diff --git a/Neural graph module/ngm.ipynb b/Neural graph module/ngm.ipynb
index a924791ec8388b0e67e4a9635a89828d25d7f9bb..d95c77016b150ff219acb1da22657a5994bbad3b 100644
--- a/Neural graph module/ngm.ipynb
+++ b/Neural graph module/ngm.ipynb
@@ -2,7 +2,7 @@
"cells": [
{
"cell_type": "code",
- "execution_count": 17,
+ "execution_count": 25,
"metadata": {},
"outputs": [],
"source": [
@@ -15,21 +15,23 @@
"from transformers import BertTokenizer, BertModel\n",
"from transformers import Seq2SeqTrainer, Seq2SeqTrainingArguments\n",
"from tqdm import tqdm\n",
- "import json\n"
+ "import json\n",
+ "import requests\n"
]
},
{
"cell_type": "code",
- "execution_count": 18,
+ "execution_count": 26,
"metadata": {},
"outputs": [],
"source": [
- "tokenizer = BertTokenizer.from_pretrained(\"bert-base-uncased\")"
+ "mod = \"bert-base-uncased\"\n",
+ "tokenizer = BertTokenizer.from_pretrained(mod)"
]
},
{
"cell_type": "code",
- "execution_count": 19,
+ "execution_count": 27,
"metadata": {},
"outputs": [
{
@@ -49,7 +51,7 @@
},
{
"cell_type": "code",
- "execution_count": 20,
+ "execution_count": 28,
"metadata": {},
"outputs": [],
"source": [
@@ -57,8 +59,12 @@
"class NgmOne(nn.Module):\n",
" def __init__(self, device, relations):\n",
" super(NgmOne, self).__init__()\n",
- " self.tokenizer = BertTokenizer.from_pretrained(\"bert-base-uncased\")\n",
- " self.bert = BertModel.from_pretrained(\"bert-base-uncased\").to(device)\n",
+ " with torch.no_grad():\n",
+ " self.tokenizer = BertTokenizer.from_pretrained(mod)\n",
+ " self.bert = BertModel.from_pretrained(mod).to(device)\n",
+ "\n",
+ " for param in self.bert.parameters():\n",
+ " param.requires_grad = False\n",
" self.linear = nn.Linear(768, len(relations)).to(device)\n",
" self.softmax = nn.Softmax(dim=1).to(device)\n",
" self.device = device\n",
@@ -68,8 +74,7 @@
" tokenized_mask = tokenized_mask.to(self.device)\n",
" with torch.no_grad():\n",
" x = self.bert.forward(tokenized_seq, attention_mask=tokenized_mask)\n",
- " x = x[0][:,0,:].to(self.device)\n",
- " \n",
+ " x = x[0][:,0,:].to(self.device)\n",
" x = self.linear(x)\n",
" x = self.softmax(x)\n",
" return x"
@@ -77,7 +82,7 @@
},
{
"cell_type": "code",
- "execution_count": 21,
+ "execution_count": 29,
"metadata": {},
"outputs": [],
"source": [
@@ -98,6 +103,25 @@
" \"http://www.w3.org/2004/02/skos/core#\": \"skos:\",\n",
"}\n",
"\n",
+ "prefixes_reverse = {v: k for k, v in prefixes.items()}\n",
+ "\n",
+ "# {\n",
+ "# \"res:\": \"http://dbpedia.org/resource/\",\n",
+ "# \"dbo:\": \"http://dbpedia.org/ontology/\",\n",
+ "# \"dbp:\": \"http://dbpedia.org/property/\",\n",
+ "# \"rdfs:\": \"http://www.w3.org/2000/01/rdf-schema#\",\n",
+ "# \"rdf:\": \"http://www.w3.org/1999/02/22-rdf-syntax-ns#\",\n",
+ "# \"yago:\": \"http://dbpedia.org/class/yago/\",\n",
+ "# \"wdt:\": \"http://www.wikidata.org/prop/direct/\",\n",
+ "# \"wd:\": \"http://www.wikidata.org/entity/\",\n",
+ "# \"p:\": \"http://www.wikidata.org/prop/\",\n",
+ "# \"ps:\": \"https://w3id.org/payswarm#\",\n",
+ "# \"pq:\": \"http://www.wikidata.org/prop/qualifier/\",\n",
+ "# \"bd:\": \"http://www.bigdata.com/rdf#\",\n",
+ "# \"wikibase:\": \"http://wikiba.se/ontology#\",\n",
+ "# \"skos:\": \"http://www.w3.org/2004/02/skos/core#\",\n",
+ "# }\n",
+ "\n",
"prefixes_end = [\"org\", \"se\", \"com\", \"nu\"]\n",
"ALL_HTTP_PREFIXES = False\n",
"\n",
@@ -119,6 +143,7 @@
" \n",
" inputs = []\n",
" correct_rels = []\n",
+ " sub_obj_ents = []\n",
" inputs_max_len = 0\n",
" for d in tqdm(pred[\"questions\"]):\n",
" question = d[\"question\"][0][\"string\"]\n",
@@ -171,17 +196,19 @@
" if triplet_i[0] is not None and triplet_i[1] is not None:\n",
" #seq += \"[SUB] [SEP] \" + triplet[0]\n",
" # , padding=True, truncation=True)\n",
- " tokenized_seq = tokenizer(question, \"[SUB]\", triplet_i[0], padding=True, truncation=True)\n",
+ " tokenized_seq = tokenizer(question.lower(), \"[OBJ] [SEP] \" + triplet_i[0].split(\":\")[1].lower(), padding=True, truncation=True)\n",
+ " sub_obj_ents.append(\"[SUB] \" + prefixes_reverse[\"\".join(triplet_i[0].split(\":\")[0]) + \":\"] + triplet_i[0].split(\":\")[1])\n",
" elif triplet_i[2] is not None and triplet_i[1] is not None:\n",
