Notebooks/Anxiety-Disorders-Training.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Bu notebook'da Anksiyete spesifiğindeki hastalıklar olan Agorafobi, Panik, Fobi, Seçici Dilsizlik ve Sosyal Anksiyete bazında bir multi-label model geliştirilmektedir.**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Import Libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "import warnings \n",
    "warnings.filterwarnings(\"ignore\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load Data and Specify Classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import DatasetDict\n",
    "\n",
    "dataset = DatasetDict.load_from_disk(\"../Data/Anxiety_Detection_Data/anxiety_balanced_not_augmented_hg_dataset\")\n",
    "\n",
    "classes = ['agoraphobia', 'panic', 'phobia', 'selectivemutism', 'socialanxiety']\n",
    "class2id = {class_:id for id, class_ in enumerate(classes)}\n",
    "id2class = {id:class_ for class_, id in class2id.items()}\n",
    "# !huggingface-cli login --token=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\n",
    "# dataset = load_dataset('halilibr/dilbazlar-anxiety-disorders-recognition-not-augmented-dataset') "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "DatasetDict({\n",
       "    train: Dataset({\n",
       "        features: ['text', 'agoraphobia', 'panic', 'phobia', 'selectivemutism', 'socialanxiety', 'all_labels', 'source'],\n",
       "        num_rows: 7676\n",
       "    })\n",
       "    test: Dataset({\n",
       "        features: ['text', 'agoraphobia', 'panic', 'phobia', 'selectivemutism', 'socialanxiety', 'all_labels', 'source'],\n",
       "        num_rows: 2293\n",
       "    })\n",
       "})"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load the Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Kullanılacak modelin yolunu belirt\n",
    "model_path = 'dbmdz/bert-base-turkish-uncased'\n",
    "\n",
    "# Belirtilen model yoluna göre bir tokenizer oluştur\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_path)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "a84bcfdf333a47c2b80672e538d41728",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Map:   0%|          | 0/7676 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "87558be682304b658a4179a157d4cbd6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Map:   0%|          | 0/2293 [00:00<?, ? examples/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Metni ön işleme tabi tutacak bir fonksiyon tanımla\n",
    "\n",
    "def preprocess_function(example):\n",
    "    \n",
    "    # Örneğin 'text' alanını al ve text değişkenine ata\n",
    "    text = f\"{example['text']}\"\n",
    "\n",
    "    # Örneğin etiketlerini al\n",
    "    all_labels = example['all_labels']\n",
    "    # Etiketlerin sayısı kadar sıfırdan oluşan bir liste oluştur\n",
    "    labels = [0. for i in range(len(classes))]\n",
    "\n",
    "    # Her bir etiket için\n",
    "    for label in all_labels:\n",
    "        # Etiketin sınıf kimliğini al\n",
    "        label_id = class2id[label]\n",
    "        # İlgili sınıf kimliğinin olduğu yere 1.0 ata\n",
    "        labels[label_id] = 1.\n",
    "\n",
    "    # Metni tokenlara dönüştür ve örneğe ata\n",
    "    example = tokenizer(text, truncation=True)\n",
    "    # Tokenize edilmiş örneğe etiketleri ekle\n",
    "    example['labels'] = labels\n",
    "    # Ön işlenmiş örneği döndür\n",
    "    return example\n",
    "\n",
    "# Dataset üzerindeki her bir örnek için preprocess_function fonksiyonunu uygula ve tokenized_dataset değişkenine ata\n",
    "tokenized_dataset = dataset.map(preprocess_function)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "# transformers kütüphanesinden DataCollatorWithPadding sınıfını içe aktar\n",
    "from transformers import DataCollatorWithPadding\n",
    "\n",
    "# Tokenizer kullanarak bir data collator oluştur\n",
    "data_collator = DataCollatorWithPadding(tokenizer=tokenizer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "# evaluate kütüphanesini ve numpy'yi içe aktar\n",
    "import evaluate\n",
    "import numpy as np\n",
    "\n",
    "# Doğruluk (accuracy), F1 skoru, hassasiyet (precision) ve duyarlılık (recall) metriklerini birleştir\n",
    "clf_metrics = evaluate.combine([\"accuracy\", \"f1\", \"precision\", \"recall\"])\n",
    "\n",
    "# Sigmoid fonksiyonunu tanımla\n",
    "def sigmoid(x):\n",
    "    return 1/(1 + np.exp(-x))\n",
    "\n",
    "# Metin sınıflandırma modeli için metrikleri hesaplayan fonksiyonu tanımla\n",
    "def compute_metrics(eval_pred):\n",
    "   # Tahminleri ve etiketleri ayır\n",
    "   predictions, labels = eval_pred\n",
    "   # Tahminler için sigmoid fonksiyonunu uygula\n",
    "   predictions = sigmoid(predictions)\n",
