fixed set_weight_decay

CHANGELOG.md

@@ -6,6 +6,18 @@ Most recent releases are shown at the top. Each release shows:
 - **Changed**: Additional parameters, changes to inputs or outputs, etc
 - **Fixed**: Bug fixes that don't change documented behaviour
 
+## 0.1.8 (2019-06-04)
+
+### New:
+- N/A
+
+### Changed:
+- N/A
+
+### Fixed:
+- ```Learner.set_weight_decay``` now works correctly
+
+
 ## 0.1.7 (2019-05-24)
 
 ### New:

ktrain/core.py

@@ -152,8 +152,11 @@ def set_weight_decay(self, wd=0.005):
         for layer in self.model.layers:
             if hasattr(layer, 'kernel_regularizer'):
                 layer.kernel_regularizer= regularizers.l2(wd)
+                layer.add_loss(regularizers.l2(wd)(layer.kernel))
             if hasattr(layer, 'bias_regularizer'):
                 layer.bias_regularizer= regularizers.l2(wd)
+                layer.add_loss(regularizers.l2(wd)(layer.bias))
+        self._recompile()
         return
 
 

ktrain/version.py

@@ -1,2 +1,2 @@
 __all__ = ['__version__']
-__version__ = '0.1.7'
+__version__ = '0.1.8'

tutorial-A1-additional-tricks.ipynb

@@ -99,9 +99,9 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "0 (trainable=True) : <keras.layers.embeddings.Embedding object at 0x7fef21ba1630>\n",
-      "1 (trainable=True) : <keras.layers.pooling.GlobalAveragePooling1D object at 0x7fef615d2978>\n",
-      "2 (trainable=True) : <keras.layers.core.Dense object at 0x7fef21ba1a20>\n"
+      "0 (trainable=True) : <keras.layers.embeddings.Embedding object at 0x7f77fa4adcf8>\n",
+      "1 (trainable=True) : <keras.layers.pooling.GlobalAveragePooling1D object at 0x7f77fa4adfd0>\n",
+      "2 (trainable=True) : <keras.layers.core.Dense object at 0x7f783cc27fd0>\n"
      ]
     }
    ],
@@ -198,9 +198,9 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "0 (trainable=True) : <keras.layers.embeddings.Embedding object at 0x7ff69cc4d9b0>\n",
-      "1 (trainable=True) : <keras.layers.pooling.GlobalAveragePooling1D object at 0x7ff69cc1c128>\n",
-      "2 (trainable=True, wd=None) : <keras.layers.core.Dense object at 0x7ff69cc1c160>\n"
+      "0 (trainable=True) : <keras.layers.embeddings.Embedding object at 0x7f65e8e2bd68>\n",
+      "1 (trainable=True) : <keras.layers.pooling.GlobalAveragePooling1D object at 0x7f65e8e2beb8>\n",
+      "2 (trainable=True, wd=None) : <keras.layers.core.Dense object at 0x7f65e8e2be80>\n"
      ]
     }
    ],
@@ -214,7 +214,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "learner.set_weight_decay(0.1)"
+    "learner.set_weight_decay(1e-3)"
    ]
   },
   {
@@ -225,7 +225,7 @@
     {
      "data": {
       "text/plain": [
-       "[array(0.1, dtype=float32)]"
+       "[array(0.001, dtype=float32)]"
       ]
      },
      "execution_count": 7,
@@ -246,9 +246,9 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "0 (trainable=True) : <keras.layers.embeddings.Embedding object at 0x7ff69cc4d9b0>\n",
-      "1 (trainable=True) : <keras.layers.pooling.GlobalAveragePooling1D object at 0x7ff69cc1c128>\n",
-      "2 (trainable=True, wd=0.1) : <keras.layers.core.Dense object at 0x7ff69cc1c160>\n"
+      "0 (trainable=True) : <keras.layers.embeddings.Embedding object at 0x7f65e8e2bd68>\n",
+      "1 (trainable=True) : <keras.layers.pooling.GlobalAveragePooling1D object at 0x7f65e8e2beb8>\n",
+      "2 (trainable=True, wd=0.001) : <keras.layers.core.Dense object at 0x7f65e8e2be80>\n"
      ]
     }
    ],
@@ -258,7 +258,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": 9,
    "metadata": {},
    "outputs": [
     {
@@ -267,30 +267,35 @@
      "text": [
       "\n",
       "\n",
-      "begin training using onecycle policy with max lr of 0.005...\n",
+      "begin training using triangular learning rate policy with max lr of 0.005...\n",
       "Train on 25000 samples, validate on 25000 samples\n",
-      "Epoch 1/3\n",
-      "25000/25000 [==============================] - 6s 251us/step - loss: 0.5552 - acc: 0.7462 - val_loss: 0.3613 - val_acc: 0.8609\n",
-      "Epoch 2/3\n",
-      "25000/25000 [==============================] - 6s 244us/step - loss: 0.2686 - acc: 0.8970 - val_loss: 0.2907 - val_acc: 0.8841\n",
-      "Epoch 3/3\n",
-      "25000/25000 [==============================] - 6s 246us/step - loss: 0.1738 - acc: 0.9406 - val_loss: 0.2780 - val_acc: 0.8902\n"
+      "Epoch 1/6\n",
+      "25000/25000 [==============================] - 7s 270us/step - loss: 0.5387 - acc: 0.7776 - val_loss: 0.4369 - val_acc: 0.8681\n",
+      "Epoch 2/6\n",
+      "25000/25000 [==============================] - 6s 255us/step - loss: 0.3808 - acc: 0.8934 - val_loss: 0.3947 - val_acc: 0.8810\n",
+      "Epoch 3/6\n",
+      "25000/25000 [==============================] - 6s 254us/step - loss: 0.3363 - acc: 0.9124 - val_loss: 0.3755 - val_acc: 0.8861\n",
+      "Epoch 4/6\n",
+      "25000/25000 [==============================] - 6s 254us/step - loss: 0.3081 - acc: 0.9220 - val_loss: 0.3623 - val_acc: 0.8881\n",
+      "Epoch 5/6\n",
+      "25000/25000 [==============================] - 6s 257us/step - loss: 0.2867 - acc: 0.9319 - val_loss: 0.3539 - val_acc: 0.8891\n",
+      "Epoch 6/6\n",
+      "25000/25000 [==============================] - 6s 257us/step - loss: 0.2703 - acc: 0.9363 - val_loss: 0.3469 - val_acc: 0.8905\n"
      ]
     },
     {
      "data": {
       "text/plain": [
-       "<keras.callbacks.History at 0x7ff697cf0550>"
+       "<keras.callbacks.History at 0x7f65df4c8278>"
       ]
      },
-     "execution_count": 10,
+     "execution_count": 9,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
-    "learner.set_model(get_model())\n",
-    "learner.fit_onecycle(0.005, 3)"
+    "learner.autofit(0.005, 6)"
    ]
   },
   {