20230329 dropout .5 public 0.951

ML2023_hw04.ipynb

@@ -68,7 +68,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": 1,
    "metadata": {
     "id": "E6burzCXIyuA"
    },
@@ -131,7 +131,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 2,
    "metadata": {
     "id": "KpuGxl4CI2pr"
    },
@@ -160,7 +160,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": 3,
    "metadata": {
     "id": "B7c2gZYoJDRS"
    },
@@ -236,7 +236,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 4,
    "metadata": {
     "id": "iXZ5B0EKJGs8"
    },
@@ -248,7 +248,7 @@
     "import torchaudio\n",
     "\n",
     "class Classifier(nn.Module):\n",
-    "    def __init__(self, d_model=160, n_spks=600, dropout=0.25):\n",
+    "    def __init__(self, d_model=160, n_spks=600, dropout=0.5):\n",
     "        super().__init__()\n",
     "        # Project the dimension of features from that of input into d_model.\n",
     "        self.prenet = nn.Linear(40, d_model)\n",
@@ -308,7 +308,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": 5,
    "metadata": {
     "id": "ykt0N1nVJJi2"
    },
@@ -485,58 +485,16 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "Train: 100% 2000/2000 [06:35<00:00,  5.05 step/s, accuracy=0.28, loss=3.44, step=2000]  \n",
-      "Valid: 100% 5664/5667 [00:56<00:00, 100.29 uttr/s, accuracy=0.25, loss=3.53]\n",
-      "Train: 100% 2000/2000 [06:25<00:00,  5.19 step/s, accuracy=0.53, loss=2.23, step=4000]  \n",
-      "Valid: 100% 5664/5667 [00:58<00:00, 96.06 uttr/s, accuracy=0.44, loss=2.51] \n",
-      "Train: 100% 2000/2000 [06:37<00:00,  5.04 step/s, accuracy=0.59, loss=1.77, step=6000] \n",
-      "Valid: 100% 5664/5667 [00:59<00:00, 95.68 uttr/s, accuracy=0.53, loss=2.01] \n",
-      "Train: 100% 2000/2000 [07:21<00:00,  4.53 step/s, accuracy=0.66, loss=1.39, step=8000] \n",
-      "Valid: 100% 5664/5667 [00:55<00:00, 101.48 uttr/s, accuracy=0.63, loss=1.60]\n",
-      "Train:   0% 0/2000 [00:00<?, ? step/s]"
-     ]
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Step 8000, best model saved. (accuracy=0.6324)\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "Train: 100% 2000/2000 [06:26<00:00,  5.17 step/s, accuracy=0.78, loss=0.82, step=1e+4]  \n",
-      "Valid: 100% 5664/5667 [00:56<00:00, 99.78 uttr/s, accuracy=0.65, loss=1.50] \n",
-      "Train: 100% 2000/2000 [05:41<00:00,  5.85 step/s, accuracy=0.75, loss=1.21, step=12000]  \n",
-      "Valid: 100% 5664/5667 [00:56<00:00, 100.25 uttr/s, accuracy=0.69, loss=1.33]\n",
-      "Train: 100% 2000/2000 [06:58<00:00,  4.78 step/s, accuracy=0.66, loss=1.26, step=14000] \n",
-      "Valid: 100% 5664/5667 [01:02<00:00, 90.35 uttr/s, accuracy=0.71, loss=1.22] \n",
-      "Train: 100% 2000/2000 [07:45<00:00,  4.29 step/s, accuracy=0.75, loss=1.04, step=16000] \n",
-      "Valid: 100% 5664/5667 [00:55<00:00, 102.93 uttr/s, accuracy=0.72, loss=1.19]\n",
-      "Train:   0% 0/2000 [00:00<?, ? step/s]"
-     ]
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Step 16000, best model saved. (accuracy=0.7225)\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "Train: 100% 2000/2000 [06:32<00:00,  5.10 step/s, accuracy=0.81, loss=0.62, step=18000]  \n",
-      "Valid: 100% 5664/5667 [01:00<00:00, 93.13 uttr/s, accuracy=0.75, loss=1.13] \n",
