clearn: cexact: properly initialize storage before learning

kamel-usp · Aug 29, 2023 · 4488c12 · 4488c12
1 parent 984f8e7
commit 4488c12
Show file tree

Hide file tree

Showing 4 changed files with 6 additions and 7 deletions.
diff --git a/pasp/cexact.c b/pasp/cexact.c
@@ -785,7 +785,7 @@ bool prob_obs(program_t *P, observations_t *obs, bool lstables_sat, prob_storage
     PyErr_SetString(PyExc_ValueError, "received NULL prob_storage_t as argument!");
     goto cleanup;
   }
-  if (!prob_obs_reuse(P, obs, lstables_sat, ret, Q, derive)) goto cleanup;
+  if (!prob_obs_reuse(P, obs, lstables_sat, ret, Q, derive, num_procs)) goto cleanup;
 
   return true;
 cleanup:
@@ -798,10 +798,9 @@ bool prob_obs(program_t *P, observations_t *obs, bool lstables_sat, prob_storage
 }
 
 bool prob_obs_reuse(program_t *P, observations_t *obs, bool lstable_sat, prob_storage_t *ret,
-    prob_storage_t Q[NUM_PROCS], bool derive) {
+    prob_storage_t Q[NUM_PROCS], bool derive, size_t num_procs) {
   total_choice_t theta;
   size_t total_choice_n = get_num_facts(P);
-  size_t num_procs = estimate_nprocs(total_choice_n + P->AD_n + P->NA_n);
   bool busy_procs[NUM_PROCS] = {0};
   storage_t S[NUM_PROCS] = {{0}};
   size_t i;

diff --git a/pasp/cexact.h b/pasp/cexact.h
@@ -19,6 +19,6 @@ bool prob_obs(program_t *P, observations_t *obs, bool lstables_sat, prob_storage
 /* Same as prob_obs, but reuse the prob_storage_t's in Q. It's memory safe to assign ret to &Q[0]
  * or NULL; the latter used if the user prefers to access data directly from Q. */
 bool prob_obs_reuse(program_t *P, observations_t *obs, bool lstable_sat, prob_storage_t *ret,
-    prob_storage_t Q[NUM_PROCS], bool derive);
+    prob_storage_t Q[NUM_PROCS], bool derive, size_t num_procs);
 
 #endif
diff --git a/pasp/clearn.c b/pasp/clearn.c
@@ -180,7 +180,7 @@ bool learn(program_t *P, PyArrayObject *obs, PyArrayObject *obs_counts,
     if (!forward_neural(P, &O)) goto cleanup;
 
     /* Compute probabilities. */
-    if (!prob_obs_reuse(P, &O, lstable_sat, NULL, Q, derive)) goto cleanup;
+    if (!prob_obs_reuse(P, &O, lstable_sat, NULL, Q, derive, num_procs)) goto cleanup;
 
     alg[which](P, &Q[0], N, eta, obs_counts, &O);
 
@@ -387,7 +387,7 @@ bool learn_batch(program_t *P, PyArrayObject *obs, size_t niters, double eta, si
       if (!forward_neural(P, &O)) goto cleanup;
 
       /* Compute probabilities. */
-      if (!prob_obs_reuse(P, &O, lstable_sat, NULL, Q, derive)) goto cleanup;
+      if (!prob_obs_reuse(P, &O, lstable_sat, NULL, Q, derive, num_procs)) goto cleanup;
 
       alg[which](P, &Q[0], &O, eta, smooth);
 

diff --git a/pasp/cstorage.c b/pasp/cstorage.c
@@ -14,7 +14,7 @@ bool prob_storage_learnable(prob_storage_t *S) { return S->n || S->m || S->pr ||
 
 size_t init_prob_storage_seq(prob_storage_t Q[NUM_PROCS], program_t *P, observations_t *O) {
   size_t total_choice_n = get_num_facts(P);
-  size_t num_procs = estimate_nprocs(total_choice_n + P->AD_n);
+  size_t num_procs = estimate_nprocs(total_choice_n + P->AD_n + P->NA_n);
   size_t i = 0;
 
   for (i = 0; i < num_procs; ++i) {