{ "cells": [ { "cell_type": "code", "execution_count": 93, "id": "fb283f23", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Total flattened samples: 342\n" ] } ], "source": [ "import json\n", "from pathlib import Path\n", "from itertools import chain\n", "\n", "RAW = json.loads(\n", " Path(\"../dataset/dev_dataset_qg.json\").read_text()\n", ") # ← file contoh Anda\n", "\n", "samples = []\n", "for item in RAW:\n", " for qp in item[\"quiz_posibility\"]:\n", " samp = {\n", " \"tokens\": [tok.lower() for tok in item[\"tokens\"]],\n", " \"ner\": item[\"ner\"],\n", " \"srl\": item[\"srl\"],\n", " \"q_type\": qp[\"type\"], # isian / opsi / benar_salah\n", " \"q_toks\": [tok.lower() for tok in qp[\"question\"]]\n", " + [\"\"], # tambahkan \n", " }\n", " # Jawaban bisa multi token\n", " if isinstance(qp[\"answer\"], list):\n", " samp[\"a_toks\"] = [tok.lower() for tok in qp[\"answer\"]] + [\"\"]\n", " else:\n", " samp[\"a_toks\"] = [qp[\"answer\"].lower(), \"\"]\n", " samples.append(samp)\n", "\n", "print(\"Total flattened samples:\", len(samples))" ] }, { "cell_type": "code", "execution_count": 94, "id": "fa4f979d", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "{'': 0, '': 1, '': 2, '': 3, 'jepara': 4, 'false': 5, 'trowulan': 6, '17': 7, 'agustus': 8, '1945': 9, 'soekarno': 10, 'mohammad hatta': 11, '365': 12, 'hari': 13, 'merkurius': 14, 'true': 15, 'mars': 16, 'jupiter': 17, 'saturnus': 18, 'uranus': 19, 'neptunus': 20, '5': 21, 'januari': 22, '2020': 23, '12': 24, 'februari': 25, '2019': 26, '23': 27, 'maret': 28, '2021': 29, '1': 30, 'april': 31, '2022': 32, '15': 33, 'mei': 34, '2023': 35, 'gunung': 36, 'everest': 37, 'amazon': 38, 'piramida': 39, 'giza': 40, 'benua': 41, 'asia': 42, 'colosseum': 43, 'taj': 44, 'mahal': 45, 'petra': 46, 'tembok': 47, 'cina': 48, 'chichen': 49, 'itza': 50, 'patung': 51, 'yesus': 52, 'penebus': 53, 'machu': 54, 'picchu': 55, 'stonehenge': 56, 'menara': 57, 'pisa': 58, 'angkot': 59, 'wat': 60, '8848': 61, 'meter': 62, '17 agustus 1945': 63, 'albert': 64, 'einstein': 65, 'jantung': 66, 'memompa darah': 67, 'tokyo': 68, '100': 69, 'derajat': 70, 'celsius': 71, 'thomas': 72, 'alva': 73, 'edison': 74, '1879': 75, 'ketiga': 76, 'leonardo': 77, 'da': 78, 'vinci': 79, 'leonardo da vinci': 80, '9,46': 81, 'triliun': 82, 'kilometer': 83, 'mahatma': 84, 'gandhi': 85, '1958': 86, 'kornea': 87, 'waterloo': 88, '1815': 89, 'indonesia': 90, 'marie': 91, 'curie': 92, 'fisika dan kimia': 93, 'inka': 94, 'oksigen': 95, 'karbon dioksida dan air': 96, 'vincent': 97, 'van': 98, 'gogh': 99, 'double': 100, 'helix': 101, 'double helix': 102, 'alexander': 103, 'fleming': 104, 'jeruk': 105, 'dan': 106, 'kiwi': 107, 'vitamin c': 108, 'nikola': 109, 'tesla': 110, 'sungai': 111, 'nil': 112, '6650 kilometer': 113, 'paus': 114, 'biru': 115, 'pankreas': 116, 'mengatur gula darah': 117, 'charles': 118, 'darwin': 119, 'shah': 120, 'jahan': 121, 'mumtaz mahal': 122, '44.58 juta km²': 123, '54': 124, 'di selatan laut mediterania': 125, 'eropa': 126, '10.18 juta km²': 127, 'atlantik': 128, 'pasifik': 129, 'hutan amazon': 130, 'australia': 131, 'belahan bumi selatan': 132, 'antartika': 133, 'kutub selatan': 134, '4.7 miliar': 135, 'kilimanjaro': 136, '5,895 meter': 137, 'sahara': 138, 'afrika': 139, 'alpen': 140, '8': 141, 'superior': 142, 'danau superior': 143, 'amerika selatan': 144, 'ali': 145, 'turnamen': 146, 'nina': 147, 'rapat': 148, 'farhan': 149, 'andi': 150, 'workshop': 151, 'lina': 152, 'pameran': 153, 'iqbal': 154, 'siti': 155, 'perlombaan': 156, 'konser': 157, 'fajar': 158, 'dina': 159, 'festival': 160, 'rian': 161, 'bazar': 162, 'tari': 163, 'seminar': 164, 'kompetisi': 165, 'rudi': 166, 'putri': 167, 'budi': 168, 'hana': 169, 'raka': 170, 'dewi': 171, 'surabaya': 172, 'yogyakarta': 173, 'kota': 174, 'jakarta': 175, 'bandung': 176, 'malang': 177, 'bali': 178, 'padang': 179, 'ibukota': 180, 'makassar': 181, 'medan': 182}\n" ] } ], "source": [ "def build_vocab(seq_iter, reserved=[\"\", \"\", \"\", \"\"]):\n", " vocab = {tok: idx for idx, tok in enumerate(reserved)}\n", " for tok in chain.from_iterable(seq_iter):\n", " if tok not in vocab:\n", " vocab[tok] = len(vocab)\n", " return