release: prepare v1.0.0-beta.8

.gitignore

@@ -6,6 +6,8 @@ my_graph.pkl.gz
 tests/_notebooks/
 .asv/results/
 CLAUDE.md
+benchmarks/
+resources/
 
 
 ########################

.pre-commit-config.yaml

@@ -20,7 +20,7 @@ repos:
     -   id: pre-commit-update
 
 -   repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.15.0
+    rev: v0.15.17
     hooks:
     -   id: ruff
         args:
@@ -44,13 +44,13 @@ repos:
     -   id: check-yaml
 
 -   repo: https://github.com/compilerla/conventional-pre-commit
-    rev: v4.3.0
+    rev: v4.4.0
     hooks:
     -   id: conventional-pre-commit
         stages: [commit-msg]
 
 -   repo: https://github.com/python-poetry/poetry
-    rev: 2.3.2
+    rev: 2.4.1
     hooks:
     -   id: poetry-check
     -   id: poetry-lock

RELEASE.md

@@ -101,8 +101,24 @@ Useful options:
 ```bash
 poetry run release v1.0.0-beta.4 --dry-run  # validate only
 poetry run release v1.0.0-beta.4 --yes      # skip confirmation prompt
+poetry run release v1.0.0-beta.4 --remote public  # use an explicit release remote
 ```
 
+The remote defaults to `origin`, which is appropriate for a normal clone of
+the public CORNETO repository. In a private development workbench where
+`origin` points to a private mirror, configure a separate release-only remote
+for the public repository and pass it explicitly:
+
+```bash
+git remote add public https://github.com/saezlab/corneto.git
+poetry run release v1.0.0-beta.8 --remote public --dry-run
+poetry run release v1.0.0-beta.8 --remote public
+```
+
+Before tagging, publish selected private changes through a clean branch and a
+pull request into public `dev`, followed by a release pull request from public
+`dev` into public `main`. Never release a private development branch directly.
+
 ### Customizing Release Notes
 
 After the automated release is created, you can:
@@ -237,13 +253,6 @@ The version switcher uses a **single, centrally-updated file hosted on GitHub Pa
    - This ensures clean regeneration for a given version (no mixing of old/new files) while leaving other versions untouched.
    - The root redirect is deployed with `keep_files: true` so it doesn’t delete any versioned folders.
 
-#### Benefits of This Approach
-
-- **Central Management**: One `switcher.json` serves all deployed documentation versions
-- **Automatic Updates**: Old deployed docs automatically show new versions in the switcher dropdown
-- **No Regeneration Needed**: Previously deployed documentation doesn't need to be rebuilt to show new versions
-- **Consistency**: All documentation versions display the same version list
-
 #### Manual Switcher Updates
 
 If the automatic workflow fails or manual intervention is needed, re-run the docs workflow or regenerate and deploy the root artifacts (redirect + `switcher.json`) using the same workflow steps.

corneto/backend/_base.py

@@ -1048,8 +1048,8 @@ def Acyclic0(
             if In is not None:
                 # NOTE: Negative flows eq. to reversed directed edge
                 # Get edges s->t that can have a positive flow
-                if hasattr(In, "lb"):
-                    e_neg = [(i, g.get_edge(i)) for i in np.flatnonzero(In.lb < 0)]
+                if hasattr(In, "ub"):
+                    e_neg = [(i, g.get_edge(i)) for i in np.flatnonzero(In.ub > 0)]
                     e_ix = np.array([i for i, (s, t) in e_neg if len(s) > 0 and len(t) > 0])
                 else:
                     e_ix = np.array([i for i, (s, t) in enumerate(g.E) if len(s) > 0 and len(t) > 0])
@@ -1192,9 +1192,9 @@ def Acyclic(
 
             if In is not None:
                 # Negative flows are handled as reversed directed edges.
-                if hasattr(In, "lb"):
-                    lb = In.lb if len(In.shape) == 1 else In.lb[:, i_sample]
-                    e_neg = [(i, g.get_edge(i)) for i in np.flatnonzero(lb < 0)]
+                if hasattr(In, "ub"):
+                    ub = In.ub if len(In.shape) == 1 else In.ub[:, i_sample]
+                    e_neg = [(i, g.get_edge(i)) for i in np.flatnonzero(ub > 0)]
                     e_ix = np.array([i for i, (s, t) in e_neg if s and t])
                 else:
                     e_ix = np.array([i for i, (s, t) in enumerate(g.E) if s and t])
@@ -1351,9 +1351,18 @@ def Indicator(
         I = self.Variable(name, S.shape, 0, 1, vartype=VarType.BINARY)
         blocked = np.isclose(ub, 0) & np.isclose(lb, 0)
         if np.sum(blocked) > 0:
-            # indexing compatible with CVXPY and PICOS
-            idx = np.where(blocked)[0]
-            c += [I[idx] == 0, S[idx] == 0]
+            if blocked.ndim == 0:
+                c += [I == 0, S == 0]
+            elif blocked.ndim == 1:
+                idx = np.where(blocked)[0]
+                c += [I[idx] == 0, S[idx] == 0]
+            elif blocked.ndim == 2:
+                for col in range(blocked.shape[1]):
+                    idx = np.where(blocked[:, col])[0]
+                    if len(idx) > 0:
+                        c += [I[idx, col] == 0, S[idx, col] == 0]
+            else:
+                raise ValueError(f"Unsupported indicator bounds dimensionality: {blocked.ndim}")
         # Add constraint: lb * I <= V <= ub * I
         if V._provided_lb is None or V._provided_ub is None:
             raise ValueError(f"The continuous variable {V.name} is unbounded, indicators cannot be created.")
@@ -1410,9 +1419,33 @@ def NonZeroIndicator(
 
         # Disable infeasible binary indicators based on bounds
         if np.sum(ub <= 0) > 0:
-            c += [I_pos[np.where(ub <= 0)[0]] == 0]
+            mask = ub <= 0
+            if mask.ndim == 0:
+                c += [I_pos == 0]
+            elif mask.ndim == 1:
+                idx = np.where(mask)[0]
+                c += [I_pos[idx] == 0]
+            elif mask.ndim == 2:
+                for col in range(mask.shape[1]):
+                    idx = np.where(mask[:, col])[0]
+                    if len(idx) > 0:
+                        c += [I_pos[idx, col] == 0]
+            else:
+                raise ValueError(f"Unsupported NonZeroIndicator ub dimensionality: {mask.ndim}")
         if np.sum(lb >= 0) > 0:
-            c += [I_neg[np.where(lb >= 0)[0]] == 0]
+            mask = lb >= 0
+            if mask.ndim == 0:
+                c += [I_neg == 0]
+            elif mask.ndim == 1:
+                idx = np.where(mask)[0]
+                c += [I_neg[idx] == 0]
+            elif mask.ndim == 2:
+                for col in range(mask.shape[1]):
+                    idx = np.where(mask[:, col])[0]
+                    if len(idx) > 0:
+                        c += [I_neg[idx, col] == 0]
+            else:
+                raise ValueError(f"Unsupported NonZeroIndicator lb dimensionality: {mask.ndim}")
 
         # Add constraints to enforce variable behavior depending on the indicator activation:
         # If I_pos = 1 and I_neg = 0: V >= tol AND V <= ub