    "   # 0.5'ten büyük olan tahminleri 1, diğerlerini 0 yap\n",
    "   predictions = (predictions > 0.5).astype(int).reshape(-1)\n",
    "   # Metrikleri hesapla ve geri döndür\n",
    "   return clf_metrics.compute(predictions=predictions, references=labels.astype(int).reshape(-1))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Some weights of BertForSequenceClassification were not initialized from the model checkpoint at dbmdz/bert-base-turkish-uncased and are newly initialized: ['classifier.bias', 'classifier.weight']\n",
      "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
     ]
    }
   ],
   "source": [
    "# transformers kütüphanesinden AutoModelForSequenceClassification, TrainingArguments ve Trainer sınıflarını içe aktar\n",
    "from transformers import AutoModelForSequenceClassification, TrainingArguments, Trainer\n",
    "\n",
    "# Sıra sınıflandırması için önceden eğitilmiş bir model oluştur\n",
    "model = AutoModelForSequenceClassification.from_pretrained(\n",
    "   model_path,                   # Model yolunu belirt\n",
    "   num_labels=len(classes),      # Sınıf sayısını belirt\n",
    "   id2label=id2class,            # Etiket ID'sinden sınıf ismine dönüştürücü harita\n",
    "   label2id=class2id,            # Sınıf isminden etiket ID'sine dönüştürücü harita\n",
    "   problem_type=\"multi_label_classification\" # Sorun tipini çoklu etiket sınıflandırması olarak ayarla\n",
    ")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ[\"WANDB_PROJECT\"]=\"Dilbazlar\"\n",
    "wandb_api_key = \"USE_YOUR_API_KEY\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "You're using a BertTokenizerFast tokenizer. Please note that with a fast tokenizer, using the `__call__` method is faster than using a method to encode the text followed by a call to the `pad` method to get a padded encoding.\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "      \n",
       "      <progress value='5118' max='5118' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      [5118/5118 12:45, Epoch 2/2]\n",
       "    </div>\n",
       "    <table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       " <tr style=\"text-align: left;\">\n",
       "      <th>Epoch</th>\n",
       "      <th>Training Loss</th>\n",
       "      <th>Validation Loss</th>\n",
       "      <th>Accuracy</th>\n",
       "      <th>F1</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.221500</td>\n",
       "      <td>0.217860</td>\n",
       "      <td>0.924030</td>\n",
       "      <td>0.807087</td>\n",
       "      <td>0.819982</td>\n",
       "      <td>0.794592</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.140300</td>\n",
       "      <td>0.229607</td>\n",
       "      <td>0.927780</td>\n",
       "      <td>0.817942</td>\n",
       "      <td>0.824834</td>\n",
       "      <td>0.811164</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table><p>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "TrainOutput(global_step=5118, training_loss=0.2135590500736572, metrics={'train_runtime': 765.2226, 'train_samples_per_second': 20.062, 'train_steps_per_second': 6.688, 'total_flos': 751222852839600.0, 'train_loss': 0.2135590500736572, 'epoch': 2.0})"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Eğitim parametrelerini belirle\n",
    "training_args = TrainingArguments(\n",
    "   output_dir=\"anxiety_disorders_model\",     # Modelin kaydedileceği dizin\n",
    "   learning_rate=2e-5,                       # Öğrenme oranı\n",
    "   per_device_train_batch_size=3,            # Eğitimde her cihaz için batch boyutu\n",
    "   per_device_eval_batch_size=3,             # Değerlendirmede her cihaz için batch boyutu\n",
    "   num_train_epochs=2,                       # Eğitimde kaç epoch kullanılacağı\n",
    "   weight_decay=0.01,                        # Ağırlık azaltma oranı\n",
    "   evaluation_strategy=\"epoch\",              # Değerlendirme stratejisi: her epoch sonunda\n",
    "   save_strategy=\"epoch\",                    # Modeli kaydetme stratejisi: her epoch sonunda\n",
    "   load_best_model_at_end=True,              # Eğitim sonunda en iyi modeli yükle\n",
    "   report_to=[\"wandb\"],                     # Weights & Biases'e raporlama yap\n",
    "   run_name=\"anxiety-specific-augmented-model\"  # Eğitim çalışması için bir isim\n",
    ")\n",
    "\n",
    "# Trainer nesnesini oluştur\n",
    "trainer = Trainer(\n",
    "   model=model,                             # Kullanılacak model\n",
    "   args=training_args,                      # Eğitim parametreleri\n",
    "   train_dataset=tokenized_dataset[\"train\"], # Eğitim veri kümesi\n",
    "   eval_dataset=tokenized_dataset[\"test\"],  # Değerlendirme veri kümesi\n",
    "   tokenizer=tokenizer,                    # Tokenizer\n",
