{
"cells": [
{
"cell_type": "code",
"execution_count": 51,
"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\n",
"from collections import Counter"
]
},
{
"cell_type": "code",
"execution_count": 52,
"id": "50118278",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n",
" Jumlah data valid: 70 / 70\n",
" Jumlah data tidak valid: 0\n",
"Counter({'tof': 30, 'isian': 30, 'opsi': 10})\n"
]
}
],
"source": [
"# Load raw data\n",
"with open(\"qg_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",
"type_counts = Counter(types)\n",
"\n",
"print(type_counts)\n"
]
},
{
"cell_type": "code",
"execution_count": 53,
"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": 54,
"id": "555f9e22",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'isian', 'tof', 'opsi'}\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": 55,
"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": 56,
"id": "255e2a9a",
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"Model: \"functional_5\"\n",
"\n"
],
"text/plain": [
"\u001b[1mModel: \"functional_5\"\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_15 │ (None, None, 128) │ 41,600 │ tok_input[0][0] │\n",
"│ (Embedding) │ │ │ │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ embedding_16 │ (None, None, 16) │ 272 │ ner_input[0][0] │\n",
"│ (Embedding) │ │ │ │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ embedding_17 │ (None, None, 16) │ 272 │ srl_input[0][0] │\n",
"│ (Embedding) │ │ │ │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ concatenate_5 │ (None, None, 160) │ 0 │ embedding_15[0][… │\n",
"│ (Concatenate) │ │ │ embedding_16[0][… │\n",
"│ │ │ │ embedding_17[0][… │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ lstm_5 (LSTM) │ (None, None, 256) │ 427,008 │ concatenate_5[0]… │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ get_item_5 │ (None, 256) │ 0 │ lstm_5[0][0] │\n",
"│ (GetItem) │ │ │ │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ question_output │ (None, None, 272) │ 69,904 │ lstm_5[0][0] │\n",
"│ (TimeDistributed) │ │ │ │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ answer_output │ (None, None, 60) │ 15,420 │ lstm_5[0][0] │\n",
"│ (TimeDistributed) │ │ │ │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ type_output (Dense) │ (None, 3) │ 771 │ get_item_5[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_15 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m128\u001b[0m) │ \u001b[38;5;34m41,600\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_16 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m272\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_17 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m272\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_5 │ (\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_15[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n",
"│ (\u001b[38;5;33mConcatenate\u001b[0m) │ │ │ embedding_16[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n",
"│ │ │ │ embedding_17[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ lstm_5 (\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_5[\u001b[38;5;34m0\u001b[0m]… │\n",
"├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
"│ get_item_5 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ lstm_5[\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;34m272\u001b[0m) │ \u001b[38;5;34m69,904\u001b[0m │ lstm_5[\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;34m60\u001b[0m) │ \u001b[38;5;34m15,420\u001b[0m │ lstm_5[\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_5[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n",
"└─────────────────────┴───────────────────┴────────────┴───────────────────┘\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" Total params: 555,247 (2.12 MB)\n",
"\n"
],
"text/plain": [
"\u001b[1m Total params: \u001b[0m\u001b[38;5;34m555,247\u001b[0m (2.12 MB)\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"data": {
"text/html": [
" Trainable params: 555,247 (2.12 MB)\n",
"\n"
],
"text/plain": [
"\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m555,247\u001b[0m (2.12 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[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 3s/step - answer_output_accuracy: 0.0030 - answer_output_loss: 4.1163 - loss: 10.8193 - question_output_accuracy: 0.0030 - question_output_loss: 5.6031 - type_output_accuracy: 0.2000 - type_output_loss: 1.0999 - val_answer_output_accuracy: 0.8833 - val_answer_output_loss: 4.0123 - val_loss: 10.6706 - val_question_output_accuracy: 0.6000 - val_question_output_loss: 5.5595 - val_type_output_accuracy: 0.1667 - val_type_output_loss: 1.0987\n",
"Epoch 2/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 95ms/step - answer_output_accuracy: 0.8800 - answer_output_loss: 4.0174 - loss: 10.6778 - question_output_accuracy: 0.5640 - question_output_loss: 5.5631 - type_output_accuracy: 0.4200 - type_output_loss: 1.0973 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 3.8939 - val_loss: 10.4860 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 5.4945 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0976\n",
"Epoch 3/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 91ms/step - answer_output_accuracy: 0.9370 - answer_output_loss: 3.9075 - loss: 10.5064 - question_output_accuracy: 0.5870 - question_output_loss: 5.5043 - type_output_accuracy: 0.6200 - type_output_loss: 1.0946 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 3.7157 - val_loss: 10.1938 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 5.3815 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0965\n",
"Epoch 4/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 90ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 3.7435 - loss: 10.2381 - question_output_accuracy: 0.5890 - question_output_loss: 5.4027 - type_output_accuracy: 0.6200 - type_output_loss: 1.0919 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 3.4257 - val_loss: 9.7085 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 5.1873 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0955\n",
"Epoch 5/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 90ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 3.4788 - loss: 9.7970 - question_output_accuracy: 0.5850 - question_output_loss: 5.2288 - type_output_accuracy: 0.6600 - type_output_loss: 1.0894 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 2.9617 - val_loss: 8.9146 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 4.8585 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0944\n",