vocab\n", "\n", "\n", "vocab_tok = build_vocab((s[\"tokens\"] for s in samples))\n", "vocab_ner = build_vocab((s[\"ner\"] for s in samples), reserved=[\"\", \"\"])\n", "vocab_srl = build_vocab((s[\"srl\"] for s in samples), reserved=[\"\", \"\"])\n", "vocab_q = build_vocab((s[\"q_toks\"] for s in samples))\n", "vocab_a = build_vocab((s[\"a_toks\"] for s in samples))\n", "\n", "vocab_typ = {\"isian\": 0, \"opsi\": 1, \"true_false\": 2}\n", "\n", "print(vocab_a)" ] }, { "cell_type": "code", "execution_count": 95, "id": "d1a5b324", "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "from tensorflow.keras.preprocessing.sequence import pad_sequences\n", "\n", "\n", "def encode(seq, vmap): # token → id\n", " return [vmap.get(t, vmap[\"\"]) for t in seq]\n", "\n", "\n", "MAX_SENT = max(len(s[\"tokens\"]) for s in samples)\n", "MAX_Q = max(len(s[\"q_toks\"]) for s in samples)\n", "MAX_A = max(len(s[\"a_toks\"]) for s in samples)\n", "\n", "X_tok = pad_sequences(\n", " [encode(s[\"tokens\"], vocab_tok) for s in samples], maxlen=MAX_SENT, padding=\"post\"\n", ")\n", "X_ner = pad_sequences(\n", " [encode(s[\"ner\"], vocab_ner) for s in samples], maxlen=MAX_SENT, padding=\"post\"\n", ")\n", "X_srl = pad_sequences(\n", " [encode(s[\"srl\"], vocab_srl) for s in samples], maxlen=MAX_SENT, padding=\"post\"\n", ")\n", "\n", "# Decoder input = + target[:-1]\n", "dec_q_in = pad_sequences(\n", " [[vocab_q[\"\"], *encode(s[\"q_toks\"][:-1], vocab_q)] for s in samples],\n", " maxlen=MAX_Q,\n", " padding=\"post\",\n", ")\n", "dec_q_out = pad_sequences(\n", " [encode(s[\"q_toks\"], vocab_q) for s in samples], maxlen=MAX_Q, padding=\"post\"\n", ")\n", "\n", "dec_a_in = pad_sequences(\n", " [[vocab_a[\"\"], *encode(s[\"a_toks\"][:-1], vocab_a)] for s in samples],\n", " maxlen=MAX_A,\n", " padding=\"post\",\n", ")\n", "dec_a_out = pad_sequences(\n", " [encode(s[\"a_toks\"], vocab_a) for s in samples], maxlen=MAX_A, padding=\"post\"\n", ")\n", "\n", "MAX_SENT = max(len(s[\"tokens\"]) for s in samples)\n", "MAX_Q = max(len(s[\"q_toks\"]) for s in samples)\n", "MAX_A = max(len(s[\"a_toks\"]) for s in samples)\n", "y_type = np.array([vocab_typ[s[\"q_type\"]] for s in samples])" ] }, { "cell_type": "code", "execution_count": 96, "id": "ff5bd85f", "metadata": {}, "outputs": [ { "data": { "text/html": [ "
Model: \"functional_8\"\n",
       "
\n" ], "text/plain": [ "\u001b[1mModel: \"functional_8\"\u001b[0m\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓\n",
       "┃ Layer (type)        ┃ Output Shape      ┃    Param # ┃ Connected to      ┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩\n",
       "│ tok_in (InputLayer) │ (None, 16)        │          0 │ -                 │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ ner_in (InputLayer) │ (None, 16)        │          0 │ -                 │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ srl_in (InputLayer) │ (None, 16)        │          0 │ -                 │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_tok       │ (None, 16, 128)   │     57,856 │ tok_in[0][0]      │\n",
       "│ (Embedding)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_ner       │ (None, 16, 32)    │      1,248 │ ner_in[0][0]      │\n",
       "│ (Embedding)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_srl       │ (None, 16, 32)    │        448 │ srl_in[0][0]      │\n",
       "│ (Embedding)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ dec_q_in            │ (None, 13)        │          0 │ -                 │\n",
       "│ (InputLayer)        │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ concatenate_8       │ (None, 16, 192)   │          0 │ embedding_tok[0]… │\n",
       "│ (Concatenate)       │                   │            │ embedding_ner[0]… │\n",
       "│                     │                   │            │ embedding_srl[0]… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ dec_a_in            │ (None, 4)         │          0 │ -                 │\n",
       "│ (InputLayer)        │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_q_decoder │ (None, 13, 128)   │     52,096 │ dec_q_in[0][0]    │\n",