    "   data_collator=data_collator,            # Veri toplama aracı\n",
    "   compute_metrics=compute_metrics,        # Metrikleri hesaplama fonksiyonu\n",
    ")\n",
    "\n",
    "# Modeli eğit\n",
    "trainer.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['agoraphobia', 'panic', 'phobia', 'selectivemutism', 'socialanxiety']"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'agoraphobia': 0,\n",
       " 'panic': 1,\n",
       " 'phobia': 2,\n",
       " 'selectivemutism': 3,\n",
       " 'socialanxiety': 4}"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class2id"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'text': 'Her yerde tam olarak aynı olan Yersiz, Büyük, Boş Bir Yerde olma korkusu, Okyanus, Gökyüzü, Dış Uzay ve hatta gerçek Zeminin metrelerce kumun altında olduğu geniş bir Çöl bile buna örnektir. Zemini bile görememek, çok daha az dokunmamak, nereye bakarsanız bakın aynı, sadece gizemli ve muazzam enginlikte yüzen, tamamen savunmasız.',\n",
       " 'agoraphobia': 0,\n",
       " 'panic': 0,\n",
       " 'phobia': 1,\n",
       " 'selectivemutism': 0,\n",
       " 'socialanxiety': 0,\n",
       " 'all_labels': ['phobia'],\n",
       " 'source': 'Reddit'}"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset['test'][200]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "# PyTorch kütüphanesini içe aktar\n",
    "import torch\n",
    "\n",
    "# CUDA uyumlu bir GPU varsa, GPU'yu kullan; yoksa CPU'yu kullan\n",
    "device = torch.device(\"cuda\") if torch.cuda.is_available() else torch.device(\"cpu\")\n",
    "\n",
    "# Metin tahmini yapan bir fonksiyon tanımla\n",
    "def predict(text):\n",
    "    # Metin girişi hazırlık\n",
    "    inputs = trainer.tokenizer(text, return_tensors=\"pt\").to(device)\n",
    "    \n",
    "    # Tahmin yap\n",
    "    with torch.no_grad():  # Hesaplamalar sırasında gradyan hesaplamalarını kapat\n",
    "        outputs = trainer.model(**inputs)  # Modelle tahmin yap\n",
    "        print(outputs)  # Model çıktısını yazdır\n",
    "        predictions = torch.argmax(outputs.logits, dim=-1)  # Logitlerden en yüksek olanı seç\n",
    "    \n",
    "    # Tahminleri işle\n",
    "    predicted_label = predictions.item()  # Tahmin edilen etiket\n",
    "    print(f\"Predicted label: {classes[predicted_label]}\")  # Tahmin edilen etiketi yazdır\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Kalbim çok hızlı atıyor, bir şey mi oluyor?\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[-2.0979,  0.9903, -2.9875, -4.4397, -6.9021]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: panic\n",
      "None\n",
      "**************************************************\n",
      "Nefes alamıyorum, boğuluyormuş gibi hissediyorum.\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[-3.3000,  1.8684, -2.4475, -3.4733, -6.9446]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: panic\n",
      "None\n",
      "**************************************************\n",
      "Başım dönüyor, bayılacak gibiyim.\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[-1.5482, -0.2935, -1.0072, -1.7374, -6.2352]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: panic\n",
      "None\n",
      "**************************************************\n",
      "Sürekli kötü bir şey olacakmış gibi hissediyorum.\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[-1.6660, -0.5807, -0.5266, -1.8419, -6.1230]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: phobia\n",
      "None\n",
      "**************************************************\n",
      "Göğsümde bir ağrı var, kalp krizi mi geçiriyorum?\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[ 2.2314, -2.5223, -4.9642, -5.3235, -4.3936]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: agoraphobia\n",
      "None\n",
      "**************************************************\n",
      "Ellerim titriyor, kontrol edemiyorum.\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[-2.2102, -1.0668,  0.2014, -1.8887, -6.6609]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: phobia\n",
      "None\n",
      "**************************************************\n",
      "Sanki her şey üzerime geliyormuş gibi hissediyorum.\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[-2.1067, -0.9843, -0.3476, -1.1846, -6.3466]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: phobia\n",
      "None\n",
      "**************************************************\n",
      "Çok fazla terliyorum, vücudum ateş gibi yanıyor.\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[-0.7661, -0.7493, -1.3760, -1.8543, -6.2085]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: panic\n",
      "None\n",
      "**************************************************\n",