" #seq += \"[OBJ] [SEP] \" + triplet[2]\n",
- " tokenized_seq = tokenizer(question, \"[OBJ]\", triplet_i[2], padding=True, truncation=True)\n",
+ " tokenized_seq = tokenizer(question.lower(), \"[OBJ] [SEP] \" + triplet_i[2].split(\":\")[1].lower(), padding=True, truncation=True)\n",
+ " sub_obj_ents.append(\"[OBJ] \" + prefixes_reverse[\"\".join(triplet_i[2].split(\":\")[0]) + \":\"] + triplet_i[2].split(\":\")[1])\n",
" else:\n",
" continue\n",
"\n",
" if inputs_max_len < len(tokenized_seq[\"input_ids\"]):\n",
" inputs_max_len = len(tokenized_seq[\"input_ids\"])\n",
" inputs.append(list(tokenized_seq.values())[0])\n",
- " correct_rels.append(triplet_i[1])\n",
+ " correct_rels.append(triplet_i[1].lower())\n",
"\n",
" inputs_padded = np.array([i + [0]*(inputs_max_len-len(i)) for i in inputs])\n",
" #correct_rels_padded = np.array([i + [0]*(correct_rels_max_len-len(i)) for i in correct_rels])\n",
@@ -189,34 +216,35 @@
" inputs_attention_mask = np.where(inputs_padded != 0, 1, 0)\n",
" #correct_rels_attention_mask = np.where(correct_rels_padded != 0, 1, 0)\n",
" print(\"Finished with batches\")\n",
- " return torch.LongTensor(inputs_padded), torch.LongTensor(inputs_attention_mask), correct_rels #torch.IntTensor(correct_rels_padded), torch.LongTensor(correct_rels_attention_mask)\n"
+ " return torch.LongTensor(inputs_padded), torch.LongTensor(inputs_attention_mask), correct_rels, sub_obj_ents #torch.IntTensor(correct_rels_padded), torch.LongTensor(correct_rels_attention_mask)\n"
]
},
{
"cell_type": "code",
- "execution_count": 22,
+ "execution_count": 30,
"metadata": {},
"outputs": [],
"source": [
"from torch.utils.data import Dataset, DataLoader\n",
"\n",
"class MyDataset(Dataset):\n",
- " def __init__(self, inputs, attention_mask, correct_rels, relations):\n",
+ " def __init__(self, inputs, attention_mask, correct_rels,ents, relations):\n",
" self.inputs = inputs\n",
" self.attention_mask = attention_mask\n",
" self.correct_rels = correct_rels\n",
+ " self.ents = ents\n",
" self.relations = relations\n",
"\n",
" def __len__(self):\n",
" return len(self.inputs)\n",
"\n",
" def __getitem__(self, idx):\n",
- " return self.inputs[idx], self.attention_mask[idx], self.relations.index(self.correct_rels[idx])"
+ " return self.inputs[idx], self.attention_mask[idx], self.relations.index(self.correct_rels[idx]), self.ents[idx]"
]
},
{
"cell_type": "code",
- "execution_count": 23,
+ "execution_count": 31,
"metadata": {},
"outputs": [
{
@@ -230,7 +258,7 @@
"name": "stderr",
"output_type": "stream",
"text": [
- "100%|██████████| 2052/2052 [00:03<00:00, 546.37it/s]\n"
+ "100%|██████████| 2052/2052 [00:00<00:00, 2906.57it/s]\n"
]
},
{
@@ -238,16 +266,17 @@
"output_type": "stream",
"text": [
"Finished with batches\n",
- "features: tensor([[ 101, 2054, 2003, 1996, 13314, 1997, 16122, 10230, 5400, 2100,\n",
- " 12378, 2136, 1029, 102, 1031, 4942, 1033, 102, 0, 0,\n",
+ "features: ['[CLS] name the home stadium of fc spartak moscow season 2011 - 12 [SEP] [ obj ] [SEP] 2011 – 12 _ fc _ spartak _ moscow _ season [SEP] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD] [PAD]'] mask: tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
+ " 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
+ " 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]) label_index tensor(431)\n",
+ "valid features: tensor([[ 101, 2054, 3063, 2001, 3378, 4876, 2007, 1996, 6436, 26785,\n",
+ " 7971, 6447, 1998, 2600, 1037, 20160, 10362, 1029, 102, 1031,\n",
+ " 27885, 3501, 1033, 102, 2600, 1011, 1037, 1011, 20160, 10362,\n",
+ " 102, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
" 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
- " 0, 0, 0, 0, 0, 0, 0]]) mask: tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0,\n",
- " 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]) label_index tensor(378)\n",
- "valid features: tensor([[ 101, 2054, 4752, 1997, 7992, 13843, 5112, 2267, 2003, 2036,\n",
- " 1996, 9353, 4215, 26432, 2278, 3037, 1997, 2703, 27668, 3077,\n",
- " 1029, 102, 1031, 4942, 1033, 102, 0, 0, 0, 0,\n",
- " 0, 0, 0, 0, 0, 0, 0]]) valid mask: tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
