{ "cells": [ { "cell_type": "code", "execution_count": 92, "id": "9bf2159a", "metadata": {}, "outputs": [], "source": [ "import json\n", "import numpy as np\n", "from pathlib import Path\n", "from sklearn.model_selection import train_test_split\n", "from tensorflow.keras.preprocessing.text import Tokenizer\n", "from tensorflow.keras.preprocessing.sequence import pad_sequences\n", "from tensorflow.keras.utils import to_categorical\n", "\n", "from tensorflow.keras.models import Model\n", "from tensorflow.keras.layers import (\n", " Input,\n", " Embedding,\n", " LSTM,\n", " Concatenate,\n", " Dense,\n", " TimeDistributed,\n", ")\n", "from tensorflow.keras.callbacks import EarlyStopping\n", "from sklearn.metrics import classification_report" ] }, { "cell_type": "code", "execution_count": 104, "id": "50118278", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\n", " Jumlah data valid: 732 / 732\n", " Jumlah data tidak valid: 0\n", "\n", "Contoh pertanyaan dari data valid:\n", "[\n", " [\n", " \"___\",\n", " \"adalah\",\n", " \"putra\",\n", " \"gph\",\n", " \"soerjaningrat\",\n", " \"dan\",\n", " \"cucu\",\n", " \"pakualam\",\n", " \"iii.\"\n", " ],\n", " [\n", " \"ia\",\n", " \"diterima\",\n", " \"belajar\",\n", " \"di\",\n", " \"___.\"\n", " ],\n", " [\n", " \"ia\",\n", " \"bersama\",\n", " \"douwes\",\n", " \"dekker\",\n", " \"dan\",\n", " \"dr.\",\n", " \"cipto\",\n", " \"mangoenkoesoemo\",\n", " \"lantas\",\n", " \"mendirikan\",\n", " \"___\",\n", " \"pada\",\n", " \"25\",\n", " \"desember\",\n", " \"1912.\"\n", " ]\n", "]\n" ] } ], "source": [ "# Load raw data\n", "with open(\"normalized_dataset.json\", encoding=\"utf-8\") as f:\n", " raw_data = json.load(f)\n", "\n", "# Validasi lengkap\n", "required_keys = {\"tokens\", \"ner\", \"srl\", \"question\", \"answer\", \"type\"}\n", "valid_data = []\n", "invalid_data = []\n", "\n", "for idx, item in enumerate(raw_data):\n", " error_messages = []\n", "\n", " if not isinstance(item, dict):\n", " error_messages.append(\"bukan dictionary\")\n", "\n", " missing_keys = required_keys - item.keys()\n", " if missing_keys:\n", " error_messages.append(f\"missing keys: {missing_keys}\")\n", "\n", " if not error_messages:\n", " # Cek tipe data dan None\n", " if (not isinstance(item[\"tokens\"], list) or\n", " not isinstance(item[\"ner\"], list) or\n", " not isinstance(item[\"srl\"], list) or\n", " not isinstance(item[\"question\"], list) or\n", " not isinstance(item[\"answer\"], list) or\n", " not isinstance(item[\"type\"], str)):\n", " error_messages.append(\"field type tidak sesuai\")\n", " \n", " if error_messages:\n", " print(f\"\\n Index {idx} | Masalah: {', '.join(error_messages)}\")\n", " print(json.dumps(item, indent=2, ensure_ascii=False))\n", " invalid_data.append(item)\n", " continue\n", "\n", " valid_data.append(item)\n", "\n", "# Statistik\n", "print(f\"\\n Jumlah data valid: {len(valid_data)} / {len(raw_data)}\")\n", "print(f\" Jumlah data tidak valid: {len(invalid_data)}\")\n", "\n", "# Proses data valid\n", "tokens = [[t.lower().strip() for t in item[\"tokens\"]] for item in valid_data]\n", "ner_tags = [item[\"ner\"] for item in valid_data]\n", "srl_tags = [item[\"srl\"] for item in valid_data]\n", "questions = [[token.lower().strip() for token in item[\"question\"]] for item in valid_data]\n", "answers = [[token.lower().strip() for token in item[\"answer\"]] for item in valid_data]\n", "types = [item[\"type\"] for item in valid_data]\n", "\n", "print(\"\\nContoh pertanyaan dari data valid:\")\n", "print(json.dumps(questions[:3], indent=2, ensure_ascii=False))\n" ] }, { "cell_type": "code", "execution_count": 