diff --git a/onn.py b/onn.py
index c33397f..e429edc 100755
--- a/onn.py
+++ b/onn.py
@@ -1153,8 +1153,8 @@ def run(program, args=None):
         #optim_decay1 = 2,
         #optim_decay2 = 100,
         optim = 'adam', # note: most features only implemented for Adam
-        optim_decay1 = 24,  #  first momentum given in epochs (optional)
-        optim_decay2 = 100, # second momentum given in epochs (optional)
+        optim_decay1 = 24,  #  first momentum given in batches (optional)
+        optim_decay2 = 100, # second momentum given in batches (optional)
         nesterov = True, # not available for all optimizers.
         batch_size = 64,
 
diff --git a/onn_core.py b/onn_core.py
index 0c2f2bf..bdbfc9a 100644
--- a/onn_core.py
+++ b/onn_core.py
@@ -322,7 +322,7 @@ class RMSprop(Optimizer):
         # mu = e**(-1/t)
         # default: t = -1/ln(0.99) = ~99.5
         # therefore the default of mu=0.99 means
-        # an input decays to 1/e its original amplitude over 99.5 epochs.
+        # an input decays to 1/e its original amplitude over 99.5 batches.
         # (this is from DSP, so how relevant it is in SGD is debatable)
 
         super().__init__(lr)