def can_recommend(self, i):
'''See if node i should be allowed to recommend the sp.'''
p = self.Ts[i] >= self.params['trust_threshold']
dist = f.d(self.current_c, self.crecv[i])
tscale = self.Ts[i] * f.sigmoid_s(f.d(self.c[i], self.c_cached[i]))
q = self.params['eta']**(dist + tscale) <= self.params['tau']
return p or q
def update_C(self, i):
'''Update the aggregated context.'''
self.C_cached[i] = self.C[i]
cs = self.cs[
i, 0:min(int(self.current_c[0]) + 1, self.params['zeta'] + 1)]
if len(cs):
dists = self.params['theta']**f.d(self.c[i], cs)
else:
dists = self.params['theta']**f.d(
self.c[i], np.zeros(self.params['number_contexts']))
self.C[i] = np.sum((dists * cs.T).T, axis=0) / np.sum(dists)
def calc_T(self):
'''Calculate the trust of sp.'''
if self.ever_got_recs:
idx = self.ever_got_recs
dists = f.sigmoid_s(-f.d(self.current_c, self.use_c[idx]))
ind_t = np.sum(self.QC[idx] * dists * self.R[idx] *
self.Ts[idx]) / np.sum(dists)
else:
ind_t = 0
alpha = f.sigmoid(2 * np.abs(self.Tdir) /
dist if (dist := f.d(self.Tdir, ind_t)) > 0 else 0)
beta = 1 - alpha
self.T = alpha * self.Tdir + beta * ind_t
return self.T
def update_rep(self, i):
'''Update the reputation.'''
forget = self.params['eta']**f.d(self.c[i], self.C_cached[i])
cached_comb = self.R_cached[i] * forget
idx = list(set(self.got_recs) - {i})
self.R_cached[i] = self.R[i]
gradient = self.params['delta_r'] * sum(self.Ts[i] * self.Ts[idx])
self.R[i] = np.clip(cached_comb + gradient, -1, 1)
def update_QC(self, i):
'''Update the quality of context.'''
self.QC[i] = f.sigmoid_s(f.d(self.c[i], self.C[i]))