def forward_ineq(Q, p, G, h, verbose=0, maxIter=100, dt=0.2):
""" Solves a QP problem with only inequality constraints, fastest version
for the dynamic solver
"""
nineq, nz, neq, nBatch = get_sizes(G)
A_T = torch.transpose(G, -1, 1)
Q_I = torch.inverse(Q)
AQ_I = G.bmm(Q_I)
D = -AQ_I.bmm(A_T)
d = -(h.unsqueeze(2) + AQ_I.bmm(p.unsqueeze(2)))
lams = torch.zeros(nBatch, nineq, 1).type_as(Q).to(Q.device)
zeros = torch.zeros(nBatch, nineq, 1).type_as(Q).to(Q.device)
for _ in range(maxIter):
dlams = dt * (D.bmm(lams) + d)
dlams = torch.max(lams + dlams, zeros) - lams
lams.add_(dlams)
# lams = torch.where(lams > 0, lams, zeros)
# lams = torch.max(lams + dlams, zeros)
zhat = -Q_I.bmm(p.unsqueeze(2) + A_T.bmm(lams))
slacks = h.unsqueeze(2) - G.bmm(zhat)
return zhat.squeeze(2), lams.squeeze(2), slacks.squeeze(2)
def solve_kkt(Q_LU, d, G, A, S_LU, rx, rs, rz, ry):
""" Solve KKT equations for the affine step"""
nineq, nz, neq, nBatch = get_sizes(G, A)
invQ_rx = rx.btrs(Q_LU)
if neq > 0:
h = torch.cat(
(invQ_rx.unsqueeze(1).bmm(A.transpose(1, 2)) - ry,
invQ_rx.unsqueeze(1).bmm(G.transpose(1, 2)) + rs / d - rz),
2).squeeze(1)
else:
h = (invQ_rx.unsqueeze(1).bmm(G.transpose(1, 2)) + rs / d -
rz).squeeze(1)
w = -(h.btrs(S_LU))
g1 = -rx - w[:, neq:].unsqueeze(1).bmm(G)
if neq > 0:
g1 -= w[:, :neq].unsqueeze(1).bmm(A)
g2 = -rs - w[:, neq:]
dx = g1.btrs(Q_LU)
ds = g2 / d
dz = w[:, neq:]
dy = w[:, :neq] if neq > 0 else None
return dx, ds, dz, dy
def pre_factor_kkt(Q, G, A):
""" Perform all one-time factorizations and cache relevant matrix products"""
nineq, nz, neq, _ = get_sizes(G, A)
# S = [ A Q^{-1} A^T A Q^{-1} G^T ]
# [ G Q^{-1} A^T G Q^{-1} G^T + D^{-1} ]
U_Q = torch.potrf(Q)
# partial cholesky of S matrix
U_S = torch.zeros(neq + nineq, neq + nineq).type_as(Q)
G_invQ_GT = torch.mm(G, torch.potrs(G.t(), U_Q))
R = G_invQ_GT
if neq > 0:
invQ_AT = torch.potrs(A.t(), U_Q)
A_invQ_AT = torch.mm(A, invQ_AT)
G_invQ_AT = torch.mm(G, invQ_AT)
# TODO: torch.potrf sometimes says the matrix is not PSD but
# numpy does? I filed an issue at
# https://github.com/pytorch/pytorch/issues/199
try:
U11 = torch.potrf(A_invQ_AT)
except:
U11 = torch.Tensor(np.linalg.cholesky(
A_invQ_AT.cpu().numpy())).type_as(A_invQ_AT)
# TODO: torch.trtrs is currently not implemented on the GPU
# and we are using gesv as a workaround.
U12 = torch.gesv(G_invQ_AT.t(), U11.t())[0]
U_S[:neq, :neq] = U11
U_S[:neq, neq:] = U12
R -= torch.mm(U12.t(), U12)
return U_Q, U_S, R
def solve_kkt(Q_LU, d, G, A, S_LU, rx, rs, rz, ry):
""" Solve KKT equations for the affine step"""
nineq, nz, neq, nBatch = get_sizes(G, A)
invQ_rx = rx.unsqueeze(2).lu_solve(*Q_LU).squeeze(2)
if neq > 0:
h = torch.cat(
(invQ_rx.unsqueeze(1).bmm(A.transpose(1, 2)).squeeze(1) - ry,
invQ_rx.unsqueeze(1).bmm(G.transpose(1, 2)).squeeze(1) + rs / d -
rz), 1)
else:
h = invQ_rx.unsqueeze(1).bmm(G.transpose(1,
2)).squeeze(1) + rs / d - rz
w = -(h.unsqueeze(2).lu_solve(*S_LU)).squeeze(2)
# if any(torch.isnan(torch.flatten(w)).tolist()):
# logging.info("**Alert** nan in param w ")
# logging.info( any(torch.isnan(torch.flatten(S_LU[0])).tolist()))
# logging.info( any(torch.isnan(torch.flatten(S_LU[1])).tolist()))
# logging.info( any(torch.isnan(torch.flatten(h)).tolist()))
# w = -sp_lu_solve(h.unsqueeze(2),*S_LU).squeeze(2)
# w[torch.isnan(w)] = 0
# logging.info(w[0:3])
g1 = -rx - w[:, neq:].unsqueeze(1).bmm(G).squeeze(1)
if neq > 0:
g1 -= w[:, :neq].unsqueeze(1).bmm(A).squeeze(1)
g2 = -rs - w[:, neq:]
dx = g1.unsqueeze(2).lu_solve(*Q_LU).squeeze(2)
ds = g2 / d
dz = w[:, neq:]
dy = w[:, :neq] if neq > 0 else None
return dx, ds, dz, dy
def factor_solve_kkt(Q, D, G, A, rx, rs, rz, ry):
nineq, nz, neq, _ = get_sizes(G, A)
if neq > 0:
H_ = torch.cat([
torch.cat([Q, torch.zeros(nz, nineq).type_as(Q)], 1),
torch.cat([torch.zeros(nineq, nz).type_as(Q), D], 1)
], 0)
A_ = torch.cat([
torch.cat([G, torch.eye(nineq).type_as(Q)], 1),
torch.cat([A, torch.zeros(neq, nineq).type_as(Q)], 1)
], 0)
g_ = torch.cat([rx, rs], 0)
h_ = torch.cat([rz, ry], 0)
else:
H_ = torch.cat([
torch.cat([Q, torch.zeros(nz, nineq).type_as(Q)], 1),
torch.cat([torch.zeros(nineq, nz).type_as(Q), D], 1)
], 0)
A_ = torch.cat([G, torch.eye(nineq).type_as(Q)], 1)
g_ = torch.cat([rx, rs], 0)
h_ = rz
U_H_ = torch.potrf(H_)
invH_A_ = torch.potrs(A_.t(), U_H_)
invH_g_ = torch.potrs(g_.view(-1, 1), U_H_).view(-1)
S_ = torch.mm(A_, invH_A_)
U_S_ = torch.potrf(S_)
t_ = torch.mv(A_, invH_g_).view(-1, 1) - h_
w_ = -torch.potrs(t_, U_S_).view(-1)
v_ = torch.potrs(-g_.view(-1, 1) - torch.mv(A_.t(), w_), U_H_).view(-1)
return v_[:nz], v_[nz:], w_[:nineq], w_[nineq:] if neq > 0 else None
def solve_kkt(U_Q, d, G, A, U_S, rx, rs, rz, ry, dbg=False):
""" Solve KKT equations for the affine step"""
nineq, nz, neq, _ = get_sizes(G, A)
invQ_rx = torch.potrs(rx.view(-1, 1), U_Q).view(-1)
if neq > 0:
h = torch.cat(
[torch.mv(A, invQ_rx) - ry,
torch.mv(G, invQ_rx) + rs / d - rz], 0)
else:
h = torch.mv(G, invQ_rx) + rs / d - rz
w = -torch.potrs(h.view(-1, 1), U_S).view(-1)
g1 = -rx - torch.mv(G.t(), w[neq:])
if neq > 0:
g1 -= torch.mv(A.t(), w[:neq])
g2 = -rs - w[neq:]
dx = torch.potrs(g1.view(-1, 1), U_Q).view(-1)
ds = g2 / d
dz = w[neq:]
dy = w[:neq] if neq > 0 else None
# if np.all(np.array([x.norm() for x in [rx, rs, rz, ry]]) != 0):
if dbg:
import IPython
import sys
IPython.embed()
sys.exit(-1)
# if rs.norm() > 0: import IPython, sys; IPython.embed(); sys.exit(-1)
return dx, ds, dz, dy
def solve_kkt_ir(Q, D, G, A, rx, rs, rz, ry, niter=1):
"""Inefficient iterative refinement."""
