SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2304.07063 · ICLR · 2024

Rethinking Complex Queries on Knowledge Graphs with Neural Link Predictors

Zihao Wang, Hang Yin, Yangqiu Song

arXiv · PDF · Open in the Atlas

Code that ran

20 of 41 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 6 have no contract verdict; 12 failures of our instrument; 21 no recorded run); 1 of them was also run on the host, not in the jail, by the peer-agreement lane that set its level, measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2304.07063.json · badge: /api/ran/2304.07063.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
hkust-knowcomp/fit canonical (classifier label)
classifier basis: classifier_2025
14 ran · 20 no recorded run
HKUST-KnowComp/FIT canonical (classifier label)
classifier basis: census
6 ran · 1 no recorded run
FunctionStatusWhere it lives
read_from_yaml Ran
ran; our contract draft was wrong, not the code
hkust-knowcomp/fit/QG_FIT.py
code served (permissive licence) · get_code("84eb10126b44cd7c")
How it ran

Ran in Syntology's jail on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned dict. The image it ran in was not recorded. The comparison behind this sample's level is a different run: the peer-agreement lane ran it again on the host, not in the jail (report dated 2026-09-05T00:43:41Z). If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import yaml
import os

# Create a temporary YAML file with test content
temp_dir = tempfile.mkdtemp()
yaml_path = os.path.join(temp_dir, "test_config.yaml")

test_data = {
    "model": "test_model",
    "learning_rate": 0.001,
    "batch_size": 32,
    "layers": [64, 128, 256],
    "nested": {
        "key1": "value1",
        "key2": 42
    }
}

with open(yaml_path, 'w') as f:
    yaml.dump(test_data, f)

ARGS = {
    "yaml_path": yaml_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
yaml — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Config Ran
ran on a synthesized input
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("2912753768ab645c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# INIT_ARGS: arguments for Config.__init__
INIT_ARGS = {
    'config_dict': {
        'learning_rate': 0.001,
        'batch_size': 2,
        'hidden_dim': 4,
        'num_layers': 2,
        'dropout': 0.1,
        'extra_param': 'test_value'
    }
}

# Config doesn't have a computation method like forward or __call__
# It's just a configuration container, so we test to_dict()
METHOD = 'to_dict'

# ARGS: arguments for the to_dict method (it takes no arguments)
ARGS = {}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
KGIndex Ran
ran on a synthesized input
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("9e89fde226cf805b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned NoneType. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json
import tempfile
import os

INIT_ARGS = {}

ARGS = {
    'name': 'entity_1',
    'eid': None
}

METHOD = 'register_entity'

# Setup code to construct and call the class
if __name__ == '__main__':
    from __main__ import KGIndex
    
    # Construct the instance
    kg_index = KGIndex(**INIT_ARGS)
    
    # Register some entities
    kg_index.register_entity('entity_1')
    kg_index.register_entity('entity_2', eid=5)
    kg_index.register_entity('entity_3')
    
    # Register some relations
    kg_index.register_relation('relation_1')
    kg_index.register_relation('relation_2', rid=10)
    
    # Test the main computation method
    kg_index.register_entity(**ARGS)
    
    # Test properties
    print(f"Number of entities: {kg_index.num_entities}")
    print(f"Number of relations: {kg_index.num_relations}")
    
    # Test dump and load
    with tempfile.TemporaryDirectory() as tmpdir:
        filepath = os.path.join(tmpdir, 'kg_index.json')
        kg_index.dump(filepath)
        loaded_kg = KGIndex.load(filepath)
        print(f"Loaded entities: {loaded_kg.map_entity_name_to_id}")
        print(f"Loaded relations: {loaded_kg.map_relation_name_to_id}")
    
    print("KGIndex test completed successfully")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
__main__ — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type NoneType — a NoneType output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a NoneType output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Kge_finetune Ran
ran on a synthesized input
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("29357053d3c8e6cf")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch import nn
from math import ceil

class MockObservedKG:
    def __init__(self, n_entity, n_rel):
        self.hr2t = {}
        for h in range(n_entity):
            for r in range(n_rel):
                self.hr2t[(h, r)] = set([h % 3, (h + 1) % 3])

