Merge pull request ilkarman#79 from ilkarman/ikdev

MXNet rerun

Author: ThomasDelteil

README.md

@@ -4,7 +4,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "# High-level Gluon Example"
+    "# MXNet/Gluon CNN example"
    ]
   },
   {
@@ -71,7 +71,7 @@
    "outputs": [],
    "source": [
     "def SymbolModule(n_classes=N_CLASSES):\n",
-    "    sym = gluon.nn.Sequential()\n",
+    "    sym = gluon.nn.HybridSequential()\n",
     "    with sym.name_scope():\n",
     "        sym.add(gluon.nn.Conv2D(channels=50, kernel_size=3, padding=1, activation='relu'))\n",
     "        sym.add(gluon.nn.Conv2D(channels=50, kernel_size=3, padding=1))\n",
@@ -121,8 +121,8 @@
       "Preparing test set...\n",
       "(50000, 3, 32, 32) (10000, 3, 32, 32) (50000,) (10000,)\n",
       "float32 float32 int32 int32\n",
-      "CPU times: user 630 ms, sys: 588 ms, total: 1.22 s\n",
-      "Wall time: 1.22 s\n"
+      "CPU times: user 708 ms, sys: 589 ms, total: 1.3 s\n",
+      "Wall time: 1.29 s\n"
      ]
     }
    ],
@@ -143,8 +143,8 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "CPU times: user 321 ms, sys: 392 ms, total: 713 ms\n",
-      "Wall time: 876 ms\n"
+      "CPU times: user 345 ms, sys: 421 ms, total: 766 ms\n",
+      "Wall time: 768 ms\n"
      ]
     }
    ],
@@ -164,8 +164,8 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "CPU times: user 203 µs, sys: 128 µs, total: 331 µs\n",
-      "Wall time: 337 µs\n"
+      "CPU times: user 683 µs, sys: 444 µs, total: 1.13 ms\n",
+      "Wall time: 406 µs\n"
      ]
     }
    ],
@@ -178,31 +178,49 @@
    "cell_type": "code",
    "execution_count": 9,
    "metadata": {},
+   "outputs": [],
+   "source": [
+    "train_loss = nd.zeros(1, ctx=ctx)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "train_loss += nd.ones(1, ctx=ctx)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Epoch   0: loss: 1.8405\n",
-      "Epoch   1: loss: 1.3773\n",
-      "Epoch   2: loss: 1.1577\n",
-      "Epoch   3: loss: 0.9811\n",
-      "Epoch   4: loss: 0.8450\n",
-      "Epoch   5: loss: 0.7354\n",
-      "Epoch   6: loss: 0.6391\n",
-      "Epoch   7: loss: 0.5559\n",
-      "Epoch   8: loss: 0.4810\n",
-      "Epoch   9: loss: 0.4157\n",
-      "CPU times: user 1min 18s, sys: 15.3 s, total: 1min 34s\n",
-      "Wall time: 1min 2s\n"
+      "Epoch   0: loss: 1.8314\n",
+      "Epoch   1: loss: 1.3397\n",
+      "Epoch   2: loss: 1.1221\n",
+      "Epoch   3: loss: 0.9576\n",
+      "Epoch   4: loss: 0.8261\n",
+      "Epoch   5: loss: 0.7215\n",
+      "Epoch   6: loss: 0.6226\n",
+      "Epoch   7: loss: 0.5389\n",
+      "Epoch   8: loss: 0.4729\n",
+      "Epoch   9: loss: 0.4072\n",
+      "CPU times: user 1min 5s, sys: 18 s, total: 1min 23s\n",
+      "Wall time: 56.6 s\n"
      ]
     }
    ],
    "source": [
     "%%time\n",
-    "# Main training loop: 62s\n",
+    "sym.hybridize()\n",
     "for j in range(EPOCHS):\n",
-    "    train_loss = 0.0\n",
+    "    train_loss = nd.zeros(1, ctx=ctx)\n",
     "    for data, target in yield_mb(x_train, y_train, BATCHSIZE, shuffle=True):\n",
     "        # Get samples\n",
     "        data = nd.array(data).as_in_context(ctx)\n",
@@ -215,22 +233,24 @@
     "        # Back-prop\n",
     "        loss.backward()\n",
     "        trainer.step(data.shape[0])\n",
-    "        train_loss += nd.sum(loss).asscalar()\n",
-    "    # Log\n",
-    "    print('Epoch %3d: loss: %5.4f'%(j, train_loss/len(x_train)))"
+    "        train_loss += nd.sum(loss)\n",
+    "    # Log    \n",
+    "    # Waiting for the operations on the     \n",
+    "    nd.waitall()\n",
+    "    print('Epoch %3d: loss: %5.4f'%(j, train_loss.asscalar()/len(x_train)))"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": 12,
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "CPU times: user 627 ms, sys: 73.1 ms, total: 700 ms\n",
-      "Wall time: 453 ms\n"
+      "CPU times: user 382 ms, sys: 115 ms, total: 496 ms\n",
+      "Wall time: 429 ms\n"
      ]
     }
    ],
@@ -254,14 +274,14 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": 13,
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Accuracy:  0.7661258012820513\n"
+      "Accuracy:  0.7675280448717948\n"
      ]
     }
    ],
@@ -273,7 +293,7 @@
  "metadata": {
   "anaconda-cloud": {},
   "kernelspec": {
-   "display_name": "Python 3",
+   "display_name": "Python [default]",
    "language": "python",
    "name": "python3"
   },