94, "id": "4e3a0088", "metadata": {}, "outputs": [], "source": [ "# tokenize\n", "token_tok = Tokenizer(lower=False, oov_token=\"UNK\")\n", "token_ner = Tokenizer(lower=False)\n", "token_srl = Tokenizer(lower=False)\n", "token_q = Tokenizer(lower=False)\n", "token_a = Tokenizer(lower=False)\n", "token_type = Tokenizer(lower=False)\n", "\n", "token_tok.fit_on_texts(tokens)\n", "token_ner.fit_on_texts(ner_tags)\n", "token_srl.fit_on_texts(srl_tags)\n", "token_q.fit_on_texts(questions)\n", "token_a.fit_on_texts(answers)\n", "token_type.fit_on_texts(types)\n", "\n", "\n", "maxlen = 20" ] }, { "cell_type": "code", "execution_count": 95, "id": "555f9e22", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "{'tof', 'none', 'ftb'}\n" ] } ], "source": [ "\n", "X_tok = pad_sequences(\n", " token_tok.texts_to_sequences(tokens), padding=\"post\", maxlen=maxlen\n", ")\n", "X_ner = pad_sequences(\n", " token_ner.texts_to_sequences(ner_tags), padding=\"post\", maxlen=maxlen\n", ")\n", "X_srl = pad_sequences(\n", " token_srl.texts_to_sequences(srl_tags), padding=\"post\", maxlen=maxlen\n", ")\n", "y_q = pad_sequences(token_q.texts_to_sequences(questions), padding=\"post\", maxlen=maxlen)\n", "y_a = pad_sequences(token_a.texts_to_sequences(answers), padding=\"post\", maxlen=maxlen)\n", "\n", "print(set(types))\n", "\n", "y_type = [seq[0] for seq in token_type.texts_to_sequences(types)] # list of int\n", "y_type = to_categorical(np.array(y_type) - 1, num_classes=len(token_type.word_index))\n", "\n" ] }, { "cell_type": "code", "execution_count": 96, "id": "f530cfe7", "metadata": {}, "outputs": [], "source": [ "X_tok_train, X_tok_test, X_ner_train, X_ner_test, X_srl_train, X_srl_test, \\\n", "y_q_train, y_q_test, y_a_train, y_a_test, y_type_train, y_type_test = train_test_split(\n", " X_tok, X_ner, X_srl, y_q, y_a, y_type, test_size=0.2, random_state=42\n", ")\n", "\n", "X_train = [X_tok_train, X_ner_train, X_srl_train]\n", "X_test = [X_tok_test, X_ner_test, X_srl_test]" ] }, { "cell_type": "code", "execution_count": 97, "id": "255e2a9a", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
Model: \"functional_10\"\n",
       "
\n" ], "text/plain": [ "\u001b[1mModel: \"functional_10\"\u001b[0m\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓\n",
       "┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩\n",
       "│ tok_input           │ (None, None)      │          0 │ -                 │\n",
       "│ (InputLayer)        │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ ner_input           │ (None, None)      │          0 │ -                 │\n",
       "│ (InputLayer)        │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ srl_input           │ (None, None)      │          0 │ -                 │\n",
       "│ (InputLayer)        │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_30        │ (None, None, 128) │    354,560 │ tok_input[0][0]   │\n",
       "│ (Embedding)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_31        │ (None, None, 16)  │        672 │ ner_input[0][0]   │\n",
       "│ (Embedding)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_32        │ (None, None, 16)  │        448 │ srl_input[0][0]   │\n",
       "│ (Embedding)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ concatenate_10      │ (None, None, 160) │          0 │ embedding_30[0][… │\n",
       "│ (Concatenate)       │                   │            │ embedding_31[0][… │\n",