def score_function_mock(rel_emb, head_emb, tail_emb, rank):
    batch_size = head_emb.shape[0]
    n_tail = tail_emb.shape[1]
    scores = torch.randn(batch_size, 1, n_tail)
    return scores

n_entity = 8
ent_dim = 6
rel_dim = 8
n_rel = 4
device = 'cpu'

ent_emb_tensor = torch.randn(n_entity, ent_dim)
rel_emb_tensor = torch.randn(n_rel, rel_dim)
relation_matrix_list = [torch.randn(n_entity, n_entity) for _ in range(n_rel)]
observed_kg = MockObservedKG(n_entity, n_rel)

INIT_ARGS = {
    'observed_kg': observed_kg,
    'ent_emb': ent_emb_tensor,
    'rel_emb': rel_emb_tensor,
    'relation_matrix_list': relation_matrix_list,
    'freeze_ent': False,
    'score_function': score_function_mock,
    'head_batch': 2,
    'sparse': True,
    'threshold': 0.1,
    'epsilon': 0.01,
    'device': device,
}

ARGS = {
    'rel_id': 1,
    'head_ent_vec': torch.tensor([1.0, 0.0, 1.0, 0.0, 0.0, 0.0, 1.0, 0.0]),
    'train_mask': False,
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
KnowledgeGraphConfig Ran
ran on a synthesized input
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("e1a678914e17d9cc")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

INIT_ARGS = {
    'config_dict': {
        'filelist': ['dummy_train.tsv', 'dummy_valid.tsv'],
        'kgindex': 'test_index',
        'tensorize': False
    }
}

ARGS = {}

METHOD = 'to_dict'
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
RaggedBatch Ran
ran on a synthesized input
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("2320379032a21fe0")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 5, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch.nn.utils.rnn import pad_sequence

class RaggedBatch:

    def __init__(self, flatten, sizes):
        self.flatten = flatten
        self.sizes = sizes

    def run_ops_on_flatten(self, opfunc):
        return RaggedBatch(flatten=opfunc(self.flatten), sizes=self.sizes)

    def to_dense_matrix(self, padding_value):
        flatten_sliced = torch.split(self.flatten, split_size_or_sections=self.sizes, dim=0)
        dense_matrix = pad_sequence(flatten_sliced, batch_first=True, padding_value=padding_value)
        return dense_matrix


# Setup: create a ragged batch with 2 sequences of different lengths
# Sequence 1: 3 elements, Sequence 2: 5 elements, each element is a 4-dim vector
flatten_data = torch.randn(8, 4)  # 3+5=8 total elements, 4 features each
sizes = [3, 5]

INIT_ARGS = {
    'flatten': flatten_data,
    'sizes': sizes
}

ARGS = {
    'padding_value': 0.0
}

METHOD = 'to_dense_matrix'

# Construct and call
rb = RaggedBatch(**INIT_ARGS)
result = rb.to_dense_matrix(**ARGS)
print(result.shape)
print(result)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 5, 4) — shape (2, 5, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
_iter_triple_from_tsv Ran
ran; our contract draft was wrong, not the code
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("c7e8b390d25b5532")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned generator. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import os

# Create a temporary TSV file with test data
temp_dir = tempfile.mkdtemp()
triple_file = os.path.join(temp_dir, "triples.tsv")

# Write test data to the file
with open(triple_file, 'w') as f:
    f.write("1 2 3\n")
    f.write("4 5 6\n")
    f.write("7 8 9\n")

ARGS = {
    "triple_file": triple_file,
    "to_int": True,
    "check_size": 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type generator — a generator output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a generator output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
align_entities_relations Ran
ran; our contract draft was wrong, not the code
HKUST-KnowComp/FIT/convert_beta_dataset.py
code served (permissive licence) · get_code("2c5d6c341e63248e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Analyzing the function:
# - labeled_beta_type: nested structure that can contain strings or other iterables
# - beta_sample: parallel nested structure with corresponding samples
# - The function recursively zips these together
# - It checks if sub_type is a string, and if so, checks if first char is 's' or 'r'
# - It stores matching sub_type -> sub_sample pairs in dictionary d

