shuffle by default

optim_nn.py

@@ -910,11 +910,6 @@ def run(program, args=None):
         measure_error()
 
     while training and learner.next():
-        indices = np.arange(inputs.shape[0])
-        np.random.shuffle(indices)
-        shuffled_inputs = inputs[indices]
-        shuffled_outputs = outputs[indices]
-
         avg_loss, losses = ritual.train_batched(
             shuffled_inputs, shuffled_outputs,
             config.batch_size,

optim_nn_core.py

@@ -946,12 +946,18 @@ class Ritual: # i'm just making up names at this point
         return avg_mloss
 
     def train_batched(self, inputs, outputs, batch_size,
-                      return_losses=False, test_only=False):
+                      return_losses=False, test_only=False, shuffle=True):
         assert isinstance(return_losses, bool) or return_losses == 'both'
 
         if not test_only:
             self.en += 1
 
+        if shuffle:
+            indices = np.arange(inputs.shape[0])
+            np.random.shuffle(indices)
+            inputs = inputs[indices]
+            outputs = outputs[indices]
+
         cumsum_loss, cumsum_mloss = _0, _0
         batch_count = inputs.shape[0] // batch_size
         losses, mlosses = [], []

optim_nn_mnist.py

@@ -196,11 +196,6 @@ while learner.next():
             if isinstance(node, ActivityRegularizer):
                 node.reg.lamb = act_t * node.reg.lamb_orig # HACK
 
-    indices = np.arange(inputs.shape[0])
-    np.random.shuffle(indices)
-    shuffled_inputs = inputs[indices]
-    shuffled_outputs = outputs[indices]
-
     avg_loss, avg_mloss, losses, mlosses = ritual.train_batched(
         shuffled_inputs, shuffled_outputs,
         batch_size=bs,