-      "Train: 100% 2000/2000 [07:12<00:00,  4.62 step/s, accuracy=0.91, loss=0.32, step=2e+4]   \n",
-      "Valid: 100% 5664/5667 [01:02<00:00, 91.35 uttr/s, accuracy=0.73, loss=1.14] \n",
-      "Train: 100% 2000/2000 [07:24<00:00,  4.50 step/s, accuracy=0.88, loss=0.53, step=22000] \n",
-      "Valid: 100% 5664/5667 [01:00<00:00, 93.62 uttr/s, accuracy=0.78, loss=1.00] \n",
-      "Train: 100% 2000/2000 [07:27<00:00,  4.47 step/s, accuracy=0.88, loss=0.32, step=24000] \n",
-      "Valid: 100% 5664/5667 [00:56<00:00, 101.13 uttr/s, accuracy=0.79, loss=0.93]\n",
+      "Train: 100% 2000/2000 [04:36<00:00,  7.24 step/s, accuracy=0.19, loss=3.68, step=2000]  \n",
+      "Valid: 100% 5664/5667 [00:52<00:00, 107.38 uttr/s, accuracy=0.23, loss=3.66]\n",
+      "Train: 100% 2000/2000 [04:33<00:00,  7.30 step/s, accuracy=0.47, loss=2.42, step=4000]  \n",
+      "Valid: 100% 5664/5667 [00:55<00:00, 101.57 uttr/s, accuracy=0.40, loss=2.75]\n",
+      "Train: 100% 2000/2000 [04:53<00:00,  6.82 step/s, accuracy=0.53, loss=1.94, step=6000] \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 103.31 uttr/s, accuracy=0.51, loss=2.22]\n",
+      "Train: 100% 2000/2000 [05:36<00:00,  5.94 step/s, accuracy=0.69, loss=1.35, step=8000] \n",
+      "Valid: 100% 5664/5667 [00:52<00:00, 108.07 uttr/s, accuracy=0.58, loss=1.88]\n",
+      "Train: 100% 2000/2000 [04:40<00:00,  7.13 step/s, accuracy=0.75, loss=0.93, step=1e+4]  \n",
+      "Valid: 100% 5664/5667 [00:53<00:00, 106.50 uttr/s, accuracy=0.63, loss=1.69]\n",
       "Train:   0% 0/2000 [00:00<?, ? step/s]"
      ]
     },
@@ -544,21 +502,23 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Step 24000, best model saved. (accuracy=0.7890)\n"
+      "Step 10000, best model saved. (accuracy=0.6262)\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "Train: 100% 2000/2000 [06:23<00:00,  5.21 step/s, accuracy=0.97, loss=0.16, step=26000]  \n",
-      "Valid: 100% 5664/5667 [00:57<00:00, 98.43 uttr/s, accuracy=0.78, loss=1.00] \n",
-      "Train: 100% 2000/2000 [06:04<00:00,  5.49 step/s, accuracy=0.84, loss=0.44, step=28000]  \n",
-      "Valid: 100% 5664/5667 [00:59<00:00, 95.85 uttr/s, accuracy=0.80, loss=0.90] \n",
-      "Train: 100% 2000/2000 [20:59<00:00,  1.59 step/s, accuracy=0.81, loss=0.52, step=3e+4]     \n",
-      "Valid: 100% 5664/5667 [04:36<00:00, 20.49 uttr/s, accuracy=0.80, loss=0.91]  \n",
-      "Train: 100% 2000/2000 [22:10<00:00,  1.50 step/s, accuracy=0.94, loss=0.22, step=32000]    \n",
-      "Valid: 100% 5664/5667 [00:51<00:00, 109.33 uttr/s, accuracy=0.80, loss=0.91] \n",
+      "Train: 100% 2000/2000 [04:29<00:00,  7.43 step/s, accuracy=0.66, loss=1.40, step=12000]  \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 104.72 uttr/s, accuracy=0.65, loss=1.62]\n",
+      "Train: 100% 2000/2000 [04:29<00:00,  7.42 step/s, accuracy=0.62, loss=1.48, step=14000] \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 103.87 uttr/s, accuracy=0.67, loss=1.49]\n",
+      "Train: 100% 2000/2000 [05:17<00:00,  6.29 step/s, accuracy=0.69, loss=1.37, step=16000] \n",
+      "Valid: 100% 5664/5667 [00:52<00:00, 108.50 uttr/s, accuracy=0.68, loss=1.48] \n",
+      "Train: 100% 2000/2000 [04:31<00:00,  7.36 step/s, accuracy=0.72, loss=1.02, step=18000]  \n",