# Create a nested structure for labeled_beta_type
# Mix of strings (some starting with 's' or 'r', some not) and nested lists
labeled_beta_type = [
    "subject_type",
    "relation_type",
    ["nested_s_entity", "other_field"],
    "random_string"
]

# Create corresponding nested structure for beta_sample
# Must have same nesting structure as labeled_beta_type
beta_sample = [
    torch.randn(2, 4),
    torch.randn(2, 4),
    [torch.randn(2, 4), torch.randn(2, 4)],
    torch.randn(2, 4)
]

ARGS = {
    "labeled_beta_type": labeled_beta_type,
    "beta_sample": beta_sample
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_candidate_set Ran
ran; our contract draft was wrong, not the code
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("c94caad91983a47e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Analyze the function:
# - Takes candidate_set as parameter
# - Iterates over keys in candidate_set (nodes)
# - Accesses candidate_set[node] and checks len() on it
# - candidate_set is a dict where values are sequences/collections with len()

# Create a valid candidate_set dict
# The function checks if any node has an empty candidate list (returns False)
# or if exactly one node has a single candidate (returns that node)

candidate_set = {
    "node_1": [1, 2, 3],      # Multiple candidates
    "node_2": [5],             # Single candidate - this should be fixed_node
    "node_3": [7, 8],          # Multiple candidates
}

ARGS = {
    "candidate_set": candidate_set
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
collect_matrix_scipy_sparse Ran
ran on a synthesized input
HKUST-KnowComp/FIT/create_matrix.py
code served (permissive licence) · get_code("330130ddfb49fc13")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import tempfile
import os

torch.manual_seed(0)

# Create a temporary directory for the matrix file
temp_dir = tempfile.mkdtemp()
matrix_path = os.path.join(temp_dir, "matrix.pt")

# Create a small dense matrix tensor with shape (2, 4, 3)
# This will be iterated over in the function, so first dimension should be > 1
dense_matrix = torch.randn(2, 4, 3)

# Save the matrix to the temporary path
torch.save(dense_matrix, matrix_path)

ARGS = {
    "matrix_path": matrix_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 3, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_batch_score_complex Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("dfaa9a08584100be")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family float, shape (2, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Analyze the function:
# - rel: shape (batch, ?, 2*rank) - split into real/img at dimension 2
# - arg1: shape (batch, ?, 2*rank) - split into real/img at dimension 2
# - arg2: shape (batch, ?, 2*rank) - split into real/img at dimension 2
# - rank: int - used to split the last dimension
# - Returns: tensor of shape (batch, ?)

batch_size = 2
num_entities = 3
rank = 2

# Create tensors with shape (batch, num_entities, 2*rank)
# This allows the function to split at rank: [:, :, :rank] and [:, :, rank:]
rel = torch.randn(batch_size, num_entities, 2 * rank)
arg1 = torch.randn(batch_size, num_entities, 2 * rank)
arg2 = torch.randn(batch_size, num_entities, 2 * rank)

ARGS = {
    "rel": rel,
    "arg1": arg1,
    "arg2": arg2,
    "rank": rank,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 3) — shape (2, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_batch_score_distmult Ran
ran; our contract draft was wrong, not the code
HKUST-KnowComp/FIT/compute_score.py
code served (permissive licence) · get_code("1869521a4be75145")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function compute_batch_score_distmult expects:
# - rel: tensor of shape (batch_size, embedding_dim)
# - h_emb: tensor of shape (batch_size, embedding_dim)
# - t_emb: tensor of shape (batch_size, embedding_dim)
# It computes element-wise multiplication and sums along the last dimension

batch_size = 2
embedding_dim = 4

rel = torch.randn(batch_size, embedding_dim)
h_emb = torch.randn(batch_size, embedding_dim)
t_emb = torch.randn(batch_size, embedding_dim)

