ANLP_WS24_CA2/cnn_1b.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CNN 1b"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load Packages"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "from torch.utils.data import DataLoader\n",
    "from torch.optim.lr_scheduler import ReduceLROnPlateau\n",
    "from sklearn.metrics import accuracy_score, f1_score, confusion_matrix\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Datensatz laden und DatenLoader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/l7/061cw0t95vz1myntpf9bj9540000gn/T/ipykernel_32265/1764171208.py:5: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.\n",
      "  train_dataset = torch.load(data_path + '/train.pt')\n",
      "/var/folders/l7/061cw0t95vz1myntpf9bj9540000gn/T/ipykernel_32265/1764171208.py:6: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.\n",
      "  test_dataset = torch.load(data_path + '/test.pt')\n",
      "/var/folders/l7/061cw0t95vz1myntpf9bj9540000gn/T/ipykernel_32265/1764171208.py:7: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.\n",
      "  val_dataset = torch.load(data_path + '/val.pt')\n"
     ]
    }
   ],
   "source": [
    "data_path = 'data/embedded_padded'\n",
    "\n",
    "BATCH_SIZE = 32\n",
    "\n",
    "train_dataset = torch.load(data_path + '/train.pt')\n",
    "test_dataset = torch.load(data_path + '/test.pt')\n",
    "val_dataset = torch.load(data_path + '/val.pt')\n",
    "\n",
    "# DataLoader vorbereiten\n",
    "\n",
    "\n",
    "def collate_fn(batch):\n",
    "    input_ids = torch.stack([item[\"input_ids\"] for item in batch])  \n",
    "    labels = torch.tensor([item[\"labels\"] for item in batch], dtype=torch.float32).unsqueeze(1)  \n",
    "    return input_ids, labels\n",
    "\n",
    "train_loader = DataLoader(train_dataset, batch_size=BATCH_SIZE, shuffle=True, collate_fn=collate_fn)\n",
    "val_loader = DataLoader(val_dataset, batch_size=BATCH_SIZE, shuffle=False, collate_fn=collate_fn)\n",
    "test_loader = DataLoader(test_dataset, batch_size=BATCH_SIZE, shuffle=False, collate_fn=collate_fn)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "### CNN-Modell definieren\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "class HumorCNN(nn.Module):\n",
    "    def __init__(self, embedding_dim=100):\n",
    "        super(HumorCNN, self).__init__()\n",
    "\n",
    "        self.conv1 = nn.Conv2d(1, 50, (3, embedding_dim))\n",
    "        self.conv2 = nn.Conv2d(1, 50, (4, embedding_dim))\n",
    "        self.conv3 = nn.Conv2d(1, 50, (5, embedding_dim))\n",
    "\n",
    "        self.bn1 = nn.BatchNorm1d(50)\n",
    "        self.bn2 = nn.BatchNorm1d(50)\n",
    "        self.bn3 = nn.BatchNorm1d(50)\n",
    "\n",
    "        self.fc = nn.Linear(150, 1)\n",
    "\n",
    "        self.dropout = nn.Dropout(0.5)\n",
    "    \n",
    "    def forward(self, x):\n",
    "        x = x.unsqueeze(1)  \n",
    "\n",
    "        x1 = F.relu(self.bn1(self.conv1(x).squeeze(3)))\n",
    "        x2 = F.relu(self.bn2(self.conv2(x).squeeze(3)))\n",
    "        x3 = F.relu(self.bn3(self.conv3(x).squeeze(3)))\n",
    "        \n",
    "        x1 = F.max_pool1d(x1, x1.size(2)).squeeze(2)\n",
    "        x2 = F.max_pool1d(x2, x2.size(2)).squeeze(2)\n",
    "        x3 = F.max_pool1d(x3, x3.size(2)).squeeze(2)\n",
    "\n",
    "        x = torch.cat((x1, x2, x3), 1)\n",
    "        \n",
    "        x = self.dropout(x)\n",
    "        x = self.fc(x)\n",
    "        return torch.sigmoid(x)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "\n",
    "### Training des Modells\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/michellegoppinger/.pyenv/versions/3.12.3/lib/python3.12/site-packages/torch/optim/lr_scheduler.py:62: UserWarning: The verbose parameter is deprecated. Please use get_last_lr() to access the learning rate.\n",