+      "Valid: 100% 5664/5667 [00:53<00:00, 105.38 uttr/s, accuracy=0.68, loss=1.51]\n",
+      "Train: 100% 2000/2000 [04:30<00:00,  7.39 step/s, accuracy=0.88, loss=0.52, step=2e+4]   \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 103.67 uttr/s, accuracy=0.70, loss=1.50]\n",
       "Train:   0% 0/2000 [00:00<?, ? step/s]"
      ]
     },
@@ -566,21 +526,23 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Step 32000, best model saved. (accuracy=0.7991)\n"
+      "Step 20000, best model saved. (accuracy=0.6974)\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "Train: 100% 2000/2000 [04:22<00:00,  7.61 step/s, accuracy=0.91, loss=0.28, step=34000]  \n",
-      "Valid: 100% 5664/5667 [08:57<00:00, 10.54 uttr/s, accuracy=0.81, loss=0.87]  \n",
-      "Train: 100% 2000/2000 [04:29<00:00,  7.42 step/s, accuracy=0.94, loss=0.21, step=36000]  \n",
-      "Valid: 100% 5664/5667 [00:53<00:00, 106.10 uttr/s, accuracy=0.82, loss=0.84]\n",
-      "Train: 100% 2000/2000 [04:33<00:00,  7.31 step/s, accuracy=0.88, loss=0.69, step=38000] \n",
-      "Valid: 100% 5664/5667 [00:54<00:00, 103.30 uttr/s, accuracy=0.83, loss=0.79]\n",
-      "Train: 100% 2000/2000 [05:15<00:00,  6.34 step/s, accuracy=0.88, loss=0.27, step=4e+4]  \n",
-      "Valid: 100% 5664/5667 [00:52<00:00, 107.15 uttr/s, accuracy=0.83, loss=0.78] \n",
+      "Train: 100% 2000/2000 [04:34<00:00,  7.29 step/s, accuracy=0.78, loss=0.67, step=22000] \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 103.27 uttr/s, accuracy=0.73, loss=1.30]\n",
+      "Train: 100% 2000/2000 [05:21<00:00,  6.22 step/s, accuracy=0.94, loss=0.38, step=24000] \n",
+      "Valid: 100% 5664/5667 [00:52<00:00, 107.76 uttr/s, accuracy=0.74, loss=1.24]\n",
+      "Train: 100% 2000/2000 [04:34<00:00,  7.29 step/s, accuracy=0.94, loss=0.32, step=26000]  \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 104.57 uttr/s, accuracy=0.73, loss=1.39]\n",
+      "Train: 100% 2000/2000 [04:34<00:00,  7.29 step/s, accuracy=0.75, loss=1.00, step=28000]  \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 103.97 uttr/s, accuracy=0.73, loss=1.30]\n",
+      "Train: 100% 2000/2000 [04:27<00:00,  7.46 step/s, accuracy=0.88, loss=0.50, step=3e+4]  \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 104.73 uttr/s, accuracy=0.74, loss=1.25]\n",
       "Train:   0% 0/2000 [00:00<?, ? step/s]"
      ]
     },
@@ -588,21 +550,23 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Step 40000, best model saved. (accuracy=0.8314)\n"
+      "Step 30000, best model saved. (accuracy=0.7438)\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "Train: 100% 2000/2000 [04:30<00:00,  7.40 step/s, accuracy=0.81, loss=0.66, step=42000]  \n",
-      "Valid: 100% 5664/5667 [00:54<00:00, 104.40 uttr/s, accuracy=0.83, loss=0.78]\n",
-      "Train: 100% 2000/2000 [04:53<00:00,  6.83 step/s, accuracy=0.91, loss=0.29, step=44000]  \n",
-      "Valid: 100% 5664/5667 [00:58<00:00, 97.09 uttr/s, accuracy=0.83, loss=0.78] \n",
-      "Train: 100% 2000/2000 [06:14<00:00,  5.34 step/s, accuracy=0.94, loss=0.32, step=46000] \n",
-      "Valid: 100% 5664/5667 [01:00<00:00, 94.26 uttr/s, accuracy=0.84, loss=0.75] \n",
-      "Train: 100% 2000/2000 [07:06<00:00,  4.69 step/s, accuracy=1.00, loss=0.07, step=48000] \n",