"Epoch 6/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 95ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 3.0565 - loss: 9.0790 - question_output_accuracy: 0.5850 - question_output_loss: 4.9355 - type_output_accuracy: 0.6600 - type_output_loss: 1.0869 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 2.3649 - val_loss: 7.8024 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 4.3441 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0933\n",
"Epoch 7/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 95ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 2.5004 - loss: 8.0585 - question_output_accuracy: 0.5850 - question_output_loss: 4.4735 - type_output_accuracy: 0.6600 - type_output_loss: 1.0845 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 1.8898 - val_loss: 6.6823 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 3.7005 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0920\n",
"Epoch 8/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 88ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 2.0239 - loss: 6.9823 - question_output_accuracy: 0.5850 - question_output_loss: 3.8764 - type_output_accuracy: 0.6600 - type_output_loss: 1.0821 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 1.5873 - val_loss: 5.7713 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 3.0934 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0906\n",
"Epoch 9/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 93ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 1.6939 - loss: 6.0594 - question_output_accuracy: 0.5850 - question_output_loss: 3.2857 - type_output_accuracy: 0.6600 - type_output_loss: 1.0798 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 1.3585 - val_loss: 5.0778 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.6303 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0890\n",
"Epoch 10/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 97ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 1.4268 - loss: 5.3244 - question_output_accuracy: 0.5850 - question_output_loss: 2.8203 - type_output_accuracy: 0.6600 - type_output_loss: 1.0774 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 1.1559 - val_loss: 4.5630 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.3200 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0871\n",
"Epoch 11/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 93ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 1.1880 - loss: 4.7795 - question_output_accuracy: 0.5850 - question_output_loss: 2.5167 - type_output_accuracy: 0.6600 - type_output_loss: 1.0748 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.9716 - val_loss: 4.2001 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.1437 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0848\n",
"Epoch 12/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 95ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.9857 - loss: 4.4356 - question_output_accuracy: 0.5850 - question_output_loss: 2.3778 - type_output_accuracy: 0.6600 - type_output_loss: 1.0721 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.8171 - val_loss: 3.9799 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.0807 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0822\n",
"Epoch 13/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 92ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.8280 - loss: 4.2715 - question_output_accuracy: 0.5850 - question_output_loss: 2.3745 - type_output_accuracy: 0.6600 - type_output_loss: 1.0690 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.6995 - val_loss: 3.8760 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.0974 - val_type_output_accuracy: 0.3333 - val_type_output_loss: 1.0790\n",
"Epoch 14/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 91ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.7110 - loss: 4.2264 - question_output_accuracy: 0.5850 - question_output_loss: 2.4498 - type_output_accuracy: 0.6400 - type_output_loss: 1.0656 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.6143 - val_loss: 3.8415 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.1518 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0754\n",
"Epoch 15/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 95ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.6254 - loss: 4.2353 - question_output_accuracy: 0.5850 - question_output_loss: 2.5482 - type_output_accuracy: 0.6000 - type_output_loss: 1.0617 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.5529 - val_loss: 3.8335 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.2091 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0714\n",
"Epoch 16/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 94ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.5623 - loss: 4.2530 - question_output_accuracy: 0.5850 - question_output_loss: 2.6334 - type_output_accuracy: 0.6000 - type_output_loss: 1.0573 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.5083 - val_loss: 3.8255 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.2502 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0670\n",
"Epoch 17/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 96ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.5150 - loss: 4.2561 - question_output_accuracy: 0.5850 - question_output_loss: 2.6886 - type_output_accuracy: 0.6000 - type_output_loss: 1.0525 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.4752 - val_loss: 3.8053 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.2678 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0623\n",
"Epoch 18/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 104ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.4790 - loss: 4.2357 - question_output_accuracy: 0.5850 - question_output_loss: 2.7094 - type_output_accuracy: 0.6000 - type_output_loss: 1.0473 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.4503 - val_loss: 3.7689 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.2612 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0573\n",
