Filter recommendations

Filter already liked items from recommendations before selecting top@k.

implicit/evaluation.pyx

@@ -225,7 +225,7 @@ def mean_average_precision_at_k(model, train_user_items, test_user_items, int K=
 @cython.wraparound(False)
 @cython.nonecheck(False)
 def ALS_recommend_all(
-        model, users_items, int k=10, int threads=1, show_progress=True, recalculate_user=False):
+        model, users_items, int k=10, int threads=1, show_progress=True, recalculate_user=False, filter_already_liked_items=False):
 
     if not isinstance(users_items, csr_matrix):
         users_items = users_items.tocsr()
@@ -251,8 +251,10 @@ def ALS_recommend_all(
             model._user_factor(u, users_items, recalculate_user)
             for u
             in range(u_low, u_high, 1)
-        ])
-        A[:u_len] = users_factors.dot(factors_items)
+        ]).astype(np.float32)
+        users_factors.dot(factors_items, out=A[:u_len])
+        if filter_already_liked_items:
+            A[users_items[u_low:u_high].nonzero()] = 0
         for u in prange(u_len, nogil=True, num_threads=threads, schedule='dynamic'):
             fargsort_c(A_mv_p, u, batch * u_b + u, items_c, k, B_mv_p)
         progress.update(u_len)