       "│ (Embedding)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ encoder_lstm (LSTM) │ [(None, 256),     │    459,776 │ concatenate_8[0]… │\n",
       "│                     │ (None, 256),      │            │                   │\n",
       "│                     │ (None, 256)]      │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ embedding_a_decoder │ (None, 4, 128)    │     23,424 │ dec_a_in[0][0]    │\n",
       "│ (Embedding)         │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ lstm_q_decoder      │ [(None, 13, 256), │    394,240 │ embedding_q_deco… │\n",
       "│ (LSTM)              │ (None, 256),      │            │ encoder_lstm[0][… │\n",
       "│                     │ (None, 256)]      │            │ encoder_lstm[0][… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ not_equal_32        │ (None, 13)        │          0 │ dec_q_in[0][0]    │\n",
       "│ (NotEqual)          │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ lstm_a_decoder      │ [(None, 4, 256),  │    394,240 │ embedding_a_deco… │\n",
       "│ (LSTM)              │ (None, 256),      │            │ encoder_lstm[0][… │\n",
       "│                     │ (None, 256)]      │            │ encoder_lstm[0][… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ not_equal_33        │ (None, 4)         │          0 │ dec_a_in[0][0]    │\n",
       "│ (NotEqual)          │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ q_output            │ (None, 13, 407)   │    104,599 │ lstm_q_decoder[0… │\n",
       "│ (TimeDistributed)   │                   │            │ not_equal_32[0][… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ a_output            │ (None, 4, 183)    │     47,031 │ lstm_a_decoder[0… │\n",
       "│ (TimeDistributed)   │                   │            │ not_equal_33[0][… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ type_output (Dense) │ (None, 3)         │        771 │ encoder_lstm[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_in (\u001b[38;5;33mInputLayer\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ - │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ ner_in (\u001b[38;5;33mInputLayer\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ - │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ srl_in (\u001b[38;5;33mInputLayer\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ - │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ embedding_tok │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m128\u001b[0m) │ \u001b[38;5;34m57,856\u001b[0m │ tok_in[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mEmbedding\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ embedding_ner │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m32\u001b[0m) │ \u001b[38;5;34m1,248\u001b[0m │ ner_in[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mEmbedding\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ embedding_srl │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m32\u001b[0m) │ \u001b[38;5;34m448\u001b[0m │ srl_in[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mEmbedding\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ dec_q_in │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m13\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ - │\n", "│ (\u001b[38;5;33mInputLayer\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ concatenate_8 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m16\u001b[0m, \u001b[38;5;34m192\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ embedding_tok[\u001b[38;5;34m0\u001b[0m]… │\n", "│ (\u001b[38;5;33mConcatenate\u001b[0m) │ │ │ embedding_ner[\u001b[38;5;34m0\u001b[0m]… │\n", "│ │ │ │ embedding_srl[\u001b[38;5;34m0\u001b[0m]… │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ dec_a_in │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m4\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ - │\n", "│ (\u001b[38;5;33mInputLayer\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ embedding_q_decoder │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m13\u001b[0m, \u001b[38;5;34m128\u001b[0m) │ \u001b[38;5;34m52,096\u001b[0m │ dec_q_in[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mEmbedding\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ encoder_lstm (\u001b[38;5;33mLSTM\u001b[0m) │ [(\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m), │ \u001b[38;5;34m459,776\u001b[0m │ concatenate_8[\u001b[38;5;34m0\u001b[0m]… │\n", "│ │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m), │ │ │\n", "│ │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m)] │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ embedding_a_decoder │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m4\u001b[0m, \u001b[38;5;34m128\u001b[0m) │ \u001b[38;5;34m23,424\u001b[0m │ dec_a_in[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mEmbedding\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ lstm_q_decoder │ [(\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m13\u001b[0m, \u001b[38;5;34m256\u001b[0m), │ \u001b[38;5;34m394,240\u001b[0m │ embedding_q_deco… │\n", "│ (\u001b[38;5;33mLSTM\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m), │ │ encoder_lstm[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "│ │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m)] │ │ encoder_lstm[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ not_equal_32 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m13\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ dec_q_in[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mNotEqual\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ lstm_a_decoder │ [(\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m4\u001b[0m, \u001b[38;5;34m256\u001b[0m), │ \u001b[38;5;34m394,240\u001b[0m │ embedding_a_deco… │\n", "│ (\u001b[38;5;33mLSTM\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m), │ │ encoder_lstm[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "│ │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m256\u001b[0m)] │ │ encoder_lstm[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ not_equal_33 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m4\u001b[0m) │ \u001b[38;5;34m0\u001b[0m │ dec_a_in[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m] │\n", "│ (\u001b[38;5;33mNotEqual\u001b[0m) │ │ │ │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ q_output │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m13\u001b[0m, \u001b[38;5;34m407\u001b[0m) │ \u001b[38;5;34m104,599\u001b[0m │ lstm_q_decoder[\u001b[38;5;34m0\u001b[0m… │\n", "│ (\u001b[38;5;33mTimeDistributed\u001b[0m) │ │ │ not_equal_32[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n", "│ a_output │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m4\u001b[0m, \u001b[38;5;34m183\u001b[0m) │ \u001b[38;5;34m47,031\u001b[0m │ lstm_a_decoder[\u001b[38;5;34m0\u001b[0m… │\n", "│ (\u001b[38;5;33mTimeDistributed\u001b[0m) │ │ │ not_equal_33[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\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 │ encoder_lstm[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m…\u001b[0m │\n", "└─────────────────────┴───────────────────┴────────────┴───────────────────┘\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
 Total params: 1,535,729 (5.86 MB)\n",
       "
\n" ], "text/plain": [ "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m1,535,729\u001b[0m (5.86 MB)\n" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ "
 Trainable params: 1,535,729 (5.86 MB)\n",
       "
\n" ], "text/plain": [ "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m1,535,729\u001b[0m (5.86 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" } ], "source": [ "import tensorflow as tf\n", "from tensorflow.keras.layers import (\n", " Input,\n", " Embedding,\n", " LSTM,\n", " Concatenate,\n", " Dense,\n", " TimeDistributed,\n", ")\n", "from tensorflow.keras.models import Model\n", "\n", "# ---- constants ---------------------------------------------------\n", "d_tok = 128 # token embedding dim\n", "d_tag = 32 # NER / SRL embedding dim\n", "units = 256\n", "\n", "# ---- encoder -----------------------------------------------------\n", "inp_tok = Input((MAX_SENT,), name=\"tok_in\")\n", "inp_ner = Input((MAX_SENT,), name=\"ner_in\")\n", "inp_srl = Input((MAX_SENT,), name=\"srl_in\")\n", "\n", "# make