def update(self, instance):
"""
instance: [UserId, ItemId, LabelId] where LabelID should be index in range(self.L)
"""
uid, iid, lid = instance
## calculate single gradient ##
# intermediate #
m = TDreconstruct(self.c, self.u[uid], self.v[iid], self.r)
expm = np.exp(m)
expmsum = np.sum(expm)
mgrad = expm / expmsum
mgrad[lid] = mgrad[lid] - 1.0
mgrad = - mgrad
### test ###
# print "m", m
# print "mgrad", mgrad
# print "lid", lid
###############
# gradient for embeddings #
delt_u = np.tensordot(a = mgrad, axes = (0,1),
b = np.tensordot(a = self.v[iid], axes = (0,1),
b = np.tensordot(a = self.c, axes = (2,1),
b = self.r)
)
)
delt_v = np.tensordot(a = mgrad, axes = (0,1),
b = np.tensordot(a = self.u[uid], axes = (0,0),
b = np.tensordot(a = self.c, axes = (2,1),
b = self.r)
)
)
delt_c = matrixTool.TensorOuter([self.u[uid], self.v[iid], np.tensordot(mgrad, self.r, axes = (0,0))])
delt_r = np.outer(mgrad, np.tensordot(a = self.u[uid], axes = (0,0),
b = np.tensordot(a = self.v[iid], axes = (0,1),
b = self.c)))
### test ###
# rela_u = np.linalg.norm(delt_u)/np.linalg.norm(self.u[uid])
# rela_v = np.linalg.norm(delt_v)/np.linalg.norm(self.v[iid])
# rela_c = np.linalg.norm(delt_c)/np.linalg.norm(self.c)
# rela_r = np.linalg.norm(delt_r)/np.linalg.norm(self.r)
# print "relative step size", rela_u, rela_v, rela_c, rela_r
# update #
self.u[uid] += (self.SGDstep * (delt_u - self.lamda * self.u[uid]))
self.v[iid] += (self.SGDstep * (delt_v - self.lamda * self.v[iid]))
self.c += (self.SGDstep * (delt_c - self.lamda * self.c))
self.r += (self.SGDstep * (delt_r - self.lamda * self.r))
### test ###
# m = TDreconstruct(self.c, self.u[uid], self.v[iid], self.r)
# print "m after update", m
return self
def update(self, instance):
uid, iid, lid = instance
## calculate single gradient ##
# intermediate #
m = TDreconstruct(self.c, self.u[uid], self.v[iid], self.r)
mgrad = -m
mgrad[lid] = 1.0 + mgrad[lid]
# gradient for embeddings #
delt_u = np.tensordot(a=mgrad,
axes=(0, 1),
b=np.tensordot(a=self.v[iid],
axes=(0, 1),
b=np.tensordot(a=self.c,
axes=(2, 1),
b=self.r)))
delt_v = np.tensordot(a=mgrad,
axes=(0, 1),
b=np.tensordot(a=self.u[uid],
axes=(0, 0),
b=np.tensordot(a=self.c,
axes=(2, 1),
b=self.r)))
delt_c = matrixTool.TensorOuter([
self.u[uid], self.v[iid],
np.tensordot(mgrad, self.r, axes=(0, 0))
])
delt_r = np.outer(
mgrad,
np.tensordot(a=self.u[uid],
axes=(0, 0),
b=np.tensordot(a=self.v[iid], axes=(0, 1), b=self.c)))
# update #
self.u[uid] += (self.SGDstep * (delt_u - self.lamda * self.u[uid]))
self.v[iid] += (self.SGDstep * (delt_v - self.lamda * self.v[iid]))
self.c += (self.SGDstep * (delt_c - self.lamda * self.c))
self.r += (self.SGDstep * (delt_r - self.lamda * self.r))
return self
def update(self, instance):
uid, iid, lid, value = instance
m = TDreconstruct(self.c, self.u[uid], self.v[iid], self.r[lid])
mgrad = 2.0 * (value - m)
# gradient for embeddings #
delt_u = mgrad * np.tensordot(
a=self.v[iid],
axes=(0, 1),
b=np.tensordot(a=self.c, axes=(2, 0), b=self.r[lid]))
delt_v = mgrad * np.tensordot(
a=self.u[uid],
axes=(0, 0),
b=np.tensordot(a=self.c, axes=(2, 0), b=self.r[lid]))
delt_c = mgrad * matrixTool.TensorOuter(
[self.u[uid], self.v[iid], self.r[lid]])
delt_r = mgrad * np.tensordot(
a=self.u[uid],
axes=(0, 0),
b=np.tensordot(a=self.v[iid], axes=(0, 1), b=self.c))
# update #
self.u[uid] += (self.SGDstep * delt_u)
self.v[iid] += (self.SGDstep * delt_v)
self.c += (self.SGDstep * delt_c)
self.r[lid] += (self.SGDstep * delt_r)
def update(self, instance):
"""
instance: [UserId, ItemId, LabelId] where LabelID should be index in range(self.L)
"""
uid, iid, lid = instance
## calculate single gradient ##
# intermediate #
m = TDreconstruct(self.c, self.u[uid], self.v[iid], self.r)
mgrad = softmaxGradient(m, lid)
# gradient for embeddings #
delt_u = np.tensordot(a=mgrad, axes=(0, 1),
b=np.tensordot(a=self.v[iid], axes=(0, 1),
b=np.tensordot(a=self.c, axes=(2, 1),
b=self.r)
)
)
delt_v = np.tensordot(a=mgrad, axes=(0, 1),
b=np.tensordot(a=self.u[uid], axes=(0, 0),
b=np.tensordot(a=self.c, axes=(2, 1),
b=self.r)
)
)
delt_c = matrixTool.TensorOuter([self.u[uid], self.v[iid], np.tensordot(mgrad, self.r, axes=(0, 0))])
delt_r = np.outer(mgrad, np.tensordot(a=self.u[uid], axes=(0, 0),
b=np.tensordot(a=self.v[iid], axes=(0, 1),
b=self.c)))
# update #
self.u[uid] += (self.SGDstep * (delt_u - self.lamda * self.u[uid]))
self.v[iid] += (self.SGDstep * (delt_v - self.lamda * self.v[iid]))
self.c += (self.SGDstep * (delt_c - self.lamda * self.c))
self.r += (self.SGDstep * (delt_r - self.lamda * self.r))
### test ###
# m = TDreconstruct(self.c, self.u[uid], self.v[iid], self.r)
# print "m after update", m
return self