       "│                     │                   │            │ embedding_32[0][… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ lstm_10 (LSTM)      │ (None, None, 256) │    427,008 │ concatenate_10[0… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ get_item_10         │ (None, 256)       │          0 │ lstm_10[0][0]     │\n",
       "│ (GetItem)           │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ question_output     │ (None, None, 727) │    186,839 │ lstm_10[0][0]     │\n",
       "│ (TimeDistributed)   │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ answer_output       │ (None, None, 367) │     94,319 │ lstm_10[0][0]     │\n",
       "│ (TimeDistributed)   │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ type_output (Dense) │ (None, 3)         │        771 │ get_item_10[0][0] │\n",
       "└─────────────────────┴───────────────────┴────────────┴───────────────────┘\n",
       "
\n" ], "text/plain": [ "┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓\n", "┃\u001b[1m \u001b[0m\u001b[1mLayer (type) \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m Param #\u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mConnected to \u001b[0m\u001b[1m \u001b[0m┃\n", "┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩\n", "│ tok_input │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ - │\n", "│ (\u001b[38;5;33mInputLayer\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ ner_input │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ - │\n", "│ (\u001b[38;5;33mInputLayer\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ srl_input │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ - │\n", "│ (\u001b[38;5;33mInputLayer\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ embedding_30 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m128\u001b[0m) │ \u001b[38;5;34m354,560\u001b[0m │ tok_input[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mEmbedding\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ embedding_31 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m672\u001b[0m │ ner_input[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mEmbedding\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ embedding_32 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m448\u001b[0m │ srl_input[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mEmbedding\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ concatenate_10 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m160\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ embedding_30[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "│ (\u001b[38;5;33mConcatenate\u001b[0m) │ │ │ embedding_31[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "│ │ │ │ embedding_32[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ lstm_10 (\u001b[38;5;33mLSTM\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m) │ \u001b[38;5;34m427,008\u001b[0m │ concatenate_10[\u001b[38;5;34m0\u001b[0m… │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ get_item_10 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ lstm_10[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mGetItem\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ question_output │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m727\u001b[0m) │ \u001b[38;5;34m186,839\u001b[0m │ lstm_10[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mTimeDistributed\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ answer_output │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m367\u001b[0m) │ \u001b[38;5;34m94,319\u001b[0m │ lstm_10[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mTimeDistributed\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ type_output (\u001b[38;5;33mDense\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m3\u001b[0m) │ \u001b[38;5;34m771\u001b[0m │ get_item_10[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "└─────────────────────┴───────────────────┴────────────┴───────────────────┘\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
