size
f = plt.figure()
f.set_size_inches(18.5, 10.5)
if savefn is not None or show_vis:
nx.draw_networkx(
G,
nx.spring_layout(G, iterations=20000),
node_color=node_colors,
node_size = node_sizes,
with_labels = False,
fontsize=1,
# font_weight = 'bold',
linewidths=.5,
width=.5
)
if savefn is not None:
plt.title('Communities in Patent {}'.format(pno))
plt.savefig(savefn, dpi=100)
if show_vis:
plt.show()
return node_color_lookup
def test():
db = MongoClient().patents
pno = 4723129
threshold = 75
print "getting lineage..."
lineage = crawl_lineage(
db,
pno,
n_generations=5,
enforce_func=lambda x: len(x.get('citedby', [])) >= threshold,
flatten=True)
adj = subnet_adj_dict(lineage)
bubblejet_color_dict_fn = '/Users/jmenick/Desktop/sandbox/jacobs_pca_dicts/bubblejet_pca_dict.p'
bubblejet_colors = load_obj(bubblejet_color_dict_fn)
savefn = '{}_force_pca_test.pdf'.format(pno)
print "making plot..."
network_plot(pno, adj, bubblejet_colors, True, savefn)
return adj, bubblejet_colors
def parent_child_trait_distance(ancestor_pno, db, trait='w2v', n_gens = 2, enforce_func = lambda x: True):
"""
Computes the sum and average of the parent-descendant trait distances for every descendant begot by the patent
with pno given by the 'ancestor_pno' argument. The appropriate distance function
for the given trait is looked up in the _trait_info dictionary. The distance function is then evaluated
for each descendent in the lineage from the ancestor, up to n_gens in the future. The returned
dictionary contains the sum and average fields, with both fields equal to zero if the patent in question has no
children, and set to -1 as a kind of error value, if none of its descendants have traits. The argument 'enforce_func'
allows the user to set a condition which must be met in order for a patent to be included in the lineage. For instance,
the enforce_func,
lambda x: len(x.get('citedby', [])) > 100
Would enforce that patents only be included in the lineage if they have more than 100 incoming citations.
"""
if trait == 'lda':
raise RuntimeError('Trait variance not currently supported for {}'.format(trait))
stats = {}
sum_fieldname = '_'.join([str(n_gens), 'gen_sum_dist', trait])
avg_fieldname = '_'.join([str(n_gens), 'gen_avg_dist', trait])
trait_field,dist_func, _, _ = _trait_info[trait]
parent = db.traits.find_one({'_id': ancestor_pno}, {'_id': 1, 'citedby': 1, trait_field:1})
if parent is None:
return None # We don't have that ancestor.
lineage = crawl_lineage(db, ancestor_pno, n_gens, fields = ['_id', 'citedby', trait_field], flatten=True, enforce_func = enforce_func)[1:]
if lineage is None:
stats[sum_fieldname] = -1
stats[avg_fieldname] = -1
return stats
if not (parent.get('citedby', None) and parent.get(trait_field, None)):
stats[sum_fieldname] = 0
stats[avg_fieldname] = 0
return stats
traits = [doc.get(trait_field, None) for doc in lineage]
traits = np.array([t for t in traits if t is not None], dtype=np.float64)
n_children_with_traits = len(traits)
if n_children_with_traits == 0:
stats[sum_fieldname] = -1
stats[avg_fieldname] = -1
return stats
dist_sum = np.sum([dist_func(parent.get(trait_field), trait) for trait in traits])
stats[sum_fieldname] = dist_sum
stats[avg_fieldname] = dist_sum/n_children_with_traits
return stats
def testCrawl2Gens(self):
for pno in self.pnos_test:
dbutil.crawl_lineage(self.db, pno, n_generations = 2)