nineq, nz, neq, nBatch = get_sizes(G, A)
eps = 1e-7
Q_tilde = Q + eps * torch.eye(nz).type_as(Q).repeat(nBatch, 1, 1)
D_tilde = D + eps * torch.eye(nineq).type_as(Q).repeat(nBatch, 1, 1)
dx, ds, dz, dy = factor_solve_kkt_reg(
Q_tilde, D_tilde, G, A, rx, rs, rz, ry, eps)
res = kkt_resid_reg(Q, D, G, A, eps,
dx, ds, dz, dy, rx, rs, rz, ry)
resx, ress, resz, resy = res
res = resx
for k in range(niter):
ddx, dds, ddz, ddy = factor_solve_kkt_reg(Q_tilde, D_tilde, G, A, -resx, -ress, -resz,
-resy if resy is not None else None,
eps)
dx, ds, dz, dy = [v + dv if v is not None else None
for v, dv in zip((dx, ds, dz, dy), (ddx, dds, ddz, ddy))]
res = kkt_resid_reg(Q, D, G, A, eps,
dx, ds, dz, dy, rx, rs, rz, ry)
resx, ress, resz, resy = res
# res = torch.cat(resx)
res = resx
return dx, ds, dz, dy
def factor_solve_kkt(Q, D, G, A, rx, rs, rz, ry):
nineq, nz, neq, nBatch = get_sizes(G, A)
H_ = torch.zeros(nBatch, nz + nineq, nz + nineq).type_as(Q)
H_[:, :nz, :nz] = Q
H_[:, -nineq:, -nineq:] = D
if neq > 0:
A_ = torch.cat([torch.cat([G, torch.eye(nineq).type_as(Q).repeat(nBatch, 1, 1)], 2),
torch.cat([A, torch.zeros(nBatch, neq, nineq).type_as(Q)], 2)], 1)
g_ = torch.cat([rx, rs], 1)
h_ = torch.cat([rz, ry], 1)
else:
A_ = torch.cat([G, torch.eye(nineq).type_as(Q)], 1)
g_ = torch.cat([rx, rs], 1)
h_ = rz
H_LU = lu_hack(H_)
invH_A_ = A_.transpose(1, 2).lu_solve(*H_LU)
invH_g_ = g_.lu_solve(*H_LU)
S_ = torch.bmm(A_, invH_A_)
S_LU = lu_hack(S_)
t_ = torch.bmm(invH_g_.unsqueeze(1), A_.transpose(1, 2)).squeeze(1) - h_
w_ = -t_.lu_solve(*S_LU)
t_ = -g_ - w_.unsqueeze(1).bmm(A_).squeeze()
v_ = t_.lu_solve(*H_LU)
dx = v_[:, :nz]
ds = v_[:, nz:]
dz = w_[:, :nineq]
dy = w_[:, nineq:] if neq > 0 else None
return dx, ds, dz, dy
def pre_factor_kkt(Q, G, A):
""" Perform all one-time factorizations and cache relevant matrix products"""
nineq, nz, neq, nBatch = get_sizes(G, A)
try:
Q_LU = lu_hack(Q)
except:
raise RuntimeError("""
qpth Error: Cannot perform LU factorization on Q.
Please make sure that your Q matrix is PSD and has
a non-zero diagonal.
""")
# S = [ A Q^{-1} A^T A Q^{-1} G^T ]
# [ G Q^{-1} A^T G Q^{-1} G^T + D^{-1} ]
#
# We compute a partial LU decomposition of the S matrix
# that can be completed once D^{-1} is known.
# See the 'Block LU factorization' part of our website
# for more details.
qlu, pivots=Q_LU
##put in a condition for m>1
if G.size(2)==2: ##something funny here
print(G.size(),qlu.size())
G_invQ_GT = torch.matmul(G, G.lu_solve(*Q_LU).transpose(1,2))
else:
print(G.size(2),G.transpose(1,2).size(),qlu.size())
G_invQ_GT = torch.bmm(G, G.transpose(1, 2).lu_solve(*Q_LU))
R = G_invQ_GT.clone()
S_LU_pivots = torch.IntTensor(range(1, 1 + neq + nineq)).unsqueeze(0) \
.repeat(nBatch, 1).type_as(Q).int()
if neq > 0:
invQ_AT = A.transpose(1, 2).lu_solve(*Q_LU)
A_invQ_AT = torch.bmm(A, invQ_AT)
G_invQ_AT = torch.bmm(G, invQ_AT)
LU_A_invQ_AT = lu_hack(A_invQ_AT)
P_A_invQ_AT, L_A_invQ_AT, U_A_invQ_AT = torch.lu_unpack(*LU_A_invQ_AT)
P_A_invQ_AT = P_A_invQ_AT.type_as(A_invQ_AT)
S_LU_11 = LU_A_invQ_AT[0]
U_A_invQ_AT_inv = (P_A_invQ_AT.bmm(L_A_invQ_AT)
).lu_solve(*LU_A_invQ_AT)
S_LU_21 = G_invQ_AT.bmm(U_A_invQ_AT_inv)
T = G_invQ_AT.transpose(1, 2).lu_solve(*LU_A_invQ_AT)
S_LU_12 = U_A_invQ_AT.bmm(T)
S_LU_22 = torch.zeros(nBatch, nineq, nineq).type_as(Q)
S_LU_data = torch.cat((torch.cat((S_LU_11, S_LU_12), 2),
torch.cat((S_LU_21, S_LU_22), 2)),
1)
S_LU_pivots[:, :neq] = LU_A_invQ_AT[1]
R -= G_invQ_AT.bmm(T)
else:
S_LU_data = torch.zeros(nBatch, nineq, nineq).type_as(Q)
S_LU = [S_LU_data, S_LU_pivots]
return Q_LU, S_LU, R
def forward_eq_conv(Q, p, G_, h_, A_, b_, verbose=0, maxIter=100, dt=0.2):
""" DOES NOT WORK, ONLY HERE TO DOCUMENT OUR PROCESS
Solves the QP problem by transforming equality constraints into two
inequality constraint, then using the ineq solver.