 Total params: 1,064,617 (4.06 MB)\n",
       "
\n" ], "text/plain": [ "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m1,064,617\u001b[0m (4.06 MB)\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
 Trainable params: 1,064,617 (4.06 MB)\n",
       "
\n" ], "text/plain": [ "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m1,064,617\u001b[0m (4.06 MB)\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
 Non-trainable params: 0 (0.00 B)\n",
       "
\n" ], "text/plain": [ "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n" ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stdout", "output_type": "stream", "text": [ "Epoch 1/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 31ms/step - answer_output_accuracy: 0.9421 - answer_output_loss: 1.0839 - loss: 3.7461 - question_output_accuracy: 0.8187 - question_output_loss: 1.8269 - type_output_accuracy: 0.5154 - type_output_loss: 0.8354 - val_answer_output_accuracy: 0.9683 - val_answer_output_loss: 0.1305 - val_loss: 1.3997 - val_question_output_accuracy: 0.8508 - val_question_output_loss: 0.5916 - val_type_output_accuracy: 0.5451 - val_type_output_loss: 0.6777\n", "Epoch 2/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m7s\u001b[0m 28ms/step - answer_output_accuracy: 0.9690 - answer_output_loss: 0.1140 - loss: 1.3590 - question_output_accuracy: 0.8490 - question_output_loss: 0.5613 - type_output_accuracy: 0.5599 - type_output_loss: 0.6837 - val_answer_output_accuracy: 0.9695 - val_answer_output_loss: 0.1072 - val_loss: 1.2185 - val_question_output_accuracy: 0.8523 - val_question_output_loss: 0.4397 - val_type_output_accuracy: 0.5484 - val_type_output_loss: 0.6715\n", "Epoch 3/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m7s\u001b[0m 28ms/step - answer_output_accuracy: 0.9704 - answer_output_loss: 0.0929 - loss: 1.1687 - question_output_accuracy: 0.8545 - question_output_loss: 0.4132 - type_output_accuracy: 0.5641 - type_output_loss: 0.6626 - val_answer_output_accuracy: 0.9699 - val_answer_output_loss: 0.0893 - val_loss: 1.1270 - val_question_output_accuracy: 0.8511 - val_question_output_loss: 0.3662 - val_type_output_accuracy: 0.5206 - val_type_output_loss: 0.6712\n", "Epoch 4/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 30ms/step - answer_output_accuracy: 0.9721 - answer_output_loss: 0.0755 - loss: 1.0676 - question_output_accuracy: 0.8573 - question_output_loss: 0.3360 - type_output_accuracy: 0.5583 - type_output_loss: 0.6562 - val_answer_output_accuracy: 0.9725 - val_answer_output_loss: 0.0819 - val_loss: 1.0924 - val_question_output_accuracy: 0.8538 - val_question_output_loss: 0.3355 - val_type_output_accuracy: 0.5217 - val_type_output_loss: 0.6720\n", "Epoch 5/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m7s\u001b[0m 29ms/step - answer_output_accuracy: 0.9736 - answer_output_loss: 0.0687 - loss: 1.0458 - question_output_accuracy: 0.8604 - question_output_loss: 0.3208 - type_output_accuracy: 0.5582 - type_output_loss: 0.6564 - val_answer_output_accuracy: 0.9727 - val_answer_output_loss: 0.0785 - val_loss: 1.0680 - val_question_output_accuracy: 0.8557 - val_question_output_loss: 0.3196 - val_type_output_accuracy: 0.5406 - val_type_output_loss: 