ALL streams mask the same way (here: no masking,\n", "# we'll just pad with 0s and let the LSTM ignore them)\n", "emb_tok = Embedding(len(vocab_tok), d_tok, mask_zero=False, name=\"embedding_tok\")(\n", " inp_tok\n", ")\n", "emb_ner = Embedding(len(vocab_ner), d_tag, mask_zero=False, name=\"embedding_ner\")(\n", " inp_ner\n", ")\n", "emb_srl = Embedding(len(vocab_srl), d_tag, mask_zero=False, name=\"embedding_srl\")(\n", " inp_srl\n", ")\n", "\n", "enc_concat = Concatenate()([emb_tok, emb_ner, emb_srl])\n", "enc_out, state_h, state_c = LSTM(units, return_state=True, name=\"encoder_lstm\")(\n", " enc_concat\n", ")\n", "\n", "\n", "# ---------- DECODER : Question ----------\n", "dec_q_inp = Input(shape=(MAX_Q,), name=\"dec_q_in\")\n", "dec_emb_q = Embedding(len(vocab_q), d_tok, mask_zero=True, name=\"embedding_q_decoder\")(\n", " dec_q_inp\n", ")\n", "dec_q, _, _ = LSTM(\n", " units, return_state=True, return_sequences=True, name=\"lstm_q_decoder\"\n", ")(dec_emb_q, initial_state=[state_h, state_c])\n", "q_out = TimeDistributed(\n", " Dense(len(vocab_q), activation=\"softmax\", name=\"dense_q_output\"), name=\"q_output\"\n", ")(dec_q)\n", "\n", "# ---------- DECODER : Answer ----------\n", "dec_a_inp = Input(shape=(MAX_A,), name=\"dec_a_in\")\n", "dec_emb_a = Embedding(len(vocab_a), d_tok, mask_zero=True, name=\"embedding_a_decoder\")(\n", " dec_a_inp\n", ")\n", "dec_a, _, _ = LSTM(\n", " units, return_state=True, return_sequences=True, name=\"lstm_a_decoder\"\n", ")(dec_emb_a, initial_state=[state_h, state_c])\n", "a_out = TimeDistributed(\n", " Dense(len(vocab_a), activation=\"softmax\", name=\"dense_a_output\"), name=\"a_output\"\n", ")(dec_a)\n", "\n", "# ---------- CLASSIFIER : Question Type ----------\n", "type_out = Dense(len(vocab_typ), activation=\"softmax\", name=\"type_output\")(enc_out)\n", "\n", "model = Model(\n", " inputs=[inp_tok, inp_ner, inp_srl, dec_q_inp, dec_a_inp],\n", " outputs=[q_out, a_out, type_out],\n", ")\n", "\n", "model.summary()" ] }, { "cell_type": "code", "execution_count": 97, "id": "fece1ae9", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Epoch 1/30\n", "\u001b[1m5/5\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m3s\u001b[0m 161ms/step - a_output_loss: 5.1540 - a_output_sparse_categorical_accuracy: 0.1507 - loss: 11.4761 - q_output_loss: 5.9970 - q_output_sparse_categorical_accuracy: 0.0600 - type_output_accuracy: 0.4506 - type_output_loss: 1.0728 - val_a_output_loss: 4.5900 - val_a_output_sparse_categorical_accuracy: 0.2500 - val_loss: 10.8292 - val_q_output_loss: 5.9316 - val_q_output_sparse_categorical_accuracy: 0.0769 - val_type_output_accuracy: 0.5143 - val_type_output_loss: 1.0253\n", "Epoch 2/30\n", "\u001b[1m5/5\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 48ms/step - a_output_loss: 4.2365 - a_output_sparse_categorical_accuracy: 0.2500 - loss: 10.2493 - q_output_loss: 5.6397 - q_output_sparse_categorical_accuracy: 0.1183 - type_output_accuracy: 0.5209 - type_output_loss: 1.2188 - val_a_output_loss: 3.2588 - val_a_output_sparse_categorical_accuracy: 0.2500 - val_loss: 9.0808 - val_q_output_loss: 5.4082 - val_q_output_sparse_categorical_accuracy: 0.0923 - val_type_output_accuracy: 0.5143 - val_type_output_loss: 1.3791\n", "Epoch 3/30\n", "\u001b[1m5/5\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 48ms/step - a_output_loss: 3.5259 - a_output_sparse_categorical_accuracy: 0.2500 - loss: 8.4974 - q_output_loss: 4.6444 - q_output_sparse_categorical_accuracy: 0.1174 - type_output_accuracy: 0.5233 - type_output_loss: 1.0788 - val_a_output_loss: 3.3879 - val_a_output_sparse_categorical_accuracy: 0.2500 - val_loss: 9.5209 - val_q_output_loss: 5.7546 - val_q_output_sparse_categorical_accuracy: 0.0769 - val_type_output_accuracy: 0.2000 - val_type_output_loss: 1.2615\n", "Epoch 4/30\n", "\u001b[1m5/5\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 48ms/step - a_output_loss: 3.3147 - a_output_sparse_categorical_accuracy: 0.2500 - loss: 8.1027 - q_output_loss: 4.4209 - q_output_sparse_categorical_accuracy: 0.1099 - type_output_accuracy: 0.3256 - type_output_loss: 1.2069 - val_a_output_loss: 3.0792 - val_a_output_sparse_categorical_accuracy: 