"""
nineq, nz, neq, nBatch = get_sizes(G_, A_)
G = torch.cat((G_, A_, -A_), dim=1)
h = torch.cat((h_, b_, -b_), dim=1)
A_T = torch.transpose(G, -1, 1)
Q_I = torch.inverse(Q)
AQ_I = G.bmm(Q_I)
D = -AQ_I.bmm(A_T)
d = -(h.unsqueeze(2) + AQ_I.bmm(p.unsqueeze(2)))
lams = torch.zeros(nBatch, nineq + neq + neq, 1).type_as(Q).to(Q.device)
zeros = torch.zeros(nBatch, nineq + neq + neq, 1).type_as(Q).to(Q.device)
for _ in range(maxIter):
dlams = dt * (D.bmm(lams) + d)
dlams = torch.max(lams + dlams, zeros) - lams
lams.add_(dlams)
zhat = -Q_I.bmm(p.unsqueeze(2) + A_T.bmm(lams))
slacks = h.unsqueeze(2) - G.bmm(zhat)
return zhat.squeeze(2), lams.squeeze(2)[:, :nineq], lams.squeeze(
2)[:, -neq:], slacks.squeeze(2)
def pre_factor_kkt(Q, G, A):
""" Perform all one-time factorizations and cache relevant matrix products"""
nineq, nz, neq, nBatch = get_sizes(G, A)
Q_LU = Q.btrf()
# S = [ A Q^{-1} A^T A Q^{-1} G^T ]
# [ G Q^{-1} A^T G Q^{-1} G^T + D^{-1} ]
#
# We compute a partial LU decomposition of S matrix
# that can be completed once D^{-1} is known.
# This is done for a general matrix by decomposing
# S using the Schur complement and then LU-factorizing
# the matrices in the middle:
#
# [ A B ] = [ I 0 ] [ A 0 ] [ I A^{-1} B ]
# [ C D ] [ C A^{-1} I ] [ 0 D - C A^{-1} B ] [ 0 I ]
G_invQ_GT = torch.bmm(G, G.transpose(1, 2).btrs(Q_LU))
R = G_invQ_GT
if neq > 0:
invQ_AT = A.transpose(1, 2).btrs(Q_LU)
A_invQ_AT = torch.bmm(A, invQ_AT)
G_invQ_AT = torch.bmm(G, invQ_AT)
LU_A_invQ_AT = A_invQ_AT.btrf()
if neq == 1:
LU_A_invQ_AT = LU_A_invQ_AT.view(-1, 1, 1)
M = torch.tril(torch.ones(neq,
neq)).unsqueeze(0).expand(nBatch, neq,
neq).type_as(Q)
L_A_invQ_AT = M * LU_A_invQ_AT
L_A_invQ_AT[torch.eye(neq).unsqueeze(0).expand(
nBatch, neq, neq).type_as(Q).byte()] = 1.0
M = torch.triu(torch.ones(neq,
neq)).unsqueeze(0).expand(nBatch, neq,
neq).type_as(Q)
U_A_invQ_AT = M * LU_A_invQ_AT
S_LU_11 = LU_A_invQ_AT
S_LU_21 = G_invQ_AT.bmm(L_A_invQ_AT.btrs(LU_A_invQ_AT))
T = G_invQ_AT.transpose(1, 2).btrs(LU_A_invQ_AT)
S_LU_12 = U_A_invQ_AT.bmm(T)
S_LU = torch.cat(
(torch.cat((S_LU_11, S_LU_12), 2),
torch.cat(
(S_LU_21, torch.zeros(nBatch, nineq, nineq).type_as(Q)), 2)),
1)
R -= G_invQ_AT.bmm(T)
else:
S_LU = torch.zeros(nBatch, nineq, nineq).type_as(Q)
return Q_LU, S_LU, R
def solve_kkt(Q_LU, d, G, A, S_LU, rx, rs, rz, ry):
""" Solve KKT equations for the affine step"""
nineq, nz, neq, nBatch = get_sizes(G, A)
qlu, pivots=Q_LU
#print("InEq cons, nz, eq cons, btch",nineq, nz, neq, nBatch) #how is G changing dim to 3D
#print("Now in solve KKt", G.size(),rx.size(), qlu.size())
if G.size(1)>=2:
invQ_rx = rx.unsqueeze(2).lu_solve(*Q_LU).squeeze(2)
else:
invQ_rx = rx.unsqueeze(2).lu_solve(*Q_LU).squeeze(2)
#print("Size after",invQ_rx.size(),G.size(),rs.size(),rz.size())
if neq > 0:
h = torch.cat((invQ_rx.unsqueeze(1).bmm(A.transpose(1, 2)).squeeze(1) - ry,
invQ_rx.unsqueeze(1).bmm(G.transpose(1, 2)).squeeze(1) + rs / d - rz), 1)
else:
h = invQ_rx.unsqueeze(1).bmm(G.transpose(1, 2)).squeeze(1) + rs / d - rz
#if(rz.size(0)==1):
# h=torch.cat(h,h,1)
slu, pv=S_LU
#print("W getting generated",h.size(),slu.size())
if G.size(1)>=2 and h.size(0)>1:
w = -(h.unsqueeze(0).lu_solve(*S_LU)).squeeze(2)
w=w[:,0,:]
else:
w = -(h.unsqueeze(2).lu_solve(*S_LU)).squeeze(2)
#print(w.size())
if G.size(1)>=2:
g1 = -rx - w[:, neq:].matmul(G).squeeze(1)
else:
g1 = -rx - w[:, neq:].unsqueeze(1).bmm(G).squeeze(1)
if neq > 0:
g1 -= w[:, :neq].unsqueeze(1).bmm(A).squeeze(1)
g2 = -rs - w[:, neq:]
#print(g1.size(),qlu.size())
if g1.dim()>2:
dx=g1.transpose(1,2).lu_solve(*Q_LU).transpose(1,2).squeeze(2)
else:
dx = g1.unsqueeze(2).lu_solve(*Q_LU).squeeze(2)
ds = g2 / d
dz = w[:, neq:]
dy = w[:, :neq] if neq > 0 else None
# print("Done Solve KKT");
return dx, ds, dz, dy
def factor_solve_kkt(Q, D, G, A, rx, rs, rz, ry):
nineq, nz, neq, nBatch = get_sizes(G, A)
if neq > 0:
# import IPython, sys; IPython.embed(); sys.exit(-1)
# H_ = torch.cat([torch.cat([Q, torch.zeros(nz,nineq).type_as(Q)], 1),
# torch.cat([torch.zeros(nineq, nz).type_as(Q), D], 1)], 0)
# A_ = torch.cat([torch.cat([G, torch.eye(nineq).type_as(Q)], 1),
# torch.cat([A, torch.zeros(neq, nineq).type_as(Q)], 1)], 0)
# g_ = torch.cat([rx, rs], 0)
# h_ = torch.cat([rz, ry], 0)
H_ = torch.zeros(nBatch, nz + nineq, nz + nineq).type_as(Q)