0.6702\n", "Epoch 6/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 38ms/step - answer_output_accuracy: 0.9744 - answer_output_loss: 0.0606 - loss: 1.0098 - question_output_accuracy: 0.8582 - question_output_loss: 0.2944 - type_output_accuracy: 0.5536 - type_output_loss: 0.6548 - val_answer_output_accuracy: 0.9730 - val_answer_output_loss: 0.0761 - val_loss: 1.0590 - val_question_output_accuracy: 0.8529 - val_question_output_loss: 0.3082 - val_type_output_accuracy: 0.5106 - val_type_output_loss: 0.6747\n", "Epoch 7/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 44ms/step - answer_output_accuracy: 0.9743 - answer_output_loss: 0.0602 - loss: 0.9826 - question_output_accuracy: 0.8617 - question_output_loss: 0.2767 - type_output_accuracy: 0.5678 - type_output_loss: 0.6457 - val_answer_output_accuracy: 0.9737 - val_answer_output_loss: 0.0735 - val_loss: 1.0377 - val_question_output_accuracy: 0.8632 - val_question_output_loss: 0.2956 - val_type_output_accuracy: 0.5617 - val_type_output_loss: 0.6667\n", "Epoch 8/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 42ms/step - answer_output_accuracy: 0.9738 - answer_output_loss: 0.0577 - loss: 0.9710 - question_output_accuracy: 0.8606 - question_output_loss: 0.2633 - type_output_accuracy: 0.5602 - type_output_loss: 0.6501 - val_answer_output_accuracy: 0.9739 - val_answer_output_loss: 0.0702 - val_loss: 1.0265 - val_question_output_accuracy: 0.8609 - val_question_output_loss: 0.2860 - val_type_output_accuracy: 0.5462 - val_type_output_loss: 0.6695\n", "Epoch 9/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 42ms/step - answer_output_accuracy: 0.9743 - answer_output_loss: 0.0522 - loss: 0.9559 - question_output_accuracy: 0.8590 - question_output_loss: 0.2549 - type_output_accuracy: 0.5509 - type_output_loss: 0.6488 - val_answer_output_accuracy: 0.9715 - val_answer_output_loss: 0.0709 - val_loss: 1.0313 - val_question_output_accuracy: 0.8511 - val_question_output_loss: 0.2829 - val_type_output_accuracy: 0.5128 - val_type_output_loss: 0.6767\n", "Epoch 10/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 42ms/step - answer_output_accuracy: 0.9747 - answer_output_loss: 0.0512 - loss: 0.9363 - question_output_accuracy: 0.8622 - question_output_loss: 0.2406 - type_output_accuracy: 0.5592 - type_output_loss: 0.6445 - val_answer_output_accuracy: 0.9737 - val_answer_output_loss: 0.0704 - val_loss: 1.0202 - val_question_output_accuracy: 0.8548 - val_question_output_loss: 0.2788 - val_type_output_accuracy: 0.5495 - val_type_output_loss: 0.6721\n", "Epoch 11/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 42ms/step - answer_output_accuracy: 0.9748 - answer_output_loss: 0.0520 - loss: 0.9421 - question_output_accuracy: 0.8594 - question_output_loss: 0.2446 - type_output_accuracy: 0.5581 - type_output_loss: 0.6454 - val_answer_output_accuracy: 0.9735 - val_answer_output_loss: 0.0696 - val_loss: 1.0231 - val_question_output_accuracy: 0.8625 - val_question_output_loss: 0.2759 - val_type_output_accuracy: 0.5484 - val_type_output_loss: 0.6794\n", "Epoch 12/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 42ms/step - answer_output_accuracy: 0.9750 - answer_output_loss: 0.0499 - loss: 0.9247 - question_output_accuracy: 0.8637 - question_output_loss: 0.2316 - type_output_accuracy: 0.5730 - type_output_loss: 0.6433 - val_answer_output_accuracy: 0.9736 - val_answer_output_loss: 0.0693 - val_loss: 1.0144 - val_question_output_accuracy: 0.8623 - val_question_output_loss: 0.2735 - val_type_output_accuracy: 0.5584 - val_type_output_loss: 0.6716\n", "Epoch 