Add code comments

Author: abdullahkady

perceptron.ipynb

@@ -44,6 +44,7 @@
    "outputs": [],
    "source": [
     "def initialize_weights_matrix():\n",
+    "    # 10 weight vectors (classes), each one is all zeros except for the first feature.\n",
     "    weight_matrix =  np.zeros((785, 10))\n",
     "    weight_matrix[0] = np.ones(10)\n",
     "    return weight_matrix.T"
@@ -90,8 +91,7 @@
    "source": [
     "# Loading in a separate cell to avoid multiple loads.\n",
     "x_delta = get_x_delta('Train', 2400)\n",
-    "training_true_labels = get_t('Train/Training Labels.txt')\n",
-    "T = get_T_matrix(training_true_labels)"
+    "T = get_T_matrix(get_t('Train/Training Labels.txt'))"
    ]
   },
   {
@@ -100,17 +100,20 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "learning_rates = [1, 10 ** -1, 10 ** -2, 10 ** -3, 10 ** -4, 10 ** -5, 10 ** -6, 10 ** -7, 10 ** -8, 10 ** -9]\n",
+    "# Train all learning rates for 500 epochs.\n",
+    "learning_rates_powers = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]\n",
     "resulting_weights = []\n",
-    "for eta in learning_rates:\n",
-    "    weights_matrix = initialize_weights_matrix()\n",
-    "    for class_number in range(0,10):\n",
-    "        for epoch in range(0,500):\n",
-    "            for img in range(0, 2400):\n",
-    "                r = 1 if np.dot(weights_matrix[class_number], x_delta[img]) >=0 else -1\n",
-    "                if r != T[:,img][class_number]:\n",
-    "                    weights_matrix[class_number] += eta * (np.multiply(x_delta[img], T[:,img][class_number]))\n",
-    "    resulting_weights += [(eta, weights_matrix)]"
+    "\n",
+    "for e_p in learning_rates_powers: # Eta Power \"e_p\"\n",
+    "    w_m = initialize_weights_matrix()\n",
+    "    for c in range(0,10): # Class \"c\"\n",
+    "        for _ in range(0,500): # Iteration/Epoch\n",
+    "            for i in range(0, 2400): # Image \"i\"\n",
+    "                r = 1 if np.dot(w_m[c], x_delta[i]) >=0 else -1\n",
+    "                if r != T[:,i][c]:\n",
+    "                    # Update weight-vector if image \"i\" is misclassified\n",
+    "                    w_m[c] += (10 ** (-1 * e_p)) * (np.multiply(x_delta[i], T[:,i][c]))\n",
+    "    resulting_weights += [(e_p, w_m)]"
    ]
   },
   {
@@ -121,10 +124,10 @@
    "source": [
     "x_delta_test = get_x_delta('Test', 200)\n",
     "test_true_labels = get_t('Test/Test Labels.txt')\n",
-    "for eta, weights_matrix in resulting_weights:\n",
-    "    conf_matrix = get_confusion_matrix(x_delta_test, weights_matrix.T, test_true_labels)\n",
-    "    plt.imshow(conf_matrix)\n",
-    "    plt.savefig('Confusion{}.jpg'.format(eta))"
+    "for eta_power, weights_matrix in resulting_weights:\n",
+    "    confusion_matrix = get_confusion_matrix(x_delta_test, weights_matrix.T, test_true_labels)\n",
+    "    plt.imshow(confusion_matrix)\n",
+    "    plt.savefig('Confusion-{}.jpg'.format(eta_power))"
    ]
   }
  ],