H_[:, :nz, :nz] = Q.repeat(nBatch, 1, 1)
H_[:, -nineq:, -nineq:] = D
from block import block
A_ = block(((G, 'I'), (A, torch.zeros(neq, nineq).type_as(Q))))
g_ = torch.cat([rx, rs], 1)
h_ = torch.cat([rz, ry], 1)
else:
H_ = torch.zeros(nBatch, nz + nineq, nz + nineq).type_as(Q)
H_[:, :nz, :nz] = Q.repeat(nBatch, 1, 1)
H_[:, -nineq:, -nineq:] = D
A_ = torch.cat([G, torch.eye(nineq).type_as(Q)], 1)
g_ = torch.cat([rx, rs], 1)
h_ = rz
H_LU = H_.btrf()
A = A_.repeat(nBatch, 1, 1)
invH_A_ = A.transpose(1, 2).btrs(H_LU)
invH_g_ = g_.btrs(H_LU)
S_ = torch.bmm(A, invH_A_)
S_LU = S_.btrf()
t_ = torch.mm(invH_g_, A_.t()) - h_
w_ = -t_.btrs(S_LU)
t_ = -g_ - w_.mm(A_)
v_ = t_.btrs(H_LU)
dx = v_[:, :nz]
ds = v_[:, nz:]
dz = w_[:, :nineq]
dy = w_[:, nineq:] if neq > 0 else None
return dx, ds, dz, dy
def factor_solve_kkt(Q, D, G, A, rx, rs, rz, ry):
nineq, nz, neq, nBatch = get_sizes(G, A)
H_ = torch.zeros(nBatch, nz + nineq, nz + nineq).type_as(Q)
H_[:, :nz, :nz] = Q
H_[:, -nineq:, -nineq:] = D
G=G.squeeze(0)#added
if neq > 0:
A_ = torch.cat([torch.cat([G, torch.eye(nineq).type_as(Q).repeat(nBatch, 1, 1)], 2),
torch.cat([A, torch.zeros(nBatch, neq, nineq).type_as(Q)], 2)], 1)
g_ = torch.cat([rx, rs], 1)
h_ = torch.cat([rz, ry], 1)
else:
#print(G.size(),torch.eye(nineq).type_as(Q).size())
A_ = torch.cat([G, torch.eye(nineq).type_as(Q)], 1)
g_ = torch.cat([rx, rs], 1)
h_ = rz
H_LU = lu_hack(H_)
lu, pv=H_LU
#
# print(A_.size())
invH_A_ = A_.lu_solve(*H_LU).unsqueeze(1)#changed from A_.transpose(1, 2).lu_solve(*H_LU)
invH_g_ = g_.unsqueeze(2).lu_solve(*H_LU).squeeze(2)
A_=A_.unsqueeze(0).transpose(1,2)#added
S_ = torch.bmm(A_, invH_A_)
S_LU = lu_hack(S_)
slu, pv=S_LU
#print(A_.size(),invH_g_.unsqueeze(1).size(), torch.bmm(invH_g_.unsqueeze(1), A_).size(),h_.size())
t_ = torch.bmm(A_,invH_g_.unsqueeze(1)).squeeze(1) - h_#changed from torch.bmm(invH_g_.unsqueeze(1), A_.transpose(1, 2)).squeeze(1) - h_
w_ = -t_.lu_solve(*S_LU).squeeze(2)#changed
#print(g_.size(),w_.size(),A_.size())
t_ = -g_ - w_.bmm(A_).squeeze()
v_ = t_.unsqueeze(2).lu_solve(*H_LU).squeeze(2)
dx = v_[:, :nz]
ds = v_[:, nz:]
dz = w_[:, :nineq]
dy = w_[:, nineq:] if neq > 0 else None
return dx, ds, dz, dy
def pre_factor_kkt(Q, G, A):
""" Perform all one-time factorizations and cache relevant matrix products"""
nineq, nz, neq, nBatch = get_sizes(G, A)
Q_LU = Q.btrifact()
# S = [ A Q^{-1} A^T A Q^{-1} G^T ]
# [ G Q^{-1} A^T G Q^{-1} G^T + D^{-1} ]
#
# We compute a partial LU decomposition of the S matrix
# that can be completed once D^{-1} is known.
# See the 'Block LU factorization' part of our website
# for more details.
G_invQ_GT = torch.bmm(G, G.transpose(1, 2).btrisolve(*Q_LU))
R = G_invQ_GT.clone()
S_LU_pivots = torch.IntTensor(range(1, 1 + neq + nineq)).unsqueeze(0) \
.repeat(nBatch, 1).type_as(Q).int()
if neq > 0:
invQ_AT = A.transpose(1, 2).btrisolve(*Q_LU)
A_invQ_AT = torch.bmm(A, invQ_AT)
G_invQ_AT = torch.bmm(G, invQ_AT)
LU_A_invQ_AT = A_invQ_AT.btrifact()
P_A_invQ_AT, L_A_invQ_AT, U_A_invQ_AT = torch.btriunpack(*LU_A_invQ_AT)
P_A_invQ_AT = P_A_invQ_AT.type_as(A_invQ_AT)
S_LU_11 = LU_A_invQ_AT[0]
U_A_invQ_AT_inv = (P_A_invQ_AT.bmm(L_A_invQ_AT)).btrisolve(
*LU_A_invQ_AT)
S_LU_21 = G_invQ_AT.bmm(U_A_invQ_AT_inv)
T = G_invQ_AT.transpose(1, 2).btrisolve(*LU_A_invQ_AT)
S_LU_12 = U_A_invQ_AT.bmm(T)
S_LU_22 = torch.zeros(nBatch, nineq, nineq).type_as(Q)
S_LU_data = torch.cat((torch.cat(
(S_LU_11, S_LU_12), 2), torch.cat((S_LU_21, S_LU_22), 2)), 1)
S_LU_pivots[:, :neq] = LU_A_invQ_AT[1]
R -= G_invQ_AT.bmm(T)
else:
S_LU_data = torch.zeros(nBatch, nineq, nineq).type_as(Q)
S_LU = [S_LU_data, S_LU_pivots]
return Q_LU, S_LU, R
def factor_solve_kkt_reg(Q_tilde, D, G, A, rx, rs, rz, ry, eps):
nineq, nz, neq, nBatch = get_sizes(G, A)
H_ = torch.zeros(nBatch, nz + nineq, nz + nineq).type_as(Q_tilde)
H_[:, :nz, :nz] = Q_tilde
H_[:, -nineq:, -nineq:] = D
if neq > 0:
# H_ = torch.cat([torch.cat([Q, torch.zeros(nz,nineq).type_as(Q)], 1),
# torch.cat([torch.zeros(nineq, nz).type_as(Q), D], 1)], 0)
A_ = torch.cat([
