new parameter added to choose between parallelization algorithms, but…

… more testing needed

.gitignore

@@ -0,0 +1 @@
+*.pyc

BNfinder/data.py

@@ -629,7 +629,7 @@ def get_prior(self,vertex,parent):
         return self.prior.get((vertex.name,parent.name),self.prior.get(parent.name,1))
 
 
-    def learn(self,score,data_factor, prior=None,cores=False, subset=False, \
+    def learn(self,score,data_factor,alg,prior=None,cores=False,subset=False, \
             verbose=None,n_min=1,limit=None,min_empty=None,min_optim=None,fpr=None,max_tries=None):
 
         scr=score
@@ -689,8 +689,14 @@ class MyPool(multiprocessing.pool.Pool):
                 # In case of limit<=3 it is efficient to use hybrid alg
                 # There are two situations: number of cores >= number of vertices and opposite one.
                 # So, we need to handle it separately to distritube cores equaly
+                lim = 0
+                if (alg=="set"):
+                    lim = 0
+                if (alg=="variable"):
+                    lim = 999
+
                 pool=MyPool(1)
-                if (limit is not None) and (int(limit)<=3) and (cores>int(limit)):
+                if (limit is not None) and (int(limit)<=lim) and (cores>int(limit)):
                     if (cores>=len(vertices)):
                         pool=MyPool(len(vertices))
                         for counter in range(1, len(vertices)+1):
@@ -726,7 +732,7 @@ class MyPool(multiprocessing.pool.Pool):
                 # Need to check if we just want to concatenate previously obtained results
                 if (subset==False) or (subset != "concat"):
                     # Hybrid alg
-                    if (limit is not None) and (int(limit)<=3) and (cores>int(limit)):
+                    if (limit is not None) and (int(limit)<=lim) and (cores>int(limit)):
                         result=pool.map(learn_x,[(x,self,min_empty,min_optim,scr,verbose,n_min,limit,fpr_factor,max_tries, y) for x, y in zip(vertices, distribs)])
                     # Simple alg    
                     else:

bnf

@@ -32,6 +32,7 @@ if __name__=="__main__":
     parser.add_option("-g", "--sloops", dest="sloops", action='store_true', default=False, help="Allow self-loops in Dynamic Bayesian Networks (no self-loops by default)")
     parser.add_option("-k", "--cpu", dest="cores", type=int, default=0, help="Number of cores you want to use for parallelization")
     parser.add_option("-j", "--subset", dest="subset", help="Specify the file with subset of genes (divided by space character), parents of which you want to find. Use 'concat' value to combine the results. BNFinder will automaticaly search for files 'genes_*'")
+    parser.add_option("-q", "--algorithm", dest="alg", default="set", help="Choose the parralelization algorithm:'set' (default, give all the cores to compute parents set of one variable) or 'variable' (evenly distributes cores between variables)")
 
     (options, args) = parser.parse_args()
 
@@ -65,6 +66,7 @@ if __name__=="__main__":
                         min_empty=options.min_empty,\
                         min_optim=options.min_optim,\
                         cores=options.cores,\
+                        alg=options.alg,\
                         subset=options.subset,\
                         fpr=options.fpr,
                         max_tries=options.max_tries,)