update tests

Author: a-martyn

Diff for: model/test_net.py

@@ -10,25 +10,6 @@
 class Tests(unittest.TestCase):
 
 
-    def test_weight_dropout(self):
-        # Instantiate an LSTM net with weight dropout on 
-        # hidden-to-hidden weights
-        model = net.AWD_LSTM(10, 20, 2)
-
-        # Apply drop connect with probabilit 1.0
-        model.weight_dropout(p=1.0) 
-        new_state = model.state_dict()
-        
-        # Test
-        # Because dropout probability is 1
-        # all hidden-to-hidden weights should be zero
-        # There are two such weight matrices because we
-        # configured a 2-layer lstm
-        self.assertTrue(new_state['layer0.h2h.weight'].sum() == 0)
-        self.assertTrue(new_state['layer1.h2h.weight'].sum() == 0)
-        # Here we test that input-to-hidden weights are unaffected
-        self.assertTrue(new_state['layer0.i2h.weight'].sum() != 0)
-        self.assertTrue(new_state['layer1.i2h.weight'].sum() != 0)
 
     def test_activation_reg(self):
         # make deterministe

Diff for: test_utils.py

@@ -19,25 +19,25 @@ def test_NT_ASGD(self):
         # ASGD used in place of SGD optimizer when
         # loss increases for n succesive calls to get_optimizer 
         self.assertEqual(nt_asgd.asgd_triggered, False)
-        nt_asgd.get_optimizer(3, model.parameters())
+        nt_asgd.get_optimizer(3)
         self.assertEqual(nt_asgd.asgd_triggered, False)
-        nt_asgd.get_optimizer(2, model.parameters())
+        nt_asgd.get_optimizer(2)
         self.assertEqual(nt_asgd.asgd_triggered, False)
-        nt_asgd.get_optimizer(3, model.parameters())
+        nt_asgd.get_optimizer(3)
         self.assertEqual(nt_asgd.asgd_triggered, False)
-        nt_asgd.get_optimizer(3, model.parameters())
+        nt_asgd.get_optimizer(3)
         self.assertEqual(nt_asgd.asgd_triggered, False)
         # ASGD Triggered because loss was lowest n+1 epochs ago
-        nt_asgd.get_optimizer(4, model.parameters())
+        nt_asgd.get_optimizer(4)
         self.assertEqual(nt_asgd.asgd_triggered, True)
-        nt_asgd.get_optimizer(2, model.parameters())
+        nt_asgd.get_optimizer(2)
         self.assertEqual(nt_asgd.asgd_triggered, True)
-        nt_asgd.get_optimizer(3, model.parameters())
+        nt_asgd.get_optimizer(3)
         self.assertEqual(nt_asgd.asgd_triggered, True)
-        nt_asgd.get_optimizer(3, model.parameters())
+        nt_asgd.get_optimizer(3)
         self.assertEqual(nt_asgd.asgd_triggered, True) 
         # Doesn't un-trigger
-        nt_asgd.get_optimizer(4, model.parameters())
+        nt_asgd.get_optimizer(4)
         self.assertEqual(nt_asgd.asgd_triggered, True)