Skip to content

Issues with pytorch bindings in the case where structural_feasibility=False. #414

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
wants to merge 5 commits into
base: devel
Choose a base branch
from
Draft

Issues with pytorch bindings in the case where structural_feasibility=False. #414

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
5b3671e
Add a test for the pytorch interface
oomcth Aug 8, 2025
10de54e
Fix batched unfeasible QP
oomcth Aug 8, 2025
ccfc91a
Rename cvxpy.py to cvxpy_test.py
oomcth Aug 8, 2025
e736b9c
Add some checks for the gradients
oomcth Aug 8, 2025
0166ef0
Fix some bugs for the backward infeasible case
oomcth Aug 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
53 changes: 35 additions & 18 deletions bindings/python/proxsuite/torch/qplayer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -255,7 +255,9 @@ def backward(ctx, dl_dzhat, dl_dlams, dl_dnus):
class QPFunctionFn_infeas(Function):
@staticmethod
def forward(ctx, Q_, p_, A_, b_, G_, l_, u_):
n_in, nz = G_.size() # true double-sided inequality size

n_in, nz = G_.size()[-2], G_.size()[-1] # true double-sided inequality size
ctx.G_size = G_.size()
nBatch = extract_nBatch(Q_, p_, A_, b_, G_, l_, u_)

Q, _ = expandParam(Q_, nBatch, 3)
Expand Down Expand Up @@ -436,9 +438,9 @@ def backward(ctx, dl_dzhat, dl_dlams, dl_dnus, dl_ds_e, dl_ds_i):
if neq > 0:
kkt[:dim, dim : dim + n_eq] = A_i.transpose()
kkt[dim : dim + n_eq, :dim] = A_i
kkt[
dim + n_eq + n_in : dim + 2 * n_eq + n_in, dim : dim + n_eq
] = -np.eye(n_eq)
kkt[dim + n_eq + n_in : dim + 2 * n_eq + n_in, dim : dim + n_eq] = (
-np.eye(n_eq)
)
kkt[
dim + n_eq + n_in : dim + 2 * n_eq + n_in,
dim + n_eq + 2 * n_in : 2 * dim + n_eq + 2 * n_in,
Expand Down Expand Up @@ -485,9 +487,9 @@ def backward(ctx, dl_dzhat, dl_dlams, dl_dnus, dl_ds_e, dl_ds_i):
rhs[dim + n_eq : dim + n_eq + n_in_sol][~active_set] = dl_dnus[
i
][~active_set]
rhs[dim + n_eq + n_in_sol : dim + n_eq + n_in][
active_set
] = -dl_dnus[i][active_set]
rhs[dim + n_eq + n_in_sol : dim + n_eq + n_in][active_set] = (
-dl_dnus[i][active_set]
)
if dl_ds_e is not None:
if dl_ds_e.shape[0] != 0:
rhs[dim + n_eq + n_in : dim + 2 * n_eq + n_in] = -dl_ds_e[i]
Expand Down Expand Up @@ -542,8 +544,11 @@ def backward(ctx, dl_dzhat, dl_dlams, dl_dnus, dl_ds_e, dl_ds_i):
)
if n_eq > 0:
dlam[i] = torch.from_numpy(
np.float64(vector_of_qps.get(i).results.x[dim : dim + n_eq])
vector_of_qps.get(i)
.results.x[dim : dim + n_eq]
.astype(np.float64)
)

dnu[i] = torch.from_numpy(
np.float64(
vector_of_qps.get(i).results.x[dim + n_eq : dim + n_eq + n_in]
Expand Down Expand Up @@ -597,16 +602,28 @@ def backward(ctx, dl_dzhat, dl_dlams, dl_dnus, dl_ds_e, dl_ds_i):
if p_e:
dps = dps.mean(0)

grads = (
dQs,
dps,
dAs,
dbs,
dGs[n_in_sol:, :],
-dhs[:n_in_sol],
dhs[n_in_sol:],
)

if len(ctx.G_size) == 2:
grads = (
dQs,
dps,
dAs,
dbs,
dGs[n_in_sol:, :],
-dhs[:n_in_sol],
dhs[n_in_sol:],
)
elif len(ctx.G_size) == 3:
grads = (
dQs,
dps,
dAs,
dbs,
dGs[:, n_in_sol:, :],
-dhs[:, :n_in_sol],
dhs[:, n_in_sol:],
)
else:
raise
return grads

if structural_feasibility:
Expand Down
Loading
Loading