0.2500 - val_loss: 9.2232 - val_q_output_loss: 5.8382 - val_q_output_sparse_categorical_accuracy: 0.0769 - val_type_output_accuracy: 0.5143 - val_type_output_loss: 1.0193\n", "Epoch 5/30\n", "\u001b[1m5/5\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 48ms/step - a_output_loss: 3.1559 - a_output_sparse_categorical_accuracy: 0.2500 - loss: 7.7733 - q_output_loss: 4.3048 - q_output_sparse_categorical_accuracy: 0.1120 - type_output_accuracy: 0.5160 - type_output_loss: 1.0414 - val_a_output_loss: 3.0450 - val_a_output_sparse_categorical_accuracy: 0.2500 - val_loss: 9.1657 - val_q_output_loss: 5.7943 - val_q_output_sparse_categorical_accuracy: 0.0923 - val_type_output_accuracy: 0.5143 - val_type_output_loss: 1.0881\n", "Epoch 6/30\n", "\u001b[1m5/5\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 48ms/step - a_output_loss: 3.0962 - a_output_sparse_categorical_accuracy: 0.2569 - loss: 7.6096 - q_output_loss: 4.1973 - q_output_sparse_categorical_accuracy: 0.1121 - type_output_accuracy: 0.5318 - type_output_loss: 1.0492 - val_a_output_loss: 3.1428 - val_a_output_sparse_categorical_accuracy: 0.3214 - val_loss: 9.2982 - val_q_output_loss: 5.8475 - val_q_output_sparse_categorical_accuracy: 0.0769 - val_type_output_accuracy: 0.5143 - val_type_output_loss: 1.0265\n" ] } ], "source": [ "losses = {\n", " \"q_output\": \"sparse_categorical_crossentropy\",\n", " \"a_output\": \"sparse_categorical_crossentropy\",\n", " \"type_output\": \"sparse_categorical_crossentropy\",\n", "}\n", "loss_weights = {\"q_output\": 1.0, \"a_output\": 1.0, \"type_output\": 0.3}\n", "\n", "model.compile(\n", " optimizer=\"adam\",\n", " loss=losses,\n", " loss_weights=loss_weights,\n", " metrics={\n", " \"q_output\": \"sparse_categorical_accuracy\",\n", " \"a_output\": \"sparse_categorical_accuracy\",\n", " \"type_output\": \"accuracy\",\n", " },\n", ")\n", "\n", "history = model.fit(\n", " [X_tok, X_ner, X_srl, dec_q_in, dec_a_in],\n", " [dec_q_out, dec_a_out, y_type],\n", " validation_split=0.1,\n", " epochs=30,\n", " batch_size=64,\n", " callbacks=[tf.keras.callbacks.EarlyStopping(patience=4, restore_best_weights=True)],\n", " verbose=1,\n", ")\n", "\n", "model.save(\"full_seq2seq.keras\")\n", "\n", "import json\n", "import pickle\n", "\n", "# def save_vocab(vocab, path):\n", "# with open(path, \"w\", encoding=\"utf-8\") as f:\n", "# json.dump(vocab, f, ensure_ascii=False, indent=2)\n", "\n", "# # Simpan semua vocab\n", "# save_vocab(vocab_tok, \"vocab_tok.json\")\n", "# save_vocab(vocab_ner, \"vocab_ner.json\")\n", "# save_vocab(vocab_srl, \"vocab_srl.json\")\n", "# save_vocab(vocab_q, \"vocab_q.json\")\n", "# save_vocab(vocab_a, \"vocab_a.json\")\n", "# save_vocab(vocab_typ, \"vocab_typ.json\")\n", "\n", "\n", "def save_vocab_pkl(vocab, path):\n", " with open(path, \"wb\") as f:\n", " pickle.dump(vocab, f)\n", "\n", "\n", "# Simpan semua vocab\n", "save_vocab_pkl(vocab_tok, \"vocab_tok.pkl\")\n", "save_vocab_pkl(vocab_ner, \"vocab_ner.pkl\")\n", "save_vocab_pkl(vocab_srl, \"vocab_srl.pkl\")\n", "save_vocab_pkl(vocab_q, \"vocab_q.pkl\")\n", "save_vocab_pkl(vocab_a, \"vocab_a.pkl\")\n", "save_vocab_pkl(vocab_typ, \"vocab_typ.pkl\")" ] }, { "cell_type": "code", "execution_count": 98, "id": "3355c0c7", "metadata": {}, "outputs": [], "source": [ "import tensorflow as tf\n", "import numpy as np\n", "import pickle\n", "from tensorflow.keras.models import load_model, Model\n", "from tensorflow.keras.layers import Input, Concatenate\n", "\n", "# === Load Model Utama ===\n", "model = load_model(\"full_seq2seq.keras\")\n", "\n", "\n", "# === Load Vocabulary dari .pkl ===\n", "def load_vocab(path):\n", " with open(path, \"rb\") as f:\n", " return pickle.load(f)\n", "\n", "\n", "vocab_tok = load_vocab(\"vocab_tok.pkl\")\n", "vocab_ner = load_vocab(\"vocab_ner.pkl\")\n", "vocab_srl = load_vocab(\"vocab_srl.pkl\")\n", "vocab_q = load_vocab(\"vocab_q.pkl\")\n", "vocab_a = load_vocab(\"vocab_a.pkl\")\n", "vocab_typ = load_vocab(\"vocab_typ.pkl\")\n", "\n", "inv_vocab_q = {v: k for k, v in vocab_q.items()}\n", "inv_vocab_a = {v: k for k, v in vocab_a.items()}\n", "\n", "# === Build Encoder Model ===\n", "MAX_SENT = model.input_shape[0][1] # Ambil shape dari model yang diload\n", "MAX_Q = model.input_shape[3][1] # Max length for question\n", "MAX_A = model.input_shape[4][1] # Max length for answer\n", "\n", "inp_tok_g = Input(shape=(MAX_SENT,), name=\"tok_in_g\")\n", "inp_ner_g = Input(shape=(MAX_SENT,), name=\"ner_in_g\")\n", "inp_srl_g = Input(shape=(MAX_SENT,), name=\"srl_in_g\")\n", "\n", "emb_tok = model.get_layer(\"embedding_tok\").call(inp_tok_g)\n", "emb_ner = model.get_layer(\"embedding_ner\").call(inp_ner_g)\n", "emb_srl = model.get_layer(\"embedding_srl\").call(inp_srl_g)\n", "\n", "enc_concat = Concatenate(name=\"concat_encoder\")([emb_tok, emb_ner, emb_srl])\n", "\n", "encoder_lstm = model.get_layer(\"encoder_lstm\")\n", "enc_out, state_h, state_c = encoder_lstm(enc_concat)\n", "\n", "# Create encoder model with full output including enc_out\n", "encoder_model = Model(\n", " inputs=[inp_tok_g, inp_ner_g, inp_srl_g],\n", " outputs=[enc_out, state_h, state_c],\n", " name=\"encoder_model\",\n", ")\n", "\n", "# === Build Decoder for Question ===\n", "dec_q_inp = Input(shape=(1,), name=\"dec_q_in\")\n", "dec_emb_q = model.get_layer(\"embedding_q_decoder\").call(dec_q_inp)\n", "\n", "state_h_dec = Input(shape=(256,), name=\"state_h_dec\")\n", "state_c_dec = Input(shape=(256,), name=\"state_c_dec\")\n", "\n", "lstm_decoder_q = model.get_layer(\"lstm_q_decoder\")\n", "\n", "dec_out_q, state_h_q, state_c_q = lstm_decoder_q(\n", " dec_emb_q, initial_state=[state_h_dec, state_c_dec]\n", ")\n", "\n", "q_time_dist_layer = model.get_layer(\"q_output\")\n", "dense_q = q_time_dist_layer.layer\n", "q_output = dense_q(dec_out_q)\n", "\n", "decoder_q = Model(\n", " inputs=[dec_q_inp, state_h_dec, state_c_dec],\n", " outputs=[q_output, state_h_q, state_c_q],\n", " name=\"decoder_question_model\",\n", ")\n", "\n", "# === Build Decoder for Answer ===\n", "dec_a_inp = Input(shape=(1,), name=\"dec_a_in\")\n", "dec_emb_a = model.get_layer(\"embedding_a_decoder\").call(dec_a_inp)\n", "\n", "state_h_a = Input(shape=(256,), name=\"state_h_a\")\n", "state_c_a = Input(shape=(256,), name=\"state_c_a\")\n", "\n", "lstm_decoder_a = model.get_layer(\"lstm_a_decoder\")\n", "\n", "dec_out_a, state_h_a_out, state_c_a_out = lstm_decoder_a(\n", " dec_emb_a, initial_state=[state_h_a, state_c_a]\n", ")\n", "\n", "a_time_dist_layer = model.get_layer(\"a_output\")\n", "dense_a = a_time_dist_layer.layer\n", "a_output = dense_a(dec_out_a)\n", "\n", "decoder_a = Model(\n", " inputs=[dec_a_inp, state_h_a, state_c_a],\n", " outputs=[a_output, state_h_a_out, state_c_a_out],\n", " name=\"decoder_answer_model\",\n", ")\n", "\n", "# === Build Classifier for Question Type ===\n", "type_dense = model.get_layer(\"type_output\")\n", "type_out = type_dense(enc_out)\n", "\n", "classifier_model = Model(\n", " inputs=[inp_tok_g, inp_ner_g, inp_srl_g], outputs=type_out, name=\"classifier_model\"\n", ")" ] }, { "cell_type": "code", "execution_count": 99, "id": "d406e6ff", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "Generated Question: menghadiri menghadiri ___ ___\n", "Generated Answer : \n", "Question Type : isian\n" ] } ], "source": [ "def encode(seq, vmap):\n", " return [vmap.get(tok, vmap[\"\"]) for tok in seq]\n", "\n", "\n", "def encode_and_pad(seq, vmap, max_len=MAX_SENT):\n", " encoded = [vmap.get(tok, vmap[\"\"]) for tok in seq]\n", " # Pad with vocab[\"\"] to the right if sequence is shorter than max_len\n", " padded = encoded + [vmap[\"\"]] * (max_len - len(encoded))\n", " return padded[:max_len] # Ensure it doesn't exceed max_len\n", "\n", "\n", "def greedy_decode(tokens, ner, srl, max_q=20, max_a=10):\n", " # --- encode encoder inputs -------------------------------------------\n", " if isinstance(tokens, np.ndarray):\n", " enc_tok = tokens\n", " enc_ner = ner\n", " enc_srl = srl\n", " else:\n", " enc_tok = np.array([encode_and_pad(tokens, vocab_tok, MAX_SENT)])\n", " enc_ner = np.array([encode_and_pad(ner, vocab_ner, MAX_SENT)])\n", " enc_srl = np.array([encode_and_pad(srl, vocab_srl, MAX_SENT)])\n", "\n", " # --- Get encoder outputs ---\n", " enc_out, h, c = encoder_model.predict([enc_tok, enc_ner, enc_srl], verbose=0)\n", "\n", " # QUESTION