-      "Valid: 100% 5664/5667 [00:54<00:00, 103.14 uttr/s, accuracy=0.84, loss=0.71]\n",
+      "Train: 100% 2000/2000 [05:38<00:00,  5.91 step/s, accuracy=0.84, loss=0.55, step=32000] \n",
+      "Valid: 100% 5664/5667 [00:53<00:00, 106.59 uttr/s, accuracy=0.75, loss=1.28]\n",
+      "Train: 100% 2000/2000 [04:40<00:00,  7.12 step/s, accuracy=0.81, loss=0.52, step=34000]  \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 104.54 uttr/s, accuracy=0.77, loss=1.22]\n",
+      "Train: 100% 2000/2000 [04:36<00:00,  7.24 step/s, accuracy=0.84, loss=0.53, step=36000]  \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 104.56 uttr/s, accuracy=0.78, loss=1.12]\n",
+      "Train: 100% 2000/2000 [04:48<00:00,  6.94 step/s, accuracy=0.78, loss=0.60, step=38000] \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 103.09 uttr/s, accuracy=0.77, loss=1.18]\n",
+      "Train: 100% 2000/2000 [05:15<00:00,  6.34 step/s, accuracy=0.97, loss=0.20, step=4e+4]  \n",
+      "Valid: 100% 5664/5667 [00:52<00:00, 108.32 uttr/s, accuracy=0.78, loss=1.14]\n",
       "Train:   0% 0/2000 [00:00<?, ? step/s]"
      ]
     },
@@ -610,80 +574,16 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Step 48000, best model saved. (accuracy=0.8448)\n"
+      "Step 40000, best model saved. (accuracy=0.7821)\n"
      ]
     },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "Train: 100% 2000/2000 [06:08<00:00,  5.43 step/s, accuracy=0.94, loss=0.40, step=5e+4]   \n",
-      "Valid: 100% 5664/5667 [00:59<00:00, 95.57 uttr/s, accuracy=0.85, loss=0.72] \n",
-      "Train: 100% 2000/2000 [06:04<00:00,  5.48 step/s, accuracy=0.94, loss=0.15, step=52000] \n",
-      "Valid: 100% 5664/5667 [00:58<00:00, 96.03 uttr/s, accuracy=0.85, loss=0.69] \n",
-      "Train: 100% 2000/2000 [06:10<00:00,  5.39 step/s, accuracy=0.97, loss=0.17, step=54000] \n",
-      "Valid: 100% 5664/5667 [00:57<00:00, 98.99 uttr/s, accuracy=0.86, loss=0.66] \n",
-      "Train: 100% 2000/2000 [06:30<00:00,  5.13 step/s, accuracy=0.97, loss=0.18, step=56000] \n",
-      "Valid: 100% 5664/5667 [00:54<00:00, 104.36 uttr/s, accuracy=0.86, loss=0.68]\n",
-      "Train:   0% 0/2000 [00:00<?, ? step/s]"
-     ]
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Step 56000, best model saved. (accuracy=0.8619)\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "Train: 100% 2000/2000 [06:18<00:00,  5.29 step/s, accuracy=0.97, loss=0.16, step=66000]  \n",
-      "Valid: 100% 5664/5667 [00:59<00:00, 94.93 uttr/s, accuracy=0.88, loss=0.60] \n",
-      "Train: 100% 2000/2000 [05:48<00:00,  5.73 step/s, accuracy=1.00, loss=0.04, step=68000] \n",
-      "Valid: 100% 5664/5667 [00:58<00:00, 96.95 uttr/s, accuracy=0.88, loss=0.62] \n",
-      "Train: 100% 2000/2000 [05:43<00:00,  5.82 step/s, accuracy=1.00, loss=0.03, step=7e+4]  \n",
-      "Valid: 100% 5664/5667 [00:58<00:00, 97.21 uttr/s, accuracy=0.87, loss=0.61] \n",
-      "Train: 100% 2000/2000 [06:33<00:00,  5.09 step/s, accuracy=1.00, loss=0.03, step=72000]  \n",
-      "Valid: 100% 5664/5667 [00:54<00:00, 104.50 uttr/s, accuracy=0.88, loss=0.58]\n",
-      "Train:   0% 0/2000 [00:00<?, ? step/s]"
-     ]