torch.cat(
[G, torch.eye(nineq).type_as(Q_tilde).repeat(nBatch, 1, 1)],
2),
torch.cat([A, torch.zeros(nBatch, neq, nineq).type_as(Q_tilde)], 2)
], 1)
g_ = torch.cat([rx, rs], 1)
h_ = torch.cat([rz, ry], 1)
else:
A_ = torch.cat(
[G, torch.eye(nineq).type_as(Q_tilde).repeat(nBatch, 1, 1)], 2)
g_ = torch.cat([rx, rs], 1)
h_ = rz
H_LU = H_.btrifact()
invH_A_ = A_.transpose(1, 2).btrisolve(*H_LU)
invH_g_ = g_.btrisolve(*H_LU)
S_ = torch.bmm(A_, invH_A_)
S_ -= eps * torch.eye(neq + nineq).type_as(Q_tilde).repeat(nBatch, 1, 1)
S_LU = S_.btrifact()
t_ = torch.bmm(invH_g_.unsqueeze(1), A_.transpose(1, 2)).squeeze(1) - h_
w_ = -t_.btrisolve(*S_LU)
t_ = -g_ - w_.unsqueeze(1).bmm(A_).squeeze()
v_ = t_.btrisolve(*H_LU)
dx = v_[:, :nz]
ds = v_[:, nz:]
dz = w_[:, :nineq]
dy = w_[:, nineq:] if neq > 0 else None
return dx, ds, dz, dy
def forward_eq_new(Q_, p_, G_, h_, A_, b_, verbose=0, maxIter=100, dt=0.2):
""" Solves equality constraints by dynamically solving both priml and
dual problems side-by-side
"""
neq, nz, _, nBatch = get_sizes(A_)
Q__ = torch.cat((Q_, -Q_), dim=1)
Q = torch.cat((Q__, -Q__), dim=2)
p = torch.cat((p_, -p_), dim=1).unsqueeze(2)
A = torch.cat((A_, -A_), dim=2)
b = b_.unsqueeze(2)
Q_T = torch.transpose(Q, -1, 1)
A_T = torch.transpose(A, -1, 1)
p_T = torch.transpose(p, -1, 1)
b_T = torch.transpose(b, -1, 1)
# Expand 'x' to allow for negative values
x = torch.zeros(nBatch, 2 * nz, 1).type_as(Q).to(Q.device)
y = torch.zeros(nBatch, neq, 1).type_as(Q).to(Q.device)
for _ in range(maxIter):
x_T = torch.transpose(x, -1, 1)
g = x_T.bmm(Q).bmm(x) + p_T.bmm(x) - b_T.bmm(y)
r = neg(Q.bmm(x) + p - A_T.bmm(y))
dFx = g.bmm(2 * Q.bmm(x) + p)
dFy = g.bmm(b)
dx = -dFx - A_T.bmm(A.bmm(x) - b) - neg(x) - Q_T.bmm(r)
dy = dFy + A.bmm(r)
x.add_(dx)
y.add_(dy)
slacks = torch.zeros(nBatch, neq).type_as(Q).to(Q.device)
return (x[:, :nz, :] - x[:, nz:, :]).squeeze(2), y.squeeze(2), slacks
def forward(Q, p, G, h, A, b, verbose=0, maxIter=100, dt=0.2):
""" Solves a given QP problem by modeling it's dynamic representation
as an RNN with 'maxIter' loops.
"""
nineq, nz, neq, nBatch = get_sizes(G, A)
# Base inverses and transposes
Q_I = torch.inverse(Q)
G_T = torch.transpose(G, -1, 1)
A_T = torch.transpose(A, -1, 1)
# Intermediate matrix expressions
GQ_I = -G.bmm(Q_I) # - G Q^{-1}
AQ_I = -A.bmm(Q_I) # - A Q^{-1}
GA = GQ_I.bmm(A_T) # - G Q^{-1} A^T
GG = GQ_I.bmm(G_T) # - G Q^{-1} G^T
AA = AQ_I.bmm(A_T) # - A Q^{-1} A^T
AG = AQ_I.bmm(G_T) # - A Q^{-1} G^T
la_d = GQ_I.bmm(p.unsqueeze(2)) - h.unsqueeze(2) # - G Q^{-1} p - h
nu_d = AQ_I.bmm(p.unsqueeze(2)) - b.unsqueeze(2) # - A Q^{-1} p - b
lams = torch.zeros(nBatch, nineq, 1).type_as(Q).to(Q.device)
nus = torch.zeros(nBatch, neq, 1).type_as(Q).to(Q.device)
zeros = torch.zeros(nBatch, nineq, 1).type_as(Q).to(Q.device)
for _ in range(maxIter):
dlams = dt * (GG.bmm(lams) + GA.bmm(nus) + la_d)
dnus = dt * (AG.bmm(lams) + AA.bmm(nus) + nu_d)
dlams = torch.max(lams + dlams, zeros) - lams
lams.add_(dlams)
nus.add_(dnus)
zhat = -Q_I.bmm(p.unsqueeze(2) + G_T.bmm(lams) + A_T.bmm(nus))
slacks = h.unsqueeze(2) - G.bmm(zhat)
return zhat.squeeze(2), lams.squeeze(2), nus.squeeze(2), slacks.squeeze(2)
def forward(Q, p, G, h, A, b, Q_LU, S_LU, R, eps=1e-12, verbose=0, notImprovedLim=3,
maxIter=20, solver=KKTSolvers.LU_PARTIAL):
"""
Q_LU, S_LU, R = pre_factor_kkt(Q, G, A)
"""
nineq, nz, neq, nBatch = get_sizes(G, A)
# Find initial values
if solver == KKTSolvers.LU_FULL:
D = torch.eye(nineq).repeat(nBatch, 1, 1).type_as(Q)
x, s, z, y = factor_solve_kkt(
Q, D, G, A, p,
torch.zeros(nBatch, nineq).type_as(Q),
-h, -b if b is not None else None)
elif solver == KKTSolvers.LU_PARTIAL:
d = torch.ones(nBatch, nineq).type_as(Q)
factor_kkt(S_LU, R, d)
x, s, z, y = solve_kkt(
Q_LU, d, G, A, S_LU,
p, torch.zeros(nBatch, nineq).type_as(Q),
-h, -b if neq > 0 else None)
elif solver == KKTSolvers.IR_UNOPT:
D = torch.eye(nineq).repeat(nBatch, 1, 1).type_as(Q)
x, s, z, y = solve_kkt_ir(
Q, D, G, A, p,
torch.zeros(nBatch, nineq).type_as(Q),
-h, -b if b is not None else None)
else:
assert False
# Make all of the slack variables >= 1.
M = torch.min(s, 1)[0]
M = M.view(M.size(0), 1).repeat(1, nineq)