      "  warnings.warn(\n",
      "/Users/michellegoppinger/Documents/Dokumente – Laptop von Michelle/Uni/Master/ANLP/CA2/ANLP_WS24_CA2/HumorDataset.py:21: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at /Users/runner/work/pytorch/pytorch/pytorch/torch/csrc/utils/tensor_new.cpp:281.)\n",
      "  item = {'input_ids': torch.tensor(self.data[idx], dtype=torch.float)}\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10, Train Loss: 1.0778, Val Loss: 0.6097, Test Acc: 0.6734, Test F1: 0.6567\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.6734279918864098\n",
      "Epoch 2/10, Train Loss: 0.7699, Val Loss: 0.5868, Test Acc: 0.7069, Test F1: 0.7175\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.7068965517241379\n",
      "Epoch 3/10, Train Loss: 0.6620, Val Loss: 0.5702, Test Acc: 0.7373, Test F1: 0.7566\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.7373225152129818\n",
      "Epoch 4/10, Train Loss: 0.6219, Val Loss: 0.5475, Test Acc: 0.7556, Test F1: 0.7692\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.755578093306288\n",
      "Epoch 5/10, Train Loss: 0.6035, Val Loss: 0.5171, Test Acc: 0.7769, Test F1: 0.7804\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.7768762677484787\n",
      "Epoch 6/10, Train Loss: 0.5956, Val Loss: 0.5026, Test Acc: 0.7926, Test F1: 0.8111\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.7925963488843814\n",
      "Epoch 7/10, Train Loss: 0.5601, Val Loss: 0.4781, Test Acc: 0.8119, Test F1: 0.7978\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.8118661257606491\n",
      "Epoch 8/10, Train Loss: 0.5375, Val Loss: 0.4429, Test Acc: 0.8281, Test F1: 0.8433\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.8280933062880325\n",
      "Epoch 9/10, Train Loss: 0.5281, Val Loss: 0.4177, Test Acc: 0.8773, Test F1: 0.8818\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.8772819472616633\n",
      "Epoch 10/10, Train Loss: 0.5041, Val Loss: 0.3977, Test Acc: 0.8813, Test F1: 0.8741\n",
      "🚀 Bestes Modell gespeichert mit Test-Accuracy: 0.8813387423935092\n"
     ]
    }
   ],
   "source": [
    "# Automatische Geräteauswahl (Apple MPS, CUDA, CPU)\n",
    "if torch.backends.mps.is_available():\n",
    "    device = torch.device(\"mps\")  \n",
    "elif torch.cuda.is_available():\n",
    "    device = torch.device(\"cuda\")  \n",
    "else:\n",
    "    device = torch.device(\"cpu\")  \n",
    "\n",
    "model = HumorCNN().to(device)\n",
    "\n",
    "criterion = nn.BCELoss()\n",
    "optimizer = optim.Adam(model.parameters(), lr=0.001, weight_decay=1e-5)  \n",
    "\n",
    "scheduler = ReduceLROnPlateau(optimizer, mode='min', factor=0.5, patience=2, verbose=True)  \n",
    "\n",
    "epochs = 10  # Nur 10 Epochen\n",
    "best_val_loss = float('inf')\n",
    "best_test_accuracy = 0\n",
    "patience = 3\n",
    "counter = 0\n",
    "\n",
    "for epoch in range(epochs):\n",
    "    model.train()\n",
    "    total_loss = 0\n",
    "\n",
    "    for texts, labels in train_loader:\n",
    "        texts, labels = texts.to(device), labels.to(device)\n",
    "        optimizer.zero_grad()\n",
    "        outputs = model(texts)\n",
    "        loss = criterion(outputs, labels)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        total_loss += loss.item()\n",
    "    \n",
    "    avg_train_loss = total_loss / len(train_loader)\n",
    "\n",
    "    # ========================\n",
    "    # Validierung\n",
    "    # ========================\n",
    "    model.eval()\n",
    "    val_loss = 0\n",
    "    with torch.no_grad():\n",
    "        for texts, labels in val_loader:\n",
    "            texts, labels = texts.to(device), labels.to(device)\n",
    "            outputs = model(texts)\n",
    "            loss = criterion(outputs, labels)\n",
    "            val_loss += loss.item()\n",
    "    \n",
