def calculatePropensities(self, emo=True, role=True):
for edge in self.edges: # for every edge, calculate propensities and append as an attribute
attributeDict = {}
emoPropList = propensities.propCalc(self, edge)[0] if emo else None
if len(emoPropList) > 0:
attributeDict['Emotion'] = emoPropList
attributeDict['emoWeight'] = propensities.aggregateProps(
emoPropList)
rolePropList = propensities.propCalc(self,
edge)[1] if role else None
if len(rolePropList) > 0:
attributeDict['Role'] = rolePropList
attributeDict['roleWeight'] = propensities.aggregateProps(
rolePropList)
inflPropList = propensities.propCalc(self,
edge)[2] if role else None
if len(inflPropList) > 0:
attributeDict['Influence'] = inflPropList
attributeDict['inflWeight'] = propensities.aggregateProps(
inflPropList)
self.G[edge[0]][edge[1]] = attributeDict
self.edges = nx.edges(self.G)
def calculatePropensities(self,
propToggle={
'emo': True,
'infl': True,
'role': True
}):
self.propToggle = propToggle
for edge in self.edges: # for every edge, calculate propensities and append as an attribute
attributeDict = {"propsFlag": True}
IO, emoPropList, rolePropList, inflPropList = propensities.propCalc(
self, edge, propToggle)
attributeDict['IO'] = IO
if propToggle['emo'] and len(emoPropList) > 0:
attributeDict['Emotion'] = emoPropList
if propToggle['role'] and len(rolePropList) > 0:
attributeDict['Role'] = rolePropList
if propToggle['infl'] and len(inflPropList) > 0:
attributeDict['Influence'] = inflPropList
self.G[edge[0]][edge[1]].update(attributeDict)
self.edges = nx.edges(self.G)
def drag_predict(self, node=None):
# ERGM generates probability matrix where order is G.nodes() x G.nodes()
ergm_prob_mat = ergm.probability(G=self.G)
if node is not None:
# Assigning propensities probabilities and generating add_node links
for target in self.G.nodes_iter():
IO, emoProps, roleProps, inflProps = propensities.propCalc(
self, (node, target))
if len(emoProps) > 0:
w = []
for prop in emoProps:
if len(prop) > 1:
w.append(
prop[1]
) # add the product of the attribute weights to a list for each prop
w_avg = np.average(
w) # find average propensity product weight
prob = np.random.binomial(
1, w_avg * 1 / 2 if w_avg * 1 / 2 > 0 else 0)
# use w_avg as the probability for a bernoulli distribution
if prob:
self.G.add_edge(node, target)
self.G[node][target]['Emotion'] = emoProps
self.G[node][target]['Role'] = roleProps if len(
roleProps) > 0 else None
self.G[node][target]['Influence'] = inflProps if len(
inflProps) > 0 else None
self.G[node][target]['Predicted'] = True
# iterate through all possible edges
for i, j in product(range(len(self.G.nodes())), repeat=2):
if i != j:
node = self.G.nodes()[i]
target = self.G.nodes()[j]
prob = ergm_prob_mat[i, j] * 0.05
# check props
if self.G[node].get(target) is not None:
## check emo props to modify adjacency prob matrix if present
if self.G[node][target].get('Emotion') is not None:
w = []
for prop in emoProps:
w.append(
prop[4] * prop[5]
) # add the product of the attribute weights to a list for each prop
w_avg = np.average(w)
prob = (prob + w_avg * 0.5) / 2
presence = np.random.binomial(1, prob) if prob < 1 else 1
# use adjacency prob mat as the probability for a bernoulli distribution
if presence:
self.G.add_edge(node, target)
self.G[node][target]['Predicted'] = True
self.feedbackUpdate()