- " 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]) valid label_index tensor(28)\n"
+ " 0, 0, 0, 0, 0, 0, 0, 0]]) valid mask: tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,\n",
+ " 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
+ " 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]]) valid label_index tensor(149)\n"
]
}
],
@@ -259,18 +288,19 @@
" with open(file, \"r\") as f:\n",
" return json.load(f)\n",
"\n",
+ "\n",
"#relations = open_json(\"../data/relations-query-qald-9-linked.json\")\n",
- "relations = open_json(\"../data/relations-all-lc-quad-no-http.json\")\n",
+ "relations = open_json(\"../data/relations-all-no-http-lowercase.json\")\n",
"\n",
"# \"../data/qald-9-train-linked.json\"\n",
"#pred = \"../LC-QuAD/combined-requeried-linked-train.json\"\n",
- "inputs, attention_mask, correct_rels = make_batch(src=\"../LC-QuAD/combined-requeried-linked-train.json\", http_prefix = True) #train\n",
+ "inputs, attention_mask, correct_rels, sub_objs = make_batch(src=\"../LC-QuAD/combined-requeried-linked-train.json\", http_prefix = True) #train\n",
"\n",
"# relations = open_json(\"../data/relations-lcquad-without-http-train-linked.json\")\n",
"# train_set = MyDataset(*make_batch(), relations=relations)\n",
"\n",
"\n",
- "dataset = MyDataset(inputs, attention_mask, correct_rels, relations=relations)\n",
+ "dataset = MyDataset(inputs, attention_mask, correct_rels,sub_objs, relations=relations)\n",
"train_size = int(0.8 * len(dataset))\n",
"valid_size = len(dataset) - train_size\n",
"\n",
@@ -278,94 +308,65 @@
"\n",
"train_dataloader = DataLoader(train_data, batch_size=1, shuffle=True)\n",
"#show first entry\n",
- "train_features, train_mask, train_label = next(iter(train_dataloader))\n",
- "print(\"features:\", train_features, \"mask:\",train_mask,\"label_index\", train_label[0])\n",
+ "train_features, train_mask, train_label, ents = next(iter(train_dataloader))\n",
+ "print(\"features:\", tokenizer.batch_decode(train_features), \"mask:\",train_mask,\"label_index\", train_label[0])\n",
"\n",
"\n",
"valid_dataloader = DataLoader(valid_data, batch_size=1, shuffle=True)\n",
- "valid_features, valid_mask, valid_label = next(iter(valid_dataloader))\n",
+ "valid_features, valid_mask, valid_label, ents = next(iter(valid_dataloader))\n",
"print(\"valid features:\", valid_features, \"valid mask:\",valid_mask,\"valid label_index\", valid_label[0])"
]
},
{
"cell_type": "code",
- "execution_count": 24,
+ "execution_count": 32,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
- "Some weights of the model checkpoint at bert-base-uncased were not used when initializing BertModel: ['cls.seq_relationship.bias', 'cls.seq_relationship.weight', 'cls.predictions.decoder.weight', 'cls.predictions.bias', 'cls.predictions.transform.LayerNorm.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.predictions.transform.dense.weight', 'cls.predictions.transform.dense.bias']\n",
+ "Some weights of the model checkpoint at bert-base-uncased were not used when initializing BertModel: ['cls.predictions.transform.dense.bias', 'cls.predictions.decoder.weight', 'cls.seq_relationship.weight', 'cls.seq_relationship.bias', 'cls.predictions.transform.LayerNorm.bias', 'cls.predictions.bias', 'cls.predictions.transform.LayerNorm.weight', 'cls.predictions.transform.dense.weight']\n",
"- This IS expected if you are initializing BertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).\n",
"- This IS NOT expected if you are initializing BertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).\n"
]
}
],
"source": [
+ "SPARQL_ENDPOINT = \"https://dbpedia.org/sparql\"\n",
+ "\n",
+ "\n",
+ "# test_data_file = sys.argv[1]\n",
+ "# predicted_data_file = sys.argv[2]\n",
+ "\n",
+ "headers = {\n",
+ " 'Accept': 'application/sparql-results+json',\n",
+ " 'Content-Type': 'application/x-www-form-urlencoded',\n",
+ "}\n",
+ "\n",
"# Initialize model\n",
"model = NgmOne(device, relations)"
]
},
{
"cell_type": "code",
- "execution_count": 25,
+ "execution_count": 33,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
- "1 Train 6.453075773575726 , Valid 7.615683317184448\n",
- "2 Train 6.444019513971665 , Valid 7.6052809953689575\n",
