Skip to content

Fix hybrid decoder: remove det-ratio floor that killed high-dim models #67

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

Merged
iahncajigas merged 1 commit into from
Mar 18, 2026
Merged

Fix hybrid decoder: remove det-ratio floor that killed high-dim models #67

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
26 changes: 24 additions & 2 deletions nstat/decoding_algorithms.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2043,7 +2043,18 @@ def PPHybridFilterLinear(
X_u[model_index][:, time_index] = upd_x
W_u[model_index][:, :, time_index] = upd_W

det_ratio = np.sqrt(max(np.linalg.det(upd_W), 0.0)) / max(np.sqrt(max(np.linalg.det(pred_W), 0.0)), 1e-15)
# Likelihood via Laplace approximation (Srinivasan et al. 2007).
# Compute sqrt(det(W_u)) / sqrt(det(W_p)) as sqrt(det(W_u)/det(W_p))
# to avoid a fixed floor that destroys the ratio for
# high-dimensional models with tiny absolute determinants.
det_upd = np.linalg.det(upd_W)
det_pred = np.linalg.det(pred_W)
if det_pred > 0.0 and det_upd >= 0.0:
det_ratio = np.sqrt(det_upd / det_pred)
elif det_upd == 0.0 and det_pred == 0.0:
det_ratio = 1.0
else:
det_ratio = 0.0
log_term = np.sum(obs[:, time_index] * np.log(np.clip(lambda_delta.reshape(-1), 1e-12, np.inf)) - lambda_delta.reshape(-1))
likelihoods[model_index] = float(det_ratio * np.exp(np.clip(log_term, -200.0, 50.0)))

Expand Down Expand Up @@ -2212,7 +2223,18 @@ def PPHybridFilter(A, Q, p_ij, Mu0, dN, lambdaCIFColl, binwidth=0.001, x0=None,
X_u[model_index][:, time_index] = upd_x
W_u[model_index][:, :, time_index] = upd_W

det_ratio = np.sqrt(max(np.linalg.det(upd_W), 0.0)) / max(np.sqrt(max(np.linalg.det(pred_W), 0.0)), 1e-15)
# Likelihood via Laplace approximation (Srinivasan et al. 2007).
# Compute sqrt(det(W_u)) / sqrt(det(W_p)) as sqrt(det(W_u)/det(W_p))
# to avoid a fixed floor that destroys the ratio for
# high-dimensional models with tiny absolute determinants.
det_upd = np.linalg.det(upd_W)
det_pred = np.linalg.det(pred_W)
if det_pred > 0.0 and det_upd >= 0.0:
det_ratio = np.sqrt(det_upd / det_pred)
elif det_upd == 0.0 and det_pred == 0.0:
det_ratio = 1.0
else:
det_ratio = 0.0
log_term = np.sum(obs[:, time_index] * np.log(np.clip(lambda_delta.reshape(-1), 1e-12, np.inf)) - lambda_delta.reshape(-1))
likelihoods[model_index] = float(det_ratio * np.exp(np.clip(log_term, -200.0, 50.0)))

Expand Down
Loading