I = M < 0
s[I] -= M[I] - 1
# Make all of the inequality dual variables >= 1.
M = torch.min(z, 1)[0]
M = M.view(M.size(0), 1).repeat(1, nineq)
I = M < 0
z[I] -= M[I] - 1
best = {'resids': None, 'x': None, 'z': None, 's': None, 'y': None}
nNotImproved = 0
for i in range(maxIter):
# affine scaling direction
rx = (torch.bmm(y.unsqueeze(1), A).squeeze(1) if neq > 0 else 0.) + \
torch.bmm(z.unsqueeze(1), G).squeeze(1) + \
torch.bmm(x.unsqueeze(1), Q.transpose(1, 2)).squeeze(1) + \
p
rs = z
rz = torch.bmm(x.unsqueeze(1), G.transpose(1, 2)).squeeze(1) + s - h
ry = torch.bmm(x.unsqueeze(1), A.transpose(
1, 2)).squeeze(1) - b if neq > 0 else 0.0
mu = torch.abs((s * z).sum(1).squeeze() / nineq)
z_resid = torch.norm(rz, 2, 1).squeeze()
y_resid = torch.norm(ry, 2, 1).squeeze() if neq > 0 else 0
pri_resid = y_resid + z_resid
dual_resid = torch.norm(rx, 2, 1).squeeze()
resids = pri_resid + dual_resid + nineq * mu
d = z / s
try:
factor_kkt(S_LU, R, d)
except:
return best['x'], best['y'], best['z'], best['s']
if verbose == 1:
print('iter: {}, pri_resid: {:.5e}, dual_resid: {:.5e}, mu: {:.5e}'.format(
i, pri_resid.mean(), dual_resid.mean(), mu.mean()))
if best['resids'] is None:
best['resids'] = resids
best['x'] = x.clone()
best['z'] = z.clone()
best['s'] = s.clone()
best['y'] = y.clone() if y is not None else None
nNotImproved = 0
else:
I = resids < best['resids']
if I.sum() > 0:
nNotImproved = 0
else:
nNotImproved += 1
I_nz = I.repeat(nz, 1).t()
I_nineq = I.repeat(nineq, 1).t()
best['resids'][I] = resids[I]
best['x'][I_nz] = x[I_nz]
best['z'][I_nineq] = z[I_nineq]
best['s'][I_nineq] = s[I_nineq]
if neq > 0:
I_neq = I.repeat(neq, 1).t()
best['y'][I_neq] = y[I_neq]
if nNotImproved == notImprovedLim or best['resids'].max() < eps or mu.min() > 1e32:
if best['resids'].max() > 1. and verbose >= 0:
print(INACC_ERR)
return best['x'], best['y'], best['z'], best['s']
if solver == KKTSolvers.LU_FULL:
D = bdiag(d)
dx_aff, ds_aff, dz_aff, dy_aff = factor_solve_kkt(
Q, D, G, A, rx, rs, rz, ry)
elif solver == KKTSolvers.LU_PARTIAL:
dx_aff, ds_aff, dz_aff, dy_aff = solve_kkt(
Q_LU, d, G, A, S_LU, rx, rs, rz, ry)
elif solver == KKTSolvers.IR_UNOPT:
D = bdiag(d)
dx_aff, ds_aff, dz_aff, dy_aff = solve_kkt_ir(
Q, D, G, A, rx, rs, rz, ry)
else:
assert False
# compute centering directions
alpha = torch.min(torch.min(get_step(z, dz_aff),
get_step(s, ds_aff)),
torch.ones(nBatch).type_as(Q))
alpha_nineq = alpha.repeat(nineq, 1).t()
t1 = s + alpha_nineq * ds_aff
t2 = z + alpha_nineq * dz_aff
t3 = torch.sum(t1 * t2, 1).squeeze()
t4 = torch.sum(s * z, 1).squeeze()
sig = (t3 / t4)**3
rx = torch.zeros(nBatch, nz).type_as(Q)
rs = ((-mu * sig).repeat(nineq, 1).t() + ds_aff * dz_aff) / s
rz = torch.zeros(nBatch, nineq).type_as(Q)
ry = torch.zeros(nBatch, neq).type_as(Q) if neq > 0 else torch.Tensor()
if solver == KKTSolvers.LU_FULL:
D = bdiag(d)
dx_cor, ds_cor, dz_cor, dy_cor = factor_solve_kkt(
Q, D, G, A, rx, rs, rz, ry)
elif solver == KKTSolvers.LU_PARTIAL:
dx_cor, ds_cor, dz_cor, dy_cor = solve_kkt(
Q_LU, d, G, A, S_LU, rx, rs, rz, ry)
elif solver == KKTSolvers.IR_UNOPT:
D = bdiag(d)
dx_cor, ds_cor, dz_cor, dy_cor = solve_kkt_ir(
Q, D, G, A, rx, rs, rz, ry)
else:
assert False
dx = dx_aff + dx_cor
ds = ds_aff + ds_cor
dz = dz_aff + dz_cor
dy = dy_aff + dy_cor if neq > 0 else None
alpha = torch.min(0.999 * torch.min(get_step(z, dz),
get_step(s, ds)),
torch.ones(nBatch).type_as(Q))
alpha_nineq = alpha.repeat(nineq, 1).t()
alpha_neq = alpha.repeat(neq, 1).t() if neq > 0 else None
alpha_nz = alpha.repeat(nz, 1).t()
x += alpha_nz * dx
s += alpha_nineq * ds
z += alpha_nineq * dz
y = y + alpha_neq * dy if neq > 0 else None
if best['resids'].max() > 1. and verbose >= 0:
print(INACC_ERR)
return best['x'], best['y'], best['z'], best['s']
def forward(inputs, Q, G, h, A, b, Q_LU, S_LU, R, verbose=False):
"""
Q_LU, S_LU, R = pre_factor_kkt(Q, G, A)
"""
nineq, nz, neq, nBatch = get_sizes(G, A)
# Find initial values
d = torch.ones(nBatch, nineq).type_as(Q)
factor_kkt(S_LU, R, d)
x, s, z, y = solve_kkt(Q_LU, d, G, A, S_LU, inputs,
torch.zeros(nBatch, nineq).type_as(Q), -h,
-b if neq > 0 else None)
# D = torch.eye(nineq).repeat(nBatch, 1, 1).type_as(Q)
# x1, s1, z1, y1 = factor_solve_kkt(
# Q, D, G, A,
# inputs, torch.zeros(nBatch, nineq).type_as(Q),
# -h.repeat(nBatch, 1),
# nb.repeat(nBatch, 1) if b is not None else None)
# U_Q, U_S, R = pre_factor_kkt(Q, G, A)
# factor_kkt(U_S, R, d[0])
# x2, s2, z2, y2 = solve_kkt(
# U_Q, d[0], G, A, U_S,
# inputs[0], torch.zeros(nineq).type_as(Q), -h, nb)
# import IPython, sys; IPython.embed(); sys.exit(-1)
M = torch.min(s, 1)[0].repeat(1, nineq)
I = M < 0
s[I] -= M[I] - 1
M = torch.min(z, 1)[0].repeat(1, nineq)
I = M < 0
z[I] -= M[I] - 1
best = {'resids': None, 'x': None, 'z': None, 's': None, 'y': None}
nNotImproved = 0
for i in range(20):
# affine scaling direction
rx = (torch.bmm(y.unsqueeze(1), A).squeeze(1) if neq > 0 else 0.) + \
torch.bmm(z.unsqueeze(1), G).squeeze(1) + \
torch.bmm(x.unsqueeze(1), Q.transpose(1,2)).squeeze(1) + \
inputs
rs = z
rz = torch.bmm(x.unsqueeze(1), G.transpose(1, 2)).squeeze(1) + s - h
ry = torch.bmm(x.unsqueeze(1), A.transpose(
1, 2)).squeeze(1) - b if neq > 0 else 0.0
mu = torch.abs((s * z).sum(1).squeeze() / nineq)
z_resid = torch.norm(rz, 2, 1).squeeze()
y_resid = torch.norm(ry, 2, 1).squeeze() if neq > 0 else 0
pri_resid = y_resid + z_resid
dual_resid = torch.norm(rx, 2, 1).squeeze()
resids = pri_resid + dual_resid + nineq * mu
d = z / s
try:
factor_kkt(S_LU, R, d)
except:
# TODO: Move this below.
print('=' * 70 + '\n' +
"TODO: Remove try/except around factor_kkt!!!" + '\n')
return best['x'], best['y'], best['z'], best['s']
if verbose:
print(
'iter: {}, pri_resid: {:.5e}, dual_resid: {:.5e}, mu: {:.5e}'.
format(i, pri_resid[0], dual_resid[0], mu[0]))
if best['resids'] is None:
best['resids'] = resids
best['x'] = x.clone()
best['z'] = z.clone()
best['s'] = s.clone()
best['y'] = y.clone() if y is not None else None
nNotImproved = 0
else:
I = resids < best['resids']
if I.sum() > 0:
nNotImproved = 0
else:
nNotImproved += 1
I_nz = I.repeat(nz, 1).t()
I_nineq = I.repeat(nineq, 1).t()
best['resids'][I] = resids[I]
best['x'][I_nz] = x[I_nz]
best['z'][I_nineq] = z[I_nineq]
best['s'][I_nineq] = s[I_nineq]
if neq > 0:
I_neq = I.repeat(neq, 1).t()
best['y'][I_neq] = y[I_neq]
if nNotImproved == 3 or best['resids'].max() < 1e-12:
return best['x'], best['y'], best['z'], best['s']
# L_Q, L_S, R_ = pre_factor_kkt(Q, G, A)
# factor_kkt(L_S, R_, d[0])
# dx_cor, ds_cor, dz_cor, dy_cor = solve_kkt(
# L_Q, d[0], G, A, L_S, rx[0], rs[0], rz[0], ry[0])
# TODO: Move factorization back here.
dx_aff, ds_aff, dz_aff, dy_aff = solve_kkt(Q_LU, d, G, A, S_LU, rx, rs,
rz, ry)
# D = diaged(d)
# dx_aff1, ds_aff1, dz_aff1, dy_aff1 = factor_solve_kkt(
# Q, D, G, A, rx, rs, rz, ry)
# dx_aff2, ds_aff2, dz_aff2, dy_aff2 = factor_solve_kkt(
# Q, D[0], G, A, rx[0], rs[0], 0, ry[0])
# compute centering directions
# alpha0 = min(min(get_step(z[0],dz_aff[0]), get_step(s[0], ds_aff[0])), 1.0)
alpha = torch.min(torch.min(get_step(z, dz_aff), get_step(s, ds_aff)),
torch.ones(nBatch).type_as(Q))
alpha_nineq = alpha.repeat(nineq, 1).t()
# alpha_nz = alpha.repeat(nz, 1).t()
# sig0 = (torch.dot(s[0] + alpha[0]*ds_aff[0],
# z[0] + alpha[0]*dz_aff[0])/(torch.dot(s[0],z[0])))**3
t1 = s + alpha_nineq * ds_aff
t2 = z + alpha_nineq * dz_aff
t3 = torch.sum(t1 * t2, 1).squeeze()
t4 = torch.sum(s * z, 1).squeeze()
sig = (t3 / t4)**3
# dx_cor, ds_cor, dz_cor, dy_cor = solve_kkt(
# U_Q, d, G, A, U_S, torch.zeros(nz).type_as(Q),
# (-mu*sig*torch.ones(nineq).type_as(Q) + ds_aff*dz_aff)/s,
# torch.zeros(nineq).type_as(Q), torch.zeros(neq).type_as(Q), neq, nz)
# D = diaged(d)
# dx_cor0, ds_cor0, dz_cor0, dy_cor0 = factor_solve_kkt(Q, D[0], G, A,
# torch.zeros(nz).type_as(Q),
# (-mu[0]*sig[0]*torch.ones(nineq).type_as(Q)+ds_aff[0]*dz_aff[0])/s[0],
# torch.zeros(nineq).type_as(Q), torch.zeros(neq).type_as(Q))
rx = torch.zeros(nBatch, nz).type_as(Q)
rs = ((-mu * sig).repeat(nineq, 1).t() + ds_aff * dz_aff) / s
rz = torch.zeros(nBatch, nineq).type_as(Q)
ry = torch.zeros(nBatch, neq).type_as(Q)
# dx_cor1, ds_cor1, dz_cor1, dy_cor1 = factor_solve_kkt(
# Q, D, G, A, rx, rs, rz, ry)
dx_cor, ds_cor, dz_cor, dy_cor = solve_kkt(Q_LU, d, G, A, S_LU, rx, rs,
rz, ry)
dx = dx_aff + dx_cor
ds = ds_aff + ds_cor
dz = dz_aff + dz_cor
dy = dy_aff + dy_cor if neq > 0 else None
# import qpth.solvers.pdipm.single as pdipm_s
# alpha0 = min(1.0, 0.999*min(pdipm_s.get_step(s[0],ds[0]), pdipm_s.get_step(z[0],dz[0])))
alpha = torch.min(0.999 * torch.min(get_step(z, dz), get_step(s, ds)),
torch.ones(nBatch).type_as(Q))
# assert(np.isnan(alpha0) or np.isinf(alpha0) or alpha0 - alpha[0] <= 1e-5) # TODO: Remove
alpha_nineq = alpha.repeat(nineq, 1).t()
alpha_neq = alpha.repeat(neq, 1).t() if neq > 0 else None
alpha_nz = alpha.repeat(nz, 1).t()
dx_norm = torch.norm(dx, 2, 1).squeeze()
dz_norm = torch.norm(dz, 2, 1).squeeze()
# if TODO ->np.any(np.isnan(dx_norm)) or \
# torch.sum(dx_norm > 1e5) > 0 or \
# torch.sum(dz_norm > 1e5):
# # Overflow, return early
# return x, y, z
x += alpha_nz * dx
s += alpha_nineq * ds
z += alpha_nineq * dz
y = y + alpha_neq * dy if neq > 0 else None
return best['x'], best['y'], best['z'], best['s']
def forward(inputs_i, Q, G, A, b, h, U_Q, U_S, R, verbose=False):
"""
b = A z_0
h = G z_0 + s_0
U_Q, U_S, R = pre_factor_kkt(Q, G, A, nineq, neq)
"""
nineq, nz, neq, _ = get_sizes(G, A)
# find initial values
d = torch.ones(nineq).type_as(Q)
nb = -b if b is not None else None
factor_kkt(U_S, R, d)
x, s, z, y = solve_kkt(U_Q, d, G, A, U_S, inputs_i,
torch.zeros(nineq).type_as(Q), -h, nb)
# x1, s1, z1, y1 = factor_solve_kkt(Q, torch.eye(nineq).type_as(Q), G, A, inputs_i,
# torch.zeros(nineq).type_as(Q), -h, nb)
if torch.min(s) < 0:
s -= torch.min(s) - 1
if torch.min(z) < 0:
z -= torch.min(z) - 1
prev_resid = None
for i in range(20):
# affine scaling direction
rx = (torch.mv(A.t(), y) if neq > 0 else 0.) + \
torch.mv(G.t(), z) + torch.mv(Q, x) + inputs_i
rs = z
rz = torch.mv(G, x) + s - h
ry = torch.mv(A, x) - b if neq > 0 else torch.Tensor([0.])