    "    avg_val_loss = val_loss / len(val_loader)\n",
    "\n",
    "    # ========================\n",
    "    # Evaluierung mit Testdaten\n",
    "    # ========================\n",
    "    test_preds = []\n",
    "    test_labels = []\n",
    "    with torch.no_grad():\n",
    "        for texts, labels in test_loader:\n",
    "            texts, labels = texts.to(device), labels.to(device)\n",
    "            outputs = model(texts)\n",
    "            predictions = (outputs > 0.5).float()\n",
    "            test_preds.extend(predictions.cpu().numpy())\n",
    "            test_labels.extend(labels.cpu().numpy())\n",
    "\n",
    "    test_accuracy = accuracy_score(test_labels, test_preds)\n",
    "    test_f1 = f1_score(test_labels, test_preds)\n",
    "\n",
    "    print(f'Epoch {epoch+1}/{epochs}, Train Loss: {avg_train_loss:.4f}, Val Loss: {avg_val_loss:.4f}, Test Acc: {test_accuracy:.4f}, Test F1: {test_f1:.4f}')\n",
    "    \n",
    "    # ========================\n",
    "    # Lernraten-Anpassung\n",
    "    # ========================\n",
    "    scheduler.step(avg_val_loss)\n",
    "\n",
    "    # ========================\n",
    "    # Bestes Modell speichern\n",
    "    # ========================\n",
    "    if test_accuracy > best_test_accuracy:\n",
    "        best_test_accuracy = test_accuracy\n",
    "        torch.save(model.state_dict(), \"best_model.pth\")\n",
    "        print(\"🚀 Bestes Modell gespeichert mit Test-Accuracy:\", test_accuracy)\n",
    "\n",
    "    # ========================\n",
    "    # Early Stopping\n",
    "    # ========================\n",
    "    if avg_val_loss < best_val_loss:\n",
    "        best_val_loss = avg_val_loss\n",
    "        counter = 0\n",
    "    else:\n",
    "        counter += 1\n",
    "        if counter >= patience:\n",
    "            print(\"Early Stopping ausgelöst!\")\n",
    "            break\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n",
    "### Finale Evaluierung & Confusion Matrix\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/l7/061cw0t95vz1myntpf9bj9540000gn/T/ipykernel_32265/3375079771.py:1: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.\n",
      "  model.load_state_dict(torch.load(\"best_model.pth\"))\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "🚀 Finale Test Accuracy: 0.8813\n",
      "🚀 Finale Test F1 Score: 0.8741\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 600x500 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model.load_state_dict(torch.load(\"best_model.pth\"))  \n",
    "model.eval()\n",
    "\n",
    "all_preds = []\n",
    "all_labels = []\n",
    "\n",
    "with torch.no_grad():  \n",
    "    for texts, labels in test_loader:\n",
    "        texts, labels = texts.to(device), labels.to(device)\n",
    "        outputs = model(texts)\n",
    "        predictions = (outputs > 0.5).float()\n",
    "        all_preds.extend(predictions.cpu().numpy())\n",
    "        all_labels.extend(labels.cpu().numpy())\n",
    "\n",
    "all_preds = [int(p[0]) for p in all_preds]\n",
    "all_labels = [int(l[0]) for l in all_labels]\n",
    "\n",
    "accuracy = accuracy_score(all_labels, all_preds)\n",
    "f1 = f1_score(all_labels, all_preds)\n",
    "\n",
    "print(f'🚀 Finale Test Accuracy: {accuracy:.4f}')\n",
    "print(f'🚀 Finale Test F1 Score: {f1:.4f}')\n",
    "\n",
    "# Confusion Matrix\n",
    "conf_matrix = confusion_matrix(all_labels, all_preds)\n",
    "\n",
    "plt.figure(figsize=(6,5))\n",
    "sns.heatmap(conf_matrix, annot=True, fmt='d', cmap=\"Blues\", xticklabels=['No Humor', 'Humor'], yticklabels=['No Humor', 'Humor'])\n",
    "plt.xlabel(\"Predicted Label\")\n",
    "plt.ylabel(\"True Label\")\n",
    "plt.title(\"Confusion Matrix\")\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.12.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}