Improve performance of hamming loss

corneto/backend/_base.py

@@ -511,7 +511,6 @@ def solve(
             if len(p.weights) != len(p.objectives):
                 raise ValueError("Number of weights must match number of objectives")
             # auto-convert to a weighted sum
-            # TODO: PICOS has issues with sum, change to matrix notation?
             # type: ignore
             o = sum(
                 p.weights[i] * p.objectives[i] if p.weights[i] != 0.0 else 0.0  # type: ignore
@@ -671,7 +670,12 @@ def _build_problem(self, other: ProblemDef) -> ProblemDef:
         idx_one = np.where(x == 1)[0]
         idx_zero = np.where(x == 0)[0]
         P = ProblemDef()
-        P.add_objectives((sum(y[idx_zero] - x[idx_zero]) + sum(x[idx_one] - y[idx_one])) * self.penalty, inplace=True)  # type: ignore
+        diff_zeros = y[idx_zero] - x[idx_zero]
+        diff_ones = x[idx_one] - y[idx_one]
+        hamming_dist = np.ones(diff_zeros.shape) @ diff_zeros + np.ones(diff_ones.shape) @ diff_ones
+        #P.add_objectives((sum(diff_zeros) + sum(diff_ones)) * self.penalty, inplace=True)  # type: ignore
+        P.add_objectives(hamming_dist, weights=self.penalty, inplace=True)  # type: ignore
+
         return P
 
 

corneto/methods/carnival.py

@@ -276,9 +276,9 @@ def carnival_constraints(
         # Constrain the product species of the reactions. They can be only up or
         # down if at least one of the reactions that have the node as product
         # carry some signal.
-        S_clipped = rn.stoichiometry.clip(0, 1)
-        p += N_act <= S_clipped @ R_act
-        p += N_inh <= S_clipped @ R_inh
+        incidence_matrix = rn.stoichiometry.clip(0, 1)
+        p += N_act <= incidence_matrix @ R_act
+        p += N_inh <= incidence_matrix @ R_inh
     return p
 
 
@@ -331,18 +331,12 @@ def carnival_loss(
             )
         # TODO: Issues with sum and PICOS (https://gitlab.com/picos-api/picos/-/issues/330)
         # override sum with picos.sum method
-        #losses.append(sum(Fi))  # type: ignore
-        #losses.append(Fi.T @ np.ones((1, Fi.shape[0])))
         losses.append(np.ones(Fi.shape) @ Fi)
         weights.append(l0_penalty_reaction)
     if l1_penalty_reaction > 0:
-        #losses.append(sum(F))  # type: ignore
-        #losses.append(F.T @ np.ones((1, F.shape[0])))
         losses.append(np.ones(F.shape) @ F)
         weights.append(l1_penalty_reaction)
     if l0_penalty_species > 0:
-        #losses.append(sum(N_act + N_inh))
-        #losses.append((N_act + N_inh).T @ np.ones((1, N_act.shape[0])))
         losses.append(np.ones(N_act.shape) @ (N_act + N_inh))
         weights.append(l0_penalty_species)
     # Add objective and weights to p