Decoding\n", " tgt = np.array([[vocab_q[\"\"]]])\n", " question_ids = []\n", " for _ in range(max_q):\n", " logits, h, c = decoder_q.predict([tgt, h, c], verbose=0)\n", " next_id = int(logits[0, 0].argmax()) # Get the predicted token ID\n", " if next_id == vocab_q[\"\"]:\n", " break\n", " question_ids.append(next_id)\n", " tgt = np.array([[next_id]]) # Feed the predicted token back as input\n", "\n", " # ANSWER Decoding - use encoder outputs again for fresh state\n", " _, h, c = encoder_model.predict([enc_tok, enc_ner, enc_srl], verbose=0)\n", " tgt = np.array([[vocab_a[\"\"]]])\n", " answer_ids = []\n", " for _ in range(max_a):\n", " logits, h, c = decoder_a.predict([tgt, h, c], verbose=0)\n", " next_id = int(logits[0, 0].argmax())\n", " if next_id == vocab_a[\"\"]:\n", " break\n", " answer_ids.append(next_id)\n", " tgt = np.array([[next_id]])\n", "\n", " # Question Type\n", " qtype_logits = classifier_model.predict([enc_tok, enc_ner, enc_srl], verbose=0)\n", " qtype_id = int(qtype_logits.argmax())\n", "\n", " # Final output\n", " question = [inv_vocab_q.get(i, \"\") for i in question_ids]\n", " answer = [inv_vocab_a.get(i, \"\") for i in answer_ids]\n", " q_type = [k for k, v in vocab_typ.items() if v == qtype_id][0]\n", "\n", " return question, answer, q_type\n", "\n", "\n", "def test_model():\n", " test_data = {\n", " \"tokens\": [\"nama\", \"lengkap\", \"saya\", \"Maya\", \"Maya\"],\n", " \"ner\": [\"O\", \"O\", \"O\", \"B-PER\", \"B-PER\"],\n", " \"srl\": [\"O\", \"O\", \"ARG0\", \"ARG0\", \"ARG0\"],\n", " }\n", " # tokens = [\n", " # \"soekarno\",\n", " # \"membacakan\",\n", " # \"teks\",\n", " # \"proklamasi\",\n", " # \"pada\",\n", " # \"17\",\n", " # \"agustus\",\n", " # \"1945\",\n", " # ]\n", " # ner_tags = [\"B-PER\", \"O\", \"O\", \"O\", \"O\", \"B-DATE\", \"I-DATE\", \"I-DATE\"]\n", " # srl_tags = [\"ARG0\", \"V\", \"ARG1\", \"ARG1\", \"O\", \"ARGM-TMP\", \"ARGM-TMP\", \"ARGM-TMP\"]\n", "\n", " question, answer, q_type = greedy_decode(\n", " test_data[\"tokens\"], test_data[\"ner\"], test_data[\"srl\"]\n", " )\n", " print(f\"Generated Question: {' '.join(question)}\")\n", " print(f\"Generated Answer : {' '.join(answer)}\")\n", " print(f\"Question Type : {q_type}\")\n", "\n", "\n", "test_model()" ] }, { "cell_type": "code", "execution_count": 100, "id": "5adde3c3", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "BLEU : 0.0385\n", "ROUGE1: 0.1052 | ROUGE-L: 0.1052\n" ] } ], "source": [ "from nltk.translate.bleu_score import sentence_bleu, SmoothingFunction\n", "from rouge_score import rouge_scorer\n", "\n", "smoothie = SmoothingFunction().method4\n", "scorer = rouge_scorer.RougeScorer([\"rouge1\", \"rougeL\"], use_stemmer=True)\n", "\n", "\n", "# Helper to strip special ids\n", "def strip_special(ids, vocab):\n", " pad = vocab[\"\"] if \"\" in vocab else None\n", " eos = vocab[\"\"]\n", " return [i for i in ids if i not in (pad, eos)]\n", "\n", "\n", "def ids_to_text(ids, inv_vocab):\n", " return \" \".join(inv_vocab[i] for i in ids)\n", "\n", "\n", "# ---- evaluation over a set of indices ----\n", "import random\n", "\n", "\n", "def evaluate(indices=None):\n", " if indices is None:\n", " indices = random.sample(range(len(X_tok)), k=min(100, len(X_tok)))\n", "\n", " bleu_scores, rou1, rouL = [], [], []\n", " for idx in indices:\n", " # Ground truth\n", " gt_q = strip_special(dec_q_out[idx], vocab_q)\n", " gt_a = strip_special(dec_a_out[idx], vocab_a)\n", " # Prediction\n", " q_pred, a_pred, _ = greedy_decode(\n", " X_tok[idx : idx + 1], X_ner[idx : idx + 1], X_srl[idx : idx + 1]\n", " )\n", "\n", " # BLEU on question tokens\n", " bleu_scores.append(\n", " sentence_bleu(\n", " [[inv_vocab_q[i] for i in gt_q]], q_pred, smoothing_function=smoothie\n", " )\n", " )\n", " # ROUGE on question strings\n", " r = scorer.score(ids_to_text(gt_q, inv_vocab_q), \" \".join(q_pred))\n", " rou1.append(r[\"rouge1\"].fmeasure)\n", " rouL.append(r[\"rougeL\"].fmeasure)\n", "\n", " print(f\"BLEU : {np.mean(bleu_scores):.4f}\")\n", " print(f\"ROUGE1: {np.mean(rou1):.4f} | ROUGE-L: {np.mean(rouL):.4f}\")\n", "\n", "\n", "evaluate()" ] } ], "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 }