def get_projection_gradient(self, data, params):
(X1, X2) = data
zmuvX1 = get_zm_uv(X1)
zmuvX2 = get_zm_uv(X2)
return (zmuvX1.T, zmuvX2.T)
def get_projected(self, data, params):
(X1, X2) = data
(Phi1, Phi2) = params
# Get empirical covariance matrices
zmuvX1 = get_zm_uv(X1)
zmuvX2 = get_zm_uv(X2)
# Get projected data
XPhi1 = np.dot(zmuvX1, Phi1)
XPhi2 = np.dot(zmuvX2, Phi2)
# Get gram matrices
S1 = np.dot(XPhi1.T, XPhi1)
S2 = np.dot(XPhi2.T, XPhi2)
# Get normalizers
inv_srqt1 = get_svd_power(S1, -0.5)
inv_srqt2 = get_svd_power(S2, -0.5)
# Get normalized
normed1 = np.dot(Phi1, inv_sqrt1)
normed2 = np.dot(Phi2, inv_sqrt2)
return (normed1, normed2)
def get_residuals(self, data, params):
(Phi1, Phi2) = params
(X1, X2) = data
zmuvX1 = get_zm_uv(X1)
zmuvX2 = get_zm_uv(X2)
XPhi1 = np.dot(zmuvX1, Phi1)
XPhi2 = np.dot(zmuvX2, Phi2)
return XPhi1 - XPhi2
def get_gradient(self, data, params):
residuals = self.get_residuals(data, params)
(X1, X2) = data
N = X1.shape[0]
zmuvX1 = get_zm_uv(X1)
zmuvX2 = get_zm_uv(X2)
grad1 = np.dot(zmuvX1.T, residuals) / N
grad2 = np.dot(zmuvX2.T, -residuals) / N
return (grad1, grad2)
def run(self):
printerval = self.num_data / 10
for t in range(self.num_data):
zipped = zip(self.agvs, self.servers)
for (agv, ds) in zipped:
agv.set_data(ds.get_data())
if self.delay is None or t % self.delay == 0:
XPhis = [agv.get_projected() for agv in self.agvs]
for agv in self.agvs:
for (n, XPhi) in enumerate(XPhis):
agv.update_neighbor_state(n, XPhi)
(self.Phis,
new_tccs) = unzip([agv.get_update() for agv in self.agvs])
self.tccs.append(new_tccs)
if False: #t % printerval == 0:
print('tccs', new_tccs)
zipped = zip(self.Phis, self.loaders)
XPhis = [np.dot(get_zm_uv(dl.get_data()), Phi) for (Phi, dl) in zipped]
tcc_mats = [
np.dot(XPhis[n].T, XPhis[m]) for n in range(self.num_views)
for m in range(n, self.num_views)
]
tccs = [np.trace(tccm) / self.k for tccm in tcc_mats]
print(tccs)
def get_prediction(self, data, params):
return np.dot(get_zm_uv(data), params)
def get_batch(self):
return get_zm_uv(self.batch)