- "3 Train 6.43224432889153 , Valid 7.602278828620911\n",
- "4 Train 6.415224019218893 , Valid 7.598743319511414\n",
- "5 Train 6.391957339118509 , Valid 7.584347605705261\n",
- "6 Train 6.368611419902129 , Valid 7.573382139205933\n",
- "7 Train 6.339372747084674 , Valid 7.557032465934753\n",
- "8 Train 6.318648646859562 , Valid 7.539492607116699\n",
- "9 Train 6.297140289755428 , Valid 7.523481965065002\n",
- "10 Train 6.281140131108901 , Valid 7.519673824310303\n",
- "11 Train 6.265462370479808 , Valid 7.5268988609313965\n",
- "12 Train 6.24550900739782 , Valid 7.504716753959656\n",
- "13 Train 6.234297051149256 , Valid 7.505620360374451\n",
- "14 Train 6.220176332137164 , Valid 7.500500917434692\n",
- "15 Train 6.204408701728372 , Valid 7.493185758590698\n",
- "16 Train 6.194800797630759 , Valid 7.488588452339172\n",
- "17 Train 6.183391935685101 , Valid 7.466232180595398\n",
- "18 Train 6.173011443194221 , Valid 7.457393527030945\n",
- "19 Train 6.168144422418931 , Valid 7.45364773273468\n",
- "20 Train 6.157315562753117 , Valid 7.452357649803162\n",
- "21 Train 6.140512690824621 , Valid 7.456966400146484\n",
- "22 Train 6.129443336935604 , Valid 7.464043855667114\n",
- "23 Train 6.12704220940085 , Valid 7.443313360214233\n",
- "24 Train 6.117525269003475 , Valid 7.4395164251327515\n",
- "25 Train 6.116069064420812 , Valid 7.456610083580017\n",
- "26 Train 6.108902790967156 , Valid 7.430782794952393\n",
- "27 Train 6.108471814323874 , Valid 7.44858992099762\n",
- "28 Train 6.100216697244083 , Valid 7.457515120506287\n",
- "29 Train 6.099361447726979 , Valid 7.438013672828674\n",
- "30 Train 6.092377662658691 , Valid 7.448408484458923\n",
- "31 Train 6.088698302998262 , Valid 7.442046403884888\n",
- "32 Train 6.083712998558493 , Valid 7.420018911361694\n",
- "33 Train 6.081799563239603 , Valid 7.426819205284119\n",
- "34 Train 6.070652428795309 , Valid 7.426627039909363\n",
- "35 Train 6.069005825940301 , Valid 7.425306558609009\n",
- "36 Train 6.059389002182904 , Valid 7.415045142173767\n",
- "37 Train 6.0618573918062095 , Valid 7.418038249015808\n",
- "38 Train 6.055309716392966 , Valid 7.434216380119324\n",
- "39 Train 6.049336994395537 , Valid 7.43299674987793\n",
- "40 Train 6.048270281623392 , Valid 7.410753607749939\n",
- "41 Train 6.043416135451373 , Valid 7.410261392593384\n",
- "42 Train 6.042512613184312 , Valid 7.4074866771698\n",
- "43 Train 6.032251245835248 , Valid 7.390785813331604\n",
- "44 Train 6.025217645308551 , Valid 7.370600938796997\n",
- "45 Train 6.027591985814712 , Valid 7.377846837043762\n",
- "46 Train 6.0253391826854035 , Valid 7.39339554309845\n",
- "47 Train 6.019261051626766 , Valid 7.407409071922302\n",
- "48 Train 6.014283713172464 , Valid 7.393349766731262\n",
- "49 Train 6.010964337517233 , Valid 7.37293815612793\n",
- "50 Train 6.010590104495778 , Valid 7.4031054973602295\n"
+ "1 Train 4.168464240410345 , Valid 6.253535368863274\n",
+ "2 Train 4.158141554688378 , Valid 6.238900885862463\n",
+ "3 Train 4.135342076527986 , Valid 6.225469000199261\n",
+ "4 Train 4.12202568705991 , Valid 6.212939725202673\n",
+ "5 Train 4.109015821552963 , Valid 6.205244933857637\n",
+ "6 Train 4.098594773587563 , Valid 6.194438050774967\n",
+ "7 Train 4.08812516884838 , Valid 6.178463767556583\n",
+ "8 Train 4.079949097667667 , Valid 6.175140450982487\n",
+ "9 Train 4.07018859094853 , Valid 6.167128044016221\n",
+ "10 Train 4.065771084037616 , Valid 6.161452812307021\n"
]
}
],
@@ -373,11 +374,15 @@
"# Train with data loader.\n",
"criterion = nn.CrossEntropyLoss()\n",
"optimizer = optim.Adam(model.parameters(), lr=0.001)\n",
+ "#optimizer = optim.SGD(model.parameters(), lr=0.0001, momentum=0.5)\n",
"\n",
- "epoch = 50\n",
- "batch_size = 64\n",
+ "epoch = 10\n",
+ "batch_size = 8\n",
+ "alpha = 0.5\n",
"train_dataloader = DataLoader(train_data, batch_size=batch_size, shuffle=True)\n",
"valid_dataloader = DataLoader(valid_data, batch_size=batch_size, shuffle=True)\n",
+ "\n",
+ "model.train()\n",
"for e in range(epoch):\n",
" train_loss_epoch = 0\n",
" valid_loss_epoch = 0\n",
@@ -386,10 +391,47 @@
" train = sample_batched_train[0]\n",
" train_mask = sample_batched_train[1]\n",
" label_index = sample_batched_train[2].to(device)\n",