mu = torch.dot(s, z) / nineq
pri_resid = torch.norm(ry) + torch.norm(rz)
dual_resid = torch.norm(rx)
resid = pri_resid + dual_resid + nineq * mu
d = z / s
if verbose:
print(("primal_res = {0:.5g}, dual_res = {1:.5g}, " +
"gap = {2:.5g}, kappa(d) = {3:.5g}").format(
pri_resid, dual_resid, mu,
min(d) / max(d)))
# if (pri_resid < 5e-4 and dual_resid < 5e-4 and mu < 4e-4):
improved = (prev_resid is None) or (resid < prev_resid + 1e-6)
if not improved or resid < 1e-6:
return x, y, z
prev_resid = resid
factor_kkt(U_S, R, d)
dx_aff, ds_aff, dz_aff, dy_aff = solve_kkt(U_Q, d, G, A, U_S, rx, rs,
rz, ry)
# D = torch.diag((z/s).cpu()).type_as(Q)
# dx_aff1, ds_aff1, dz_aff1, dy_aff1 = factor_solve_kkt(Q, D, G, A, rx, rs, rz, ry)
# compute centering directions
alpha = min(min(get_step(z, dz_aff), get_step(s, ds_aff)), 1.0)
sig = (torch.dot(s + alpha * ds_aff, z + alpha * dz_aff) /
(torch.dot(s, z)))**3
dx_cor, ds_cor, dz_cor, dy_cor = solve_kkt(
U_Q, d, G, A, U_S,
torch.zeros(nz).type_as(Q),
(-mu * sig * torch.ones(nineq).type_as(Q) + ds_aff * dz_aff) / s,
torch.zeros(nineq).type_as(Q),
torch.zeros(neq).type_as(Q))
# dx_cor, ds_cor, dz_cor, dy_cor = factor_solve_kkt(Q, D, G, A,
# torch.zeros(nz).type_as(Q),
# (-mu*sig*torch.ones(nineq).type_as(Q) + ds_aff*dz_aff)/s,
# torch.zeros(nineq).type_as(Q), torch.zeros(neq).type_as(Q))
dx = dx_aff + dx_cor
ds = ds_aff + ds_cor
dz = dz_aff + dz_cor
dy = dy_aff + dy_cor if neq > 0 else None
alpha = min(1.0, 0.999 * min(get_step(s, ds), get_step(z, dz)))
dx_norm = torch.norm(dx)
dz_norm = torch.norm(dz)
if np.isnan(dx_norm) or dx_norm > 1e5 or dz_norm > 1e5:
# Overflow, return early
return x, y, z
x += alpha * dx
s += alpha * ds
z += alpha * dz
y = y + alpha * dy if neq > 0 else None
return x, y, z
def pre_factor_kkt(Q, G, A):
""" Perform all one-time factorizations and cache relevant matrix products"""
nineq, nz, neq, nBatch = get_sizes(G, A)
try:
Q_LU = lu_hack(Q)
except:
raise RuntimeError("""
qpth Error: Cannot perform LU factorization on Q.
Please make sure that your Q matrix is PSD and has
a non-zero diagonal.
""")
# S = [ A Q^{-1} A^T A Q^{-1} G^T ]
# [ G Q^{-1} A^T G Q^{-1} G^T + D^{-1} ]
#
# We compute a partial LU decomposition of the S matrix
# that can be completed once D^{-1} is known.
# See the 'Block LU factorization' part of our website
# for more details.
G_invQ_GT = torch.bmm(G, G.transpose(1, 2).lu_solve(*Q_LU))
R = G_invQ_GT.clone()
S_LU_pivots = torch.IntTensor(range(1, 1 + neq + nineq)).unsqueeze(0) \
.repeat(nBatch, 1).type_as(Q).int()
if neq > 0:
invQ_AT = A.transpose(1, 2).lu_solve(*Q_LU)
# if any(torch.isnan(torch.flatten(invQ_AT)).tolist()):
# logging.info("nan comes in invq AT")
# else:
# logging.info("non NAN in invq AT")
A_invQ_AT = torch.bmm(A, invQ_AT)
G_invQ_AT = torch.bmm(G, invQ_AT)
# if any(torch.isnan(torch.flatten(G_invQ_AT)).tolist()):
# logging.info("nan comes in G_invQ_AT")
# else:
# logging.info("non NAN in G_invQ_AT")
LU_A_invQ_AT = lu_hack(A_invQ_AT)
P_A_invQ_AT, L_A_invQ_AT, U_A_invQ_AT = torch.lu_unpack(*LU_A_invQ_AT)
P_A_invQ_AT = P_A_invQ_AT.type_as(A_invQ_AT)
S_LU_11 = LU_A_invQ_AT[0]
U_A_invQ_AT_inv = (P_A_invQ_AT.bmm(L_A_invQ_AT)).lu_solve(
*LU_A_invQ_AT)
# if any(torch.isnan(torch.flatten(U_A_invQ_AT_inv)).tolist()):
# logging.info("nan in U_A_invQ_AT_inv")
S_LU_21 = G_invQ_AT.bmm(U_A_invQ_AT_inv)
T = sp_lu_solve(G_invQ_AT.transpose(1, 2), *LU_A_invQ_AT)
# T = G_invQ_AT.transpose(1, 2).lu_solve(*LU_A_invQ_AT)
S_LU_12 = U_A_invQ_AT.bmm(T)
S_LU_22 = torch.zeros(nBatch, nineq, nineq).type_as(Q)
S_LU_data = torch.cat((torch.cat(
(S_LU_11, S_LU_12), 2), torch.cat((S_LU_21, S_LU_22), 2)), 1)
S_LU_pivots[:, :neq] = LU_A_invQ_AT[1]
# if any(torch.isnan(torch.flatten(T)).tolist()):
# logging.info("nan comes in T")
# else:
# logging.info("non NAN in T")
R -= G_invQ_AT.bmm(T)
# if any(torch.isnan(torch.flatten(R)).tolist()):
# logging.info("nan is here")
# R[torch.isnan(R)] = 0
else:
S_LU_data = torch.zeros(nBatch, nineq, nineq).type_as(Q)
# S_LU_data[torch.isnan(S_LU_data)] = 0
S_LU = [S_LU_data, S_LU_pivots]
return Q_LU, S_LU, R