13/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m11s\u001b[0m 45ms/step - answer_output_accuracy: 0.9743 - answer_output_loss: 0.0546 - loss: 0.9290 - question_output_accuracy: 0.8637 - question_output_loss: 0.2325 - type_output_accuracy: 0.5665 - type_output_loss: 0.6419 - val_answer_output_accuracy: 0.9721 - val_answer_output_loss: 0.0683 - val_loss: 1.0072 - val_question_output_accuracy: 0.8631 - val_question_output_loss: 0.2707 - val_type_output_accuracy: 0.5595 - val_type_output_loss: 0.6674\n", "Epoch 14/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m12s\u001b[0m 47ms/step - answer_output_accuracy: 0.9749 - answer_output_loss: 0.0482 - loss: 0.9212 - question_output_accuracy: 0.8620 - question_output_loss: 0.2314 - type_output_accuracy: 0.5596 - type_output_loss: 0.6417 - val_answer_output_accuracy: 0.9725 - val_answer_output_loss: 0.0687 - val_loss: 1.0154 - val_question_output_accuracy: 0.8514 - val_question_output_loss: 0.2748 - val_type_output_accuracy: 0.5339 - val_type_output_loss: 0.6714\n", "Epoch 15/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m10s\u001b[0m 39ms/step - answer_output_accuracy: 0.9742 - answer_output_loss: 0.0493 - loss: 0.9149 - question_output_accuracy: 0.8617 - question_output_loss: 0.2254 - type_output_accuracy: 0.5621 - type_output_loss: 0.6402 - val_answer_output_accuracy: 0.9719 - val_answer_output_loss: 0.0692 - val_loss: 1.0235 - val_question_output_accuracy: 0.8524 - val_question_output_loss: 0.2776 - val_type_output_accuracy: 0.5317 - val_type_output_loss: 0.6747\n", "Epoch 16/30\n", "\u001b[1m253/253\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m8s\u001b[0m 31ms/step - answer_output_accuracy: 0.9752 - answer_output_loss: 0.0441 - loss: 0.9078 - question_output_accuracy: 0.8643 - question_output_loss: 0.2212 - type_output_accuracy: 0.5728 - type_output_loss: 0.6425 - val_answer_output_accuracy: 0.9722 - val_answer_output_loss: 0.0681 - val_loss: 1.0137 - val_question_output_accuracy: 0.8541 - val_question_output_loss: 0.2713 - val_type_output_accuracy: 0.5306 - val_type_output_loss: 0.6736\n" ] } ], "source": [ "\n", "inp_tok = Input(shape=(None,), name=\"tok_input\")\n", "inp_ner = Input(shape=(None,), name=\"ner_input\")\n", "inp_srl = Input(shape=(None,), name=\"srl_input\")\n", "\n", "emb_tok = Embedding(input_dim=len(token_tok.word_index) + 1, output_dim=128)(inp_tok)\n", "emb_ner = Embedding(input_dim=len(token_ner.word_index) + 1, output_dim=16)(inp_ner)\n", "emb_srl = Embedding(input_dim=len(token_srl.word_index) + 1, output_dim=16)(inp_srl)\n", "\n", "# emb_tok = Embedding(input_dim=..., output_dim=..., mask_zero=True)(inp_tok)\n", "# emb_ner = Embedding(input_dim=..., output_dim=..., mask_zero=True)(inp_ner)\n", "# emb_srl = Embedding(input_dim=..., output_dim=..., mask_zero=True)(inp_srl)\n", "\n", "merged = Concatenate()([emb_tok, emb_ner, emb_srl])\n", "\n", "x = LSTM(256, return_sequences=True)(merged)\n", "\n", "out_question = TimeDistributed(Dense(len(token_q.word_index) + 1, activation=\"softmax\"), name=\"question_output\")(x)\n", "out_answer = TimeDistributed(Dense(len(token_a.word_index) + 1, activation=\"softmax\"), name=\"answer_output\")(x)\n", "out_type = Dense(len(token_type.word_index), activation=\"softmax\", name=\"type_output\")(\n", " x[:, 0, :]\n", ") # gunakan step pertama\n", "\n", "model = Model(\n", " inputs=[inp_tok, inp_ner, inp_srl], outputs=[out_question, out_answer, out_type]\n", ")\n", "model.compile(\n", " optimizer=\"adam\",\n", " loss={\n", " \"question_output\": \"sparse_categorical_crossentropy\",\n", " \"answer_output\": \"sparse_categorical_crossentropy\",\n", " \"type_output\": \"categorical_crossentropy\",\n", " },\n", " metrics={\n", " \"question_output\": \"accuracy\",\n", " \"answer_output\": \"accuracy\",\n", " \"type_output\": \"accuracy\",\n", " },\n", ")\n", "\n", "model.summary()\n", "\n", "# ----------------------------------------------------------------------------\n", "# 5. TRAINING\n", "# ----------------------------------------------------------------------------\n", "model.fit(\n", " X_train,\n", " {\n", " \"question_output\": np.expand_dims(y_q_train, -1),\n", " \"answer_output\": np.expand_dims(y_a_train, -1),\n", " \"type_output\": y_type_train,\n", " },\n", " batch_size=32,\n", " epochs=30,\n", " validation_split=0.1,\n", " callbacks=[EarlyStopping(patience=3, restore_best_weights=True)],\n", ")\n", "\n", "import pickle\n", "\n", "\n", "model.save(\"new_model_lstm_qg.keras\")\n", "with open(\"tokenizers.pkl\", \"wb\") as f:\n", " pickle.dump({\n", " \"token\": token_tok,\n", " \"ner\": token_ner,\n", " \"srl\": token_srl,\n", " \"question\": token_q,\n", " \"answer\": token_a,\n", " \"type\": token_type\n", " }, f)\n", "\n" ] }, { "cell_type": "code", "execution_count": 98, "id": "06fd86c7", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "\u001b[1m71/71\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 13ms/step\n", "\n", "=== Akurasi Detail ===\n", "Question Accuracy (Token-level): 0.3\n", "Answer Accuracy (Token-level) : 0.7\n", "Type Accuracy (Class-level) : 0.55\n" ] } ], "source": [ "\n", "def token_level_accuracy(y_true, y_pred):\n", " correct = 0\n", " total = 0\n", " for true_seq, pred_seq in zip(y_true, y_pred):\n", " for t, p in zip(true_seq, pred_seq):\n", " if t != 0: # ignore padding\n", " total += 1\n", " if t == p:\n", " correct += 1\n", " return correct / total if total > 0 else 0\n", "\n", "\n", "# Predict on test set\n", "y_pred_q, y_pred_a, y_pred_type = model.predict(X_test)\n", "\n", "# Decode predictions to class indices\n", "y_pred_q = np.argmax(y_pred_q, axis=-1)\n", "y_pred_a = np.argmax(y_pred_a, axis=-1)\n", "y_pred_type = np.argmax(y_pred_type, axis=-1)\n", "y_true_type = np.argmax(y_type_test, axis=-1)\n", "\n", "# Calculate token-level accuracy\n", "acc_q = token_level_accuracy(y_q_test, y_pred_q)\n", "acc_a = token_level_accuracy(y_a_test, y_pred_a)\n", "\n", "# Type classification report\n", "report_type = classification_report(y_true_type, y_pred_type, zero_division=0)\n", "\n", "# Print Results\n", "print(\"\\n=== Akurasi Detail ===\")\n", "print(f\"Question Accuracy (Token-level): {acc_q:.1f}\")\n", "print(f\"Answer Accuracy (Token-level) : {acc_a:.1f}\")\n", "print(f\"Type Accuracy (Class-level) : {np.mean(y_true_type == y_pred_type):.2f}\")\n", "# print(\"\\n=== Classification Report (TYPE) ===\")\n", "# print(report_type)" ] }, { "cell_type": "code", "execution_count": 99, "id": "d5ed106c", "metadata": {}, "outputs": [], "source": [ "\n", "# flat_true_a, flat_pred_a = flatten_valid(y_a_test, y_pred_a_class)\n", "# print(\"\\n=== Classification Report: ANSWER ===\")\n", "# print(classification_report(flat_true_a, flat_pred_a))\n" ] }, { "cell_type": "code", "execution_count": 100, "id": "aa3860de", "metadata": {}, "outputs": [], "source": [ "\n", "# print(\"\\n=== Classification Report: TYPE ===\")\n", "# print(classification_report(y_true_type_class, y_pred_type_class))" ] } ], "metadata": { "kernelspec": { "display_name": "myenv", "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": "3.10.16" } }, "nbformat": 4, "nbformat_minor": 5 }