+ " sub_objs = sample_batched_train[3]\n",
" \n",
" # Forward pass\n",
" output = model(train, train_mask)\n",
- " loss = criterion(output, label_index)\n",
+ " loss_gs = []\n",
+ " for j in range(len(sub_objs)):\n",
+ " if not (sub_objs[j].split(\" \")[0] == \"[SUB]\" or sub_objs[j].split(\" \")[0] == \"[OBJ]\"):\n",
+ " continue\n",
+ "\n",
+ " if sub_objs[j].split(\" \")[0] == \"[SUB]\":\n",
+ " sub = sub_objs[j].split(\" \")[1]\n",
+ " q = \"SELECT ?r WHERE { <\" + sub + \"> ?r ?o }\"\n",
+ "\n",
+ " if sub_objs[j].split(\" \")[0] == \"[OBJ]\":\n",
+ " obj = sub_objs[j].split(\" \")[1]\n",
+ " q = \"SELECT ?r WHERE { ?s ?r <\" + obj + \"> }\"\n",
+ " \n",
+ " params = {\n",
+ " \"default-graph-uri\": \"http://dbpedia.org\",\n",
+ " \"query\": q,\n",
+ " \"format\": \"json\"\n",
+ " }\n",
+ "\n",
+ " response = requests.get(SPARQL_ENDPOINT, headers=headers, params=params, timeout=15)\n",
+ " results = response.json()\n",
+ " res_rels = {}\n",
+ " for i in range(len(list(results[\"results\"].values())[2])):\n",
+ " res_rels[list(results[\"results\"].values())[2][i][\"r\"][\"value\"].lower()] = \"True\"\n",
+ " \n",
+ " loss_rels = []\n",
+ " for i in range(len(relations)):\n",
+ " if prefixes_reverse[\"\".join(relations[i].split(\":\")[0]) + \":\"] + relations[i].split(\":\")[1] in list(res_rels.keys()):\n",
+ " loss_rels.append(5) \n",
+ " else:\n",
+ " loss_rels.append(1/20)\n",
+ " loss_gs.append(loss_rels)\n",
+ " loss_gs = torch.FloatTensor(loss_gs).to(device)\n",
+ "\n",
+ " #print(response)\n",
+ " output_gs = output * (1-alpha) + (loss_gs) * alpha\n",
+ " loss = criterion(output_gs, label_index)\n",
"\n",
" # backward and optimize\n",
" loss.backward()\n",
@@ -413,7 +455,7 @@
},
{
"cell_type": "code",
- "execution_count": 31,
+ "execution_count": 35,
"metadata": {},
"outputs": [
{
@@ -427,7 +469,7 @@
"name": "stderr",
"output_type": "stream",
"text": [
- "100%|██████████| 2052/2052 [00:03<00:00, 577.85it/s]\n"
+ "100%|██████████| 2052/2052 [00:00<00:00, 2654.58it/s]\n"
]
},
{
@@ -442,7 +484,7 @@
"name": "stderr",
"output_type": "stream",
"text": [
- "100%|██████████| 1161/1161 [00:02<00:00, 566.88it/s]\n"
+ "100%|██████████| 1161/1161 [00:00<00:00, 2831.73it/s]\n"
]
},
{
@@ -450,22 +492,22 @@
"output_type": "stream",
"text": [
"Finished with batches\n",
- "test loss 6.031976699829102\n",
- "lowest confidence train 0.08024344\n",
- "lowest confidence test 0.08576707\n",
- "Accuracy train: 0.3939828080229226\n",
- "Accuracy test: 0.006361323155216285\n"
+ "test loss 6.039710998535156\n",
+ "lowest confidence train 0.07557418\n",
+ "lowest confidence test 0.08204544\n",
+ "Accuracy train: 0.2349570200573066\n",
+ "Accuracy test: 0.007633587786259542\n"
]
}
],
"source": [
"# Predict\n",
- "train, train_mask, corr_rels = make_batch(src=\"../LC-QuAD/combined-requeried-linked-train.json\", http_prefix = True)\n",
- "test, test_mask, corr_rels_test = make_batch(src=\"../LC-QuAD/combined-requeried-linked-test.json\", http_prefix = True)\n",
- "test_data = MyDataset(test, test_mask, corr_rels_test, relations=relations)\n",
+ "train, train_mask, corr_rels,ents = make_batch(src=\"../LC-QuAD/combined-requeried-linked-train.json\", http_prefix = True)\n",
+ "test, test_mask, corr_rels_test, ents_test = make_batch(src=\"../LC-QuAD/combined-requeried-linked-test.json\", http_prefix = True)\n",
+ "test_data = MyDataset(test, test_mask, corr_rels_test, ents=ents_test, relations=relations)\n",
"test_dataloader = DataLoader(test_data, batch_size=len(test_data), shuffle=True)\n",
"\n",
- "test_batch, test_mask_batch, corr_rels_test_batch = next(iter(test_dataloader))\n",
+ "test_batch, test_mask_batch, corr_rels_test_batch, sub_objs = next(iter(test_dataloader))\n",
"corr_rels_test_batch = corr_rels_test_batch.to(device)\n",
"with torch.no_grad():\n",
" output_train = model(train, train_mask)\n",
@@ -473,8 +515,44 @@
" loss = criterion(output_test, corr_rels_test_batch)\n",
" print(\"test loss\", loss.item())\n",
"\n",
+ " loss_gs = []\n",
+ " for j in range(len(sub_objs)):\n",
+ " if not (sub_objs[j].split(\" \")[0] == \"[SUB]\" or sub_objs[j].split(\" \")[0] == \"[OBJ]\"):\n",
+ " continue\n",