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Step 72000, best model saved. (accuracy=0.8810)\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "Train: 100% 2000/2000 [05:44<00:00,  5.80 step/s, accuracy=1.00, loss=0.03, step=74000]  \n",
-      "Valid: 100% 5664/5667 [00:56<00:00, 99.98 uttr/s, accuracy=0.89, loss=0.58] \n",
-      "Train: 100% 2000/2000 [05:47<00:00,  5.76 step/s, accuracy=1.00, loss=0.02, step=76000] \n",
-      "Valid: 100% 5664/5667 [00:58<00:00, 97.10 uttr/s, accuracy=0.88, loss=0.57] \n",
-      "Train: 100% 2000/2000 [05:54<00:00,  5.64 step/s, accuracy=1.00, loss=0.06, step=78000] \n",
-      "Valid: 100% 5664/5667 [00:58<00:00, 96.74 uttr/s, accuracy=0.88, loss=0.60] \n",
-      "Train: 100% 2000/2000 [06:37<00:00,  5.03 step/s, accuracy=1.00, loss=0.02, step=8e+4]   \n",
-      "Valid: 100% 5664/5667 [00:54<00:00, 103.68 uttr/s, accuracy=0.89, loss=0.58]\n",
-      "Train:   0% 0/2000 [00:00<?, ? step/s]"
-     ]
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Step 80000, best model saved. (accuracy=0.8868)\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "Train:   0% 0/2000 [00:12<?, ? step/s]\n"
+      "Train: 100% 2000/2000 [04:34<00:00,  7.30 step/s, accuracy=0.78, loss=0.78, step=42000]  \n",
+      "Valid: 100% 5664/5667 [00:54<00:00, 103.96 uttr/s, accuracy=0.76, loss=1.29]\n",
+      "Train:  43% 867/2000 [01:45<02:21,  8.03 step/s, accuracy=0.91, loss=0.28, step=42866]"
      ]
     }
    ],
@@ -704,8 +604,8 @@
     "        \"n_workers\": 8,\n",
     "        \"valid_steps\": 2000,\n",
     "        \"warmup_steps\": 1000,\n",
-    "        \"save_steps\": 8000,\n",
-    "        \"total_steps\": 80000,\n",
+    "        \"save_steps\": 10000,\n",
+    "        \"total_steps\": 160000,\n",
     "    }\n",
     "\n",
     "    return config\n",
@@ -802,20 +702,9 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": 11,
    "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "0"
-      ]
-     },
-     "execution_count": 14,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
+   "outputs": [],
    "source": [
     "import gc\n",
     "\n",
@@ -836,7 +725,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": 8,
    "metadata": {
     "id": "efS4pCmAJXJH"
    },
@@ -863,7 +752,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 9,
    "metadata": {
     "colab": {
      "base_uri": "https://localhost:8080/",
@@ -898,7 +787,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "397ac09d6c52481cbac427496d266b3e",
+       "model_id": "4a12f429161a47e2832ec39d1fe0b251",
        "version_major": 2,
        "version_minor": 0
       },
@@ -979,9 +868,21 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 10,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "ename": "NameError",
+     "evalue": "name 'dataloader' is not defined",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
+      "Cell \u001b[0;32mIn[10], line 2\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mimport\u001b[39;00m \u001b[38;5;21;01mgc\u001b[39;00m\n\u001b[0;32m----> 2\u001b[0m \u001b[38;5;28;01mdel\u001b[39;00m dataloader\n\u001b[1;32m      3\u001b[0m gc\u001b[38;5;241m.\u001b[39mcollect()\n",
+      "\u001b[0;31mNameError\u001b[0m: name 'dataloader' is not defined"
+     ]
+    }
+   ],
    "source": [
     "import gc\n",
     "del dataloader\n",