diff --git a/bin/cluster_kmeans.py b/bin/cluster_kmeans.py
index 63e951a..230757b 100644
--- a/bin/cluster_kmeans.py
+++ b/bin/cluster_kmeans.py
@@ -7,6 +7,7 @@ import argparse
 import numpy as np
 from sklearn.cluster import KMeans
 from os import path
+from os import mkdir
 
 import pickle
 from data import read_file, index_by_id
@@ -18,6 +19,8 @@ parser.add_argument("list", type=str, help="List on which apply kmeans")
 parser.add_argument("outdir", type=str, help="Output directory for k-means models")
 parser.add_argument("--kmin", type=int, help="minimum k", default=2)
 parser.add_argument("--kmax", type=int, help="maximum k", default=100)
+parser.add_argument("--allindir", type=bool, default=False,
+                    help="all in same dir or separed ?")
 
 args = vars(parser.parse_args())
 FEATURES = args["features"]
@@ -25,6 +28,7 @@ LST = args["list"]
 OUTDIR = args["outdir"]
 KMIN = args["kmin"]
 KMAX = args["kmax"]
+ALLINDIR = args["allindir"]
 
 # -- READE FILES --
 features = read_file(FEATURES)
@@ -32,11 +36,15 @@ feat_ind = index_by_id(features)
 
 lst = read_file(LST)
 
+subdir = ""
 # -- TRANSFORM INTO NUMPY --
 X = np.asarray([feat_ind[x[0][0]][x[0][3]][1] for x in lst])
-
 Ks = range(KMIN, KMAX+1)
 for k in Ks:
     kmeans = KMeans(n_clusters=k, n_init=10, random_state=0).fit(X)
-    pickle.dump(kmeans, open(path.join(OUTDIR, "clustering_" + str(k) + ".pkl"), "wb"))
-
+    if ALLINDIR is False:
+        subdir = str(k)
+        dirname=path.join(OUTDIR, subdir)
+        if not path.exists(dirname):
+            mkdir(dirname)
+    pickle.dump(kmeans, open(path.join(OUTDIR, subdir, "clustering_" + str(k) + ".pkl"), "wb"))