+ "\n",
+ " if sub_objs[j].split(\" \")[0] == \"[SUB]\":\n",
+ " sub = sub_objs[j].split(\" \")[1]\n",
+ " q = \"SELECT ?r WHERE { <\" + sub + \"> ?r ?o }\"\n",
+ "\n",
+ " if sub_objs[j].split(\" \")[0] == \"[OBJ]\":\n",
+ " obj = sub_objs[j].split(\" \")[1]\n",
+ " q = \"SELECT ?r WHERE { ?s ?r <\" + obj + \"> }\"\n",
+ "\n",
+ " params = {\n",
+ " \"default-graph-uri\": \"http://dbpedia.org\",\n",
+ " \"query\": q,\n",
+ " \"format\": \"json\"\n",
+ " }\n",
+ "\n",
+ " response = requests.get(\n",
+ " SPARQL_ENDPOINT, headers=headers, params=params, timeout=15)\n",
+ " results = response.json()\n",
+ " res_rels = {}\n",
+ " for i in range(len(list(results[\"results\"].values())[2])):\n",
+ " res_rels[list(results[\"results\"].values())[2][i]\n",
+ " [\"r\"][\"value\"].lower()] = \"True\"\n",
+ "\n",
+ " loss_rels = []\n",
+ " for i in range(len(relations)):\n",
+ " if prefixes_reverse[\"\".join(relations[i].split(\":\")[0]) + \":\"] + relations[i].split(\":\")[1] in list(res_rels.keys()):\n",
+ " loss_rels.append(5)\n",
+ " else:\n",
+ " loss_rels.append(1/20)\n",
+ " loss_gs.append(loss_rels)\n",
+ " loss_gs = torch.FloatTensor(loss_gs).to(device)\n",
+ "output_gs_test = output_test# * loss_gs\n",
"output_train = output_train.detach().cpu().numpy()\n",
- "output_test = output_test.detach().cpu().numpy()\n",
+ "output_test = output_gs_test.detach().cpu().numpy()\n",
"\n",
"\n",
"prediction_train = [relations[np.argmax(pred).item()]for pred in output_train]\n",
@@ -506,15 +584,55 @@
},
{
"cell_type": "code",
- "execution_count": null,
+ "execution_count": 190,
"metadata": {},
- "outputs": [],
- "source": []
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "+---------------+-------------------+\n",
+ "| Mod name | Parameters Listed |\n",
+ "+---------------+-------------------+\n",
+ "| linear.weight | 338688 |\n",
+ "| linear.bias | 441 |\n",
+ "+---------------+-------------------+\n",
+ "Sum of trained parameters: 339129\n"
+ ]
+ },
+ {
+ "data": {
+ "text/plain": [
+ "339129"
+ ]
+ },
+ "execution_count": 190,
+ "metadata": {},
+ "output_type": "execute_result"
+ }
+ ],
+ "source": [
+ "from prettytable import PrettyTable\n",
+ "def count_parameters(model):\n",
+ " table = PrettyTable([\"Mod name\", \"Parameters Listed\"])\n",
+ " t_params = 0\n",
+ " for name, parameter in model.named_parameters():\n",
+ " if not parameter.requires_grad:\n",
+ " continue\n",
+ " param = parameter.numel()\n",
+ " table.add_row([name, param])\n",
+ " t_params += param\n",
+ " print(table)\n",
+ " print(f\"Sum of trained parameters: {t_params}\")\n",
+ " return t_params\n",
+ "\n",
+ "count_parameters(model)"
+ ]
}
],
"metadata": {
"kernelspec": {
- "display_name": "Python 3.10.4 ('tdde19')",
+ "display_name": "Python 3.9.11 64-bit",
"language": "python",
"name": "python3"
},
@@ -528,12 +646,12 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
- "version": "3.10.4"
+ "version": "3.9.11"
},
"orig_nbformat": 4,
"vscode": {
"interpreter": {
- "hash": "8e4aa0e1a1e15de86146661edda0b2884b54582522f7ff2b916774ba6b8accb1"
+ "hash": "64e7cd3b4b88defe39dd61a4584920400d6beb2615ab2244e340c2e20eecdfe9"
}
}
},
diff --git a/bart/sparql.ipynb b/bart/sparql.ipynb
index adaf70cfc3a9b55cb2a1499a3d81c2e673ec1f0c..6dcb66eb8825c68573affc46257035764f82681b 100644
--- a/bart/sparql.ipynb
+++ b/bart/sparql.ipynb
@@ -4075,15 +4075,13 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
- "version": "3.9.10"
+ "version": "3.9.11"
},
"vscode": {
- "interpreter": {
"hash": "0988fb18a177ab47253348976d615106e52998fa9d41728b0d76cf08f2eafea6"
}
},
"widgets": {
- "application/vnd.jupyter.widget-state+json": {
"04562bf2742444049a073935c78ad553": {
"model_module": "@jupyter-widgets/base",
"model_module_version": "1.2.0",
diff --git a/data/relations-all-no-http-lowercase.json b/data/relations-all-no-http-lowercase.json
new file mode 100644
index 0000000000000000000000000000000000000000..90d64557b4730b9cf007741d100ac3a8b89f775b
--- /dev/null
+++ b/data/relations-all-no-http-lowercase.json
@@ -0,0 +1,443 @@
+[
+ "dbp:firstdriver",
+ "dbp:governingbody",
+ "dbp:mission",
+ "dbp:flagbearer",
+ "dbp:manager",
+ "dbo:foundationplace",
+ "dbo:related",