"Epoch 19/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 94ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.4511 - loss: 4.1904 - question_output_accuracy: 0.5850 - question_output_loss: 2.6974 - type_output_accuracy: 0.5600 - type_output_loss: 1.0419 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.4313 - val_loss: 3.7162 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.2327 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0523\n",
"Epoch 20/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 94ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.4293 - loss: 4.1218 - question_output_accuracy: 0.5850 - question_output_loss: 2.6564 - type_output_accuracy: 0.5600 - type_output_loss: 1.0361 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.4164 - val_loss: 3.6494 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.1859 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0471\n",
"Epoch 21/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 93ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.4118 - loss: 4.0332 - question_output_accuracy: 0.5850 - question_output_loss: 2.5912 - type_output_accuracy: 0.5600 - type_output_loss: 1.0302 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.4046 - val_loss: 3.5722 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.1256 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0420\n",
"Epoch 22/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 95ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.3975 - loss: 3.9297 - question_output_accuracy: 0.5850 - question_output_loss: 2.5080 - type_output_accuracy: 0.5600 - type_output_loss: 1.0242 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.3951 - val_loss: 3.4909 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.0587 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0370\n",
"Epoch 23/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 94ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.3858 - loss: 3.8184 - question_output_accuracy: 0.5850 - question_output_loss: 2.4147 - type_output_accuracy: 0.5600 - type_output_loss: 1.0180 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.3875 - val_loss: 3.4143 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 1.9948 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0321\n",
"Epoch 24/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 91ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.3759 - loss: 3.7097 - question_output_accuracy: 0.5850 - question_output_loss: 2.3222 - type_output_accuracy: 0.5600 - type_output_loss: 1.0116 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.3812 - val_loss: 3.3557 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 1.9473 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0273\n",
"Epoch 25/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 95ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.3674 - loss: 3.6180 - question_output_accuracy: 0.5850 - question_output_loss: 2.2455 - type_output_accuracy: 0.5600 - type_output_loss: 1.0051 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.3759 - val_loss: 3.3316 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 1.9330 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0227\n",
"Epoch 26/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 96ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.3600 - loss: 3.5615 - question_output_accuracy: 0.5850 - question_output_loss: 2.2030 - type_output_accuracy: 0.5400 - type_output_loss: 0.9985 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.3715 - val_loss: 3.3519 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 1.9622 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0183\n",
"Epoch 27/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 90ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.3534 - loss: 3.5516 - question_output_accuracy: 0.5850 - question_output_loss: 2.2064 - type_output_accuracy: 0.5400 - type_output_loss: 0.9917 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.3677 - val_loss: 3.4014 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.0195 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0141\n",
"Epoch 28/30\n",
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 96ms/step - answer_output_accuracy: 0.9380 - answer_output_loss: 0.3474 - loss: 3.5737 - question_output_accuracy: 0.5850 - question_output_loss: 2.2414 - type_output_accuracy: 0.5400 - type_output_loss: 0.9848 - val_answer_output_accuracy: 0.9250 - val_answer_output_loss: 0.3645 - val_loss: 3.4429 - val_question_output_accuracy: 0.6250 - val_question_output_loss: 2.0682 - val_type_output_accuracy: 0.5000 - val_type_output_loss: 1.0102\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=64,\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": 57,
"id": "06fd86c7",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\u001b[1m1/1\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 239ms/step\n",
"\n",
"=== Akurasi Detail ===\n",
"Question Accuracy (Token-level): 0.0000\n",
"Answer Accuracy (Token-level) : 0.0000\n",
"Type Accuracy (Class-level) : 0.29\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:.4f}\")\n",
"print(f\"Answer Accuracy (Token-level) : {acc_a:.4f}\")\n",
"print(f\"Type Accuracy (Class-level) : {np.mean(y_true_type == y_pred_type):.2f}\")"
]
},
{
"cell_type": "code",
"execution_count": 58,
"id": "b17b6470",
"metadata": {},
"outputs": [],
"source": [
"# import sacrebleu\n",
"# from sacrebleu.metrics import BLEU # optional kalau mau smoothing/effective_order\n",
"\n",
"# idx2tok = {v:k for k,v in word2idx.items()}\n",
"# PAD_ID = word2idx[\"PAD\"]\n",
"# SOS_ID = word2idx.get(\"SOS\", None)\n",
"# EOS_ID = word2idx.get(\"EOS\", None)\n",
"\n",
"# def seq2str(seq):\n",
"# \"\"\"Konversi list index -> kalimat string, sambil buang token spesial.\"\"\"\n",
"# toks = [idx2tok[i] for i in seq\n",
"# if i not in {PAD_ID, SOS_ID, EOS_ID}]\n",
"# return \" \".join(toks).strip().lower()\n",
"\n",
"# bleu_metric = BLEU(effective_order=True) # lebih stabil utk kalimat pendek\n",
"\n",
"# def bleu_corpus(pred_seqs, true_seqs):\n",
"# preds = [seq2str(p) for p in pred_seqs]\n",
"# refs = [[seq2str(t)] for t in true_seqs] # list‑of‑list, satu ref/kalimat\n",
"# return bleu_metric.corpus_score(preds, refs).score\n"
]
},
{
"cell_type": "code",
"execution_count": 59,
"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": 60,
"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
}