[feat] implement cross validation of softmax classifier #1

jeongjae96 · Feb 6, 2023 · 0da89ea · 0da89ea
1 parent fcc518f
commit 0da89ea
Showing 1 changed file with 68 additions and 9 deletions.
diff --git a/cs231n_2022/assignment1/softmax.ipynb b/cs231n_2022/assignment1/softmax.ipynb
@@ -279,10 +279,21 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 14,
    "id": "434e757f",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "naive loss: 2.386387e+00 computed in 0.042010s\n",
+      "vectorized loss: 2.386387e+00 computed in 0.004000s\n",
+      "Loss difference: 0.000000\n",
+      "Gradient difference: 0.000000\n"
+     ]
+    }
+   ],
    "source": [
     "# Now that we have a naive implementation of the softmax loss function and its gradient,\n",
     "# implement a vectorized version in softmax_loss_vectorized.\n",
@@ -308,15 +319,41 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 15,
    "id": "2e5e374d",
    "metadata": {
     "tags": [
      "code"
     ],
     "test": "tuning"
    },
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "lr 2.000000e-06 reg 1.000000e+03 train accuracy: 0.392286 val accuracy: 0.376000\n",
+      "lr 2.000000e-06 reg 1.000000e+04 train accuracy: 0.333816 val accuracy: 0.346000\n",
+      "lr 2.000000e-06 reg 2.000000e+04 train accuracy: 0.299286 val accuracy: 0.317000\n",
+      "lr 2.000000e-06 reg 2.500000e+04 train accuracy: 0.305898 val accuracy: 0.316000\n",
+      "lr 2.000000e-06 reg 3.000000e+04 train accuracy: 0.309490 val accuracy: 0.307000\n",
+      "lr 2.000000e-06 reg 5.000000e+04 train accuracy: 0.280694 val accuracy: 0.283000\n",
+      "lr 2.500000e-06 reg 1.000000e+03 train accuracy: 0.389408 val accuracy: 0.377000\n",
+      "lr 2.500000e-06 reg 1.000000e+04 train accuracy: 0.332571 val accuracy: 0.337000\n",
+      "lr 2.500000e-06 reg 2.000000e+04 train accuracy: 0.315347 val accuracy: 0.338000\n",
+      "lr 2.500000e-06 reg 2.500000e+04 train accuracy: 0.289531 val accuracy: 0.288000\n",
+      "lr 2.500000e-06 reg 3.000000e+04 train accuracy: 0.300184 val accuracy: 0.327000\n",
+      "lr 2.500000e-06 reg 5.000000e+04 train accuracy: 0.269163 val accuracy: 0.267000\n",
+      "lr 3.000000e-06 reg 1.000000e+03 train accuracy: 0.381061 val accuracy: 0.382000\n",
+      "lr 3.000000e-06 reg 1.000000e+04 train accuracy: 0.323408 val accuracy: 0.342000\n",
+      "lr 3.000000e-06 reg 2.000000e+04 train accuracy: 0.274347 val accuracy: 0.293000\n",
+      "lr 3.000000e-06 reg 2.500000e+04 train accuracy: 0.289388 val accuracy: 0.310000\n",
+      "lr 3.000000e-06 reg 3.000000e+04 train accuracy: 0.264939 val accuracy: 0.269000\n",
+      "lr 3.000000e-06 reg 5.000000e+04 train accuracy: 0.264510 val accuracy: 0.279000\n",
+      "best validation accuracy achieved during cross-validation: 0.382000\n"
+     ]
+    }
+   ],
    "source": [
     "# Use the validation set to tune hyperparameters (regularization strength and\n",
     "# learning rate). You should experiment with different ranges for the learning\n",
@@ -336,12 +373,26 @@
     "################################################################################\n",
     "\n",
     "# Provided as a reference. You may or may not want to change these hyperparameters\n",
-    "learning_rates = [1e-7, 5e-7]\n",
-    "regularization_strengths = [2.5e4, 5e4]\n",
+    "learning_rates = [2e-6, 2.5e-6, 3e-6]\n",
+    "regularization_strengths = [1e3, 1e4, 2e4, 2.5e4, 3e4, 5e4]\n",
     "\n",
     "# *****START OF YOUR CODE (DO NOT DELETE/MODIFY THIS LINE)*****\n",
     "\n",
-    "pass\n",
+    "for lr in learning_rates:\n",
+    "    for rs in regularization_strengths:\n",
+    "        softmax = Softmax()\n",
+    "        softmax.train(X_train, y_train, lr, rs, 1000)\n",
+    "\n",
+    "        y_train_pred = softmax.predict(X_train)\n",
+    "        y_val_pred = softmax.predict(X_val)\n",
+    "        train_accuracy = np.mean(y_train == y_train_pred)\n",
+    "        val_accuracy = np.mean(y_val == y_val_pred)\n",
+    "\n",
+    "        if val_accuracy > best_val:\n",
+    "            best_val = val_accuracy\n",
+    "            best_softmax = softmax\n",
+    "\n",
+    "        results[(lr, rs)] = train_accuracy, val_accuracy\n",
     "\n",
     "# *****END OF YOUR CODE (DO NOT DELETE/MODIFY THIS LINE)*****\n",
     "    \n",
@@ -356,12 +407,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 16,
    "id": "deb37cc6",
    "metadata": {
     "test": "test"
    },
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "softmax on raw pixels final test set accuracy: 0.373000\n"
+     ]
+    }
+   ],
    "source": [
     "# evaluate on test set\n",
     "# Evaluate the best softmax on test set\n",