+ "dbo:lieutenant",
+ "dbo:formerpartner",
+ "dbp:combatant",
+ "dbo:spouse",
+ "dbo:division",
+ "dbp:religion",
+ "dbp:affiliations",
+ "dbo:predecessor",
+ "dbp:knownfor",
+ "dbp:cinematography",
+ "dbo:ideology",
+ "dbp:meaning",
+ "dbp:constituency",
+ "dbo:maintainedby",
+ "dbo:layout",
+ "dbp:address",
+ "dbo:computingplatform",
+ "dbp:position",
+ "dbo:party",
+ "dbp:religiousaffiliation",
+ "dbp:genre",
+ "dbp:successor",
+ "dbp:nationality",
+ "dbo:parentorganisation",
+ "dbp:assembly",
+ "dbp:subject",
+ "dbp:headquarters",
+ "dbp:placeofbirth",
+ "dbp:team",
+ "dbp:director",
+ "dbo:family",
+ "dbp:district",
+ "dbp:creators",
+ "dbo:firstdriver",
+ "dbo:architecturalstyle",
+ "dbo:mountainrange",
+ "dbp:leadername",
+ "dbo:product",
+ "dbp:origin",
+ "dbp:doctoralstudents",
+ "dbo:trainer",
+ "dbo:order",
+ "dbp:birthplace",
+ "dbo:builder",
+ "dbo:voice",
+ "dbp:majorshrine",
+ "dbp:office",
+ "dbp:leader",
+ "dbp:locationcountry",
+ "dbo:manager",
+ "dbp:draftteam",
+ "dbp:license",
+ "dbp:editor",
+ "dbp:trainer",
+ "dbp:stadium",
+ "dbp:domain",
+ "dbo:gender",
+ "dbo:highschool",
+ "dbo:associatedband",
+ "dbo:coach",
+ "dbp:broadcastarea",
+ "dbo:automobileplatform",
+ "dbo:lyrics",
+ "dbo:training",
+ "dbo:hubairport",
+ "dbo:militaryunit",
+ "dbp:membership",
+ "dbo:firstascentperson",
+ "dbo:operator",
+ "dbo:notablework",
+ "dbp:teamname",
+ "dbp:creator",
+ "dbp:architecturalstyle",
+ "dbp:owners",
+ "dbo:language",
+ "dbp:state",
+ "dbo:mouthmountain",
+ "dbp:archipelago",
+ "dbp:publisher",
+ "dbp:format",
+ "dbp:products",
+ "dbp:predecessor",
+ "dbo:season",
+ "dbo:bronzemedalist",
+ "dbp:discipline",
+ "dbp:lyrics",
+ "dbo:bandmember",
+ "dbo:knownfor",
+ "dbo:silvermedalist",
+ "dbp:architecture",
+ "dbo:stateoforigin",
+ "dbo:network",
+ "dbo:cinematography",
+ "dbp:coach",
+ "dbo:nearestcity",
+ "dbp:author",
+ "dbp:debutteam",
+ "dbp:prizes",
+ "dbo:mayor",
+ "dbo:denomination",
+ "dbo:currency",
+ "dbp:workinstitutions",
+ "dbp:spouse",
+ "dbo:ethnicity",
+ "dbp:notablecommanders",
+ "dbp:starring",
+ "dbp:battles",
+ "dbo:formerteam",
+ "dbo:foundedby",
+ "dbo:leader",
+ "dbo:veneratedin",
+ "dbo:restingplace",
+ "dbp:licensee",
+ "dbp:appointer",
+ "dbp:keypeople",
+ "dbo:relative",
+ "dbp:training",
+ "dbp:presenter",
+ "dbo:managerclub",
+ "dbo:writer",
+ "dbo:partner",
+ "dbo:operatingsystem",
+ "dbp:country",
+ "dbp:music",
+ "dbo:basedon",
+ "dbp:deathdate",
+ "dbp:line",
+ "dbo:creator",
+ "dbo:series",
+ "dbo:previouswork",
+ "dbp:founded",
+ "dbp:jurisdiction",
+ "dbo:opponent",
+ "dbp:area",
+ "dbo:garrison",
+ "dbp:gender",
+ "dbo:stadium",
+ "dbp:place",
+ "dbp:chancellor",
+ "dbp:hostcity",
+ "dbp:placeofburial",
+ "dbp:cityserved",
+ "dbo:starring",
+ "dbo:profession",
+ "dbp:agencyname",
+ "dbo:federalstate",
+ "dbo:origin",
+ "dbp:party",
+ "dbp:tenants",
+ "dbp:schooltype",
+ "dbp:magazine",
+ "dbo:distributinglabel",
+ "dbp:purpose",
+ "dbo:incumbent",
+ "dbp:homestadium",
+ "dbp:canonizedby",
+ "dbo:deathplace",
+ "dbp:design",
+ "dbo:producer",
+ "dbo:officiallanguage",
+ "dbp:foundation",
+ "dbp:champion",
+ "dbo:author",
+ "dbp:hometown",
+ "dbp:doctoraladvisor",
+ "dbp:affiliation",
+ "dbp:children",
+ "dbp:occupation",
+ "dbo:affiliation",
+ "dbo:doctoralstudent",
+ "dbo:authority",
+ "dbp:employer",
+ "dbp:partner",
+ "dbo:architect",
+ "dbo:deathcause",
+ "dbo:jurisdiction",
+ "dbo:battle",
+ "dbp:manufacturer",
+ "dbp:relatives",
+ "dbp:deathplace",
+ "dbp:pastmembers",
+ "dbp:operator",
+ "dbo:honours",
+ "dbp:owner",
+ "dbo:headquarter",
+ "dbo:ceremonialcounty",
+ "dbp:firstteam",
+ "dbp:college",
+ "dbp:label",
+ "dbp:poledriver",
+ "dbo:president",
+ "dbo:destination",
+ "dbp:screenplay",
+ "dbp:material",
+ "dbp:programminglanguage",
+ "dbp:sisterstations",
+ "dbo:founder",
+ "dbo:targetairport",
+ "dbp:officialname",
+ "dbp:hubs",
+ "dbo:academicadvisor",
+ "dbo:locationcity",
+ "dbp:parent",
+ "dbp:currentclub",
+ "dbp:services",
+ "dbo:discoverer",
+ "dbo:colour",
+ "dbo:team",
+ "dbo:largestcity",
+ "dbo:recordedin",
+ "dbp:governor",
+ "dbp:university",
+ "dbo:servingrailwayline",
+ "dbo:commander",
+ "dbo:species",
+ "dbo:poledriver",