      "Etrafımdaki her şey bulanıklaşıyor, gerçek değilmiş gibi.\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[-2.2910, -0.9358, -1.1200, -0.5143, -6.6203]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: selectivemutism\n",
      "None\n",
      "**************************************************\n",
      "Bunu atlatamayacağım, buradan çıkmalıyım.\n",
      "SequenceClassifierOutput(loss=None, logits=tensor([[ 2.4416, -3.7576, -4.2823, -3.9320, -5.4922]], device='cuda:0'), hidden_states=None, attentions=None)\n",
      "Predicted label: agoraphobia\n",
      "None\n",
      "**************************************************\n"
     ]
    }
   ],
   "source": [
    "# Panik atak ile ilgili örnek cümleleri tanımla\n",
    "panic_sentences = [\n",
    "    \"Kalbim çok hızlı atıyor, bir şey mi oluyor?\",\n",
    "    \"Nefes alamıyorum, boğuluyormuş gibi hissediyorum.\",\n",
    "    \"Başım dönüyor, bayılacak gibiyim.\",\n",
    "    \"Sürekli kötü bir şey olacakmış gibi hissediyorum.\",\n",
    "    \"Göğsümde bir ağrı var, kalp krizi mi geçiriyorum?\",\n",
    "    \"Ellerim titriyor, kontrol edemiyorum.\",\n",
    "    \"Sanki her şey üzerime geliyormuş gibi hissediyorum.\",\n",
    "    \"Çok fazla terliyorum, vücudum ateş gibi yanıyor.\",\n",
    "    \"Etrafımdaki her şey bulanıklaşıyor, gerçek değilmiş gibi.\",\n",
    "    \"Bunu atlatamayacağım, buradan çıkmalıyım.\"\n",
    "]\n",
    "\n",
    "# Her bir cümle için tahmin yap\n",
    "for text in panic_sentences:\n",
    "    print(text)  # Cümleyi yazdır\n",
    "    print(predict(text))  # Tahmini yap ve yazdır\n",
    "    print(\"*\"*50)  # Ayırıcı çizgi\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(11408, 5)"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions[0].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([2, 4, 1, ..., 4, 2, 2], dtype=int64)"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([2, 4, 1, 3, 3, 1, 4, 3, 4, 0, 2, 4, 0, 0, 2, 4, 1, 1, 1, 4],\n",
       "      dtype=int64)"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.argmax(predictions[1], axis=1)[:20]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([2, 4, 1, 3, 3, 1, 4, 3, 4, 0, 2, 4, 0, 0, 2, 4, 1, 1, 1, 4],\n",
       "      dtype=int64)"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.argmax(predictions.label_ids, axis=1)[:20]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x2193f0df610>"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay\n",
    "import numpy as np\n",
    "\n",
    "# Step 1: Evaluate the model and get predictions\n",
    "predictions = trainer.predict(tokenized_dataset[\"test\"])\n",
    "\n",
    "# Step 2: Extract the true labels and predictions\n",
    "true_labels = [np.argmax(tokenized_dataset['test'][i]['labels']) for i in range(len(tokenized_dataset['test']))]\n",
    "\n",
    "# Step 3: Create the confusion matrix\n",
    "conf_matrix = confusion_matrix(true_labels, np.argmax(predictions[0],  axis=1))\n",
    "\n",
    "# Display the confusion matrix\n",
    "disp = ConfusionMatrixDisplay(confusion_matrix=conf_matrix)\n",
    "disp.plot()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['agoraphobia', 'panic', 'phobia', 'selectivemutism', 'socialanxiety']"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "classes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Push To Hub"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Token will not been saved to git credential helper. Pass `add_to_git_credential=True` if you want to set the git credential as well.\n",
      "Token is valid (permission: write).\n",
      "Your token has been saved to C:\\Users\\halilibrahim.hatun\\.cache\\huggingface\\token\n",
      "Login successful\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "35327c5d7caf44b5bf5673aa245b898e",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "model.safetensors:   0%|          | 0.00/443M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "It was published :)\n"
     ]
    }
   ],
   "source": [
    "!huggingface-cli login --token=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\n",
    "\n",
    "trainer.model.push_to_hub(\"halilibr/dilbazlar-bert-uncased-anxiety-disorders-recognition-balanced-tr-model-acc-92-7\")\n",
    "trainer.tokenizer.push_to_hub(\"halilibr/dilbazlar-bert-uncased-anxiety-disorders-recognition-balanced-tr-model-acc-92-7\")\n",
    "print(\"It was published :)\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kaggle": {
   "accelerator": "none",
   "dataSources": [],
   "dockerImageVersionId": 30746,
   "isGpuEnabled": false,
   "isInternetEnabled": true,
   "language": "python",
   "sourceType": "notebook"
  },
  "kernelspec": {
   "display_name": "Psynexa Torch GPU (Python 3.11)",
   "language": "python",
   "name": "your_env_name"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}