+ "dbp:athletics",
+ "dbo:race",
+ "dbp:league",
+ "dbp:languages",
+ "dbo:militarybranch",
+ "dbp:leadertitle",
+ "dbp:notableworks",
+ "dbo:literarygenre",
+ "dbo:genre",
+ "dbp:placeofdeath",
+ "dbp:president",
+ "dbo:manufacturer",
+ "dbo:significantbuilding",
+ "dbp:mascot",
+ "dbp:borough",
+ "dbp:race",
+ "dbp:language",
+ "dbp:birthname",
+ "dbp:commander",
+ "dbo:distributor",
+ "dbp:beatifiedby",
+ "dbp:managerclubs",
+ "dbp:firstaired",
+ "dbo:countyseat",
+ "dbp:artist",
+ "dbo:majorshrine",
+ "dbo:athletics",
+ "dbo:education",
+ "dbo:programmeformat",
+ "dbp:branch",
+ "dbo:location",
+ "dbo:cpu",
+ "dbp:characters",
+ "dbp:inflow",
+ "dbp:outflow",
+ "dbp:houses",
+ "dbp:mainingredient",
+ "dbp:restingplace",
+ "dbo:wineregion",
+ "dbp:producer",
+ "dbp:primeminister",
+ "dbp:related",
+ "dbp:recorded",
+ "dbp:guests",
+ "dbo:director",
+ "dbp:style",
+ "dbo:successor",
+ "dbo:parentcompany",
+ "dbo:university",
+ "dbo:editor",
+ "dbo:timezone",
+ "dbo:hometown",
+ "dbo:commandstructure",
+ "dbo:monarch",
+ "dbo:publisher",
+ "dbo:residence",
+ "dbo:rivermouth",
+ "dbo:otherparty",
+ "dbo:designer",
+ "dbp:titles",
+ "dbo:nationality",
+ "dbo:stylisticorigin",
+ "dbo:movement",
+ "dbp:club",
+ "dbp:neighboringmunicipalities",
+ "dbp:locationtown",
+ "dbp:venue",
+ "dbp:destinations",
+ "dbo:board",
+ "dbp:residence",
+ "dbp:pastteams",
+ "dbo:editing",
+ "dbo:nonfictionsubject",
+ "dbp:lieutenant",
+ "dbp:order",
+ "dbp:title",
+ "dbo:routeend",
+ "dbo:launchsite",
+ "dbo:doctoraladvisor",
+ "dbp:writer",
+ "dbo:owningcompany",
+ "dbo:anthem",
+ "dbo:placeofburial",
+ "dbp:province",
+ "dbo:executiveproducer",
+ "dbp:distributor",
+ "dbo:debutteam",
+ "dbo:developer",
+ "dbo:chairman",
+ "dbp:ground",
+ "dbp:buildingtype",
+ "dbo:programminglanguage",
+ "dbp:nearestcity",
+ "dbp:international",
+ "dbp:almamater",
+ "dbo:academicdiscipline",
+ "dbp:deputy",
+ "dbo:river",
+ "dbp:garrison",
+ "dbo:associatedmusicalartist",
+ "dbp:school",
+ "dbo:breeder",
+ "dbp:chairman",
+ "dbo:sport",
+ "dbo:club",
+ "dbo:illustrator",
+ "dbp:founder",
+ "dbo:school",
+ "dbp:citizenship",
+ "dbp:designer",
+ "dbp:playedfor",
+ "dbp:arena",
+ "dbo:parent",
+ "dbo:ingredient",
+ "dbp:junction",
+ "dbo:institution",
+ "dbp:engine",
+ "dbo:regionserved",
+ "dbp:rank",
+ "dbp:operatingsystem",
+ "dbo:type",
+ "dbp:carries",
+ "dbo:citizenship",
+ "dbp:locationcity",
+ "dbp:nickname",
+ "dbo:governmenttype",
+ "dbp:commandstructure",
+ "dbo:militaryrank",
+ "dbo:formerbandmember",
+ "dbp:maininterests",
+ "dbo:almamater",
+ "dbo:country",
+ "dbp:narrated",
+ "dbo:occupation",
+ "dbo:primeminister",
+ "dbo:field",
+ "dbo:composer",
+ "dbo:routestart",
+ "dbo:league",
+ "dbo:outflow",
+ "dbo:subsidiary",
+ "dbo:child",
+ "dbp:highschool",
+ "dbp:sisternames",
+ "dbo:portrayer",
+ "dbo:award",
+ "dbp:crosses",
+ "dbo:employer",
+ "dbp:name",
+ "dbp:city",
+ "dbo:birthplace",
+ "dbo:musicby",
+ "dbp:editing",
+ "dbo:license",
+ "dbo:relation",
+ "dbp:fields",
+ "dbp:nationalorigin",
+ "dbp:location",
+ "dbo:kingdom",
+ "dbp:largestcity",
+ "dbo:service",
+ "dbo:capital",
+ "dbp:field",
+ "dbp:currency",
+ "dbo:phylum",
+ "dbp:haircolor",
+ "dbp:deathcause",
+ "dbo:ground",
+ "dbp:education",
+ "dbo:broadcastarea",
+ "dbo:territory",
+ "dbp:mother",
+ "dbo:narrator",
+ "dbp:allegiance",
+ "dbo:assembly",
+ "dbo:college",
+ "dbp:region",
+ "dbo:region",
+ "dbo:album",
+ "dbo:county",
+ "dbo:tenant",
+ "dbp:executiveproducer",
+ "dbp:awards",
+ "dbp:developer",
+ "dbp:writers",
+ "dbp:architect",
+ "dbp:thememusiccomposer",
+ "dbo:presenter",
+ "dbp:company",
+ "dbo:owner",
+ "dbp:birthdate",
+ "dbp:nationalteam",
+ "dbo:keyperson",
+ "dbp:cities",
+ "dbo:campus",
+ "dbo:city",
+ "dbo:languagefamily",
+ "dbo:sisterstation",
+ "dbp:coverartist",
+ "dbp:veneratedin",
+ "dbp:animator",
+ "dbp:type",
+ "dbp:notableinstruments",
+ "dbo:locatedinarea",
+ "dbo:religion",
+ "dbo:binomialauthority",
+ "dbo:neighboringmunicipality",
+ "dbp:role",
+ "dbo:homestadium",
+ "dbp:os",
+ "dbp:youthclubs",
+ "dbp:currentmembers",
+ "dbp:album",
+ "dbo:instrument",
+ "dbo:subsequentwork",
+ "dbo:inflow",
+ "dbo:artist",
+ "dbp:associatedacts"
+]