ARGS = {
    "rel": rel,
    "h_emb": h_emb,
    "t_emb": t_emb,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_batch_score_transe Ran
ran; our contract draft was wrong, not the code
HKUST-KnowComp/FIT/compute_score.py
code served (permissive licence) · get_code("8c0698d90001cd35")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function expects:
# - rel: tensor of shape (batch_size, embedding_dim)
# - h_emb: tensor of shape (batch_size, embedding_dim)
# - t_emb: tensor of shape (batch_size, embedding_dim)
# All tensors are used in arithmetic operations and norm computation

batch_size = 2
embedding_dim = 4

rel = torch.randn(batch_size, embedding_dim)
h_emb = torch.randn(batch_size, embedding_dim)
t_emb = torch.randn(batch_size, embedding_dim)

ARGS = {
    'rel': rel,
    'h_emb': h_emb,
    't_emb': t_emb,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
fixed_depth_nested_dict Ran
ran; our contract draft was wrong, not the code
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("7bd57dd5eaa81647")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned defaultdict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import collections
from functools import partial
from itertools import repeat

def fixed_depth_nested_dict(default_factory, depth=1):
    result = partial(collections.defaultdict, default_factory)
    for _ in repeat(None, depth - 1):
        result = partial(collections.defaultdict, result)
    return result()

# Test with different depths and default factories
ARGS = {
    "default_factory": list,
    "depth": 2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type defaultdict — a defaultdict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a defaultdict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
iter_triple_from_tsv Ran
ran; our contract draft was wrong, not the code
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("b1fab42528218ff3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned generator. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import os

# Create temporary TSV files for testing
temp_dir = tempfile.mkdtemp()

# Create a single TSV file with triples
single_file = os.path.join(temp_dir, "triples1.tsv")
with open(single_file, 'w') as f:
    f.write("1 2 3\n")
    f.write("4 5 6\n")
    f.write("7 8 9\n")

# Create another TSV file for list testing
second_file = os.path.join(temp_dir, "triples2.tsv")
with open(second_file, 'w') as f:
    f.write("10 11 12\n")
    f.write("13 14 15\n")

ARGS = {
    "triple_files": single_file,
    "to_int": True,
    "check_size": 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type generator — a generator output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a generator output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_beta_model Ran
ran, honoured the contract we drafted
hkust-knowcomp/fit/QG_FIT.py
code served (permissive licence) · get_code("25412604ebe7d821")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import tempfile
import os

torch.manual_seed(0)

# Create a minimal model
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create model and optimizer instances
model = SimpleModel()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

# Create a temporary directory and checkpoint file
temp_dir = tempfile.mkdtemp()
checkpoint_data = {
    'step': 100,
    'model_state_dict': model.state_dict(),
    'current_learning_rate': 0.001,
    'warm_up_steps': 1000,
    'optimizer_state_dict': optimizer.state_dict(),
}
torch.save(checkpoint_data, os.path.join(temp_dir, 'checkpoint'))

ARGS = {
    'checkpoint_path': temp_dir,
    'model': model,
    'optimizer': optimizer,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_model Ran
ran, honoured the contract we drafted
hkust-knowcomp/fit/QG_FIT.py
code served (permissive licence) · get_code("4c08e440ba7f3e8c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import tempfile
import os

torch.manual_seed(0)

# Create a minimal model
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create a minimal optimizer
model = SimpleModel()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

# Create a temporary checkpoint directory and file
temp_dir = tempfile.mkdtemp()
checkpoint_path = temp_dir

# Create a valid checkpoint file
step = 100
checkpoint_file = os.path.join(checkpoint_path, f'{step}.ckpt')

checkpoint_data = {
    'model_parameter': model.state_dict(),
    'optimizer_parameter': optimizer.state_dict(),
    'learning_rate': 0.001,
    'warm_up_steps': 1000
}

torch.save(checkpoint_data, checkpoint_file)

# Define arguments
ARGS = {
    'step': step,
    'checkpoint_path': checkpoint_path,
    'model': model,
    'opt': optimizer,
    'load_device': 'cpu'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
log_add_metric Ran
ran; our contract draft was wrong, not the code
HKUST-KnowComp/FIT/evaluate_lmpnn.py
code served (permissive licence) · get_code("8e7c46d561388d36")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Construct the arguments for log_add_metric
add_log = {
    'MRR': 0.0,
    'HITS1': 0,
    'HITS3': 0,
    'HITS10': 0,
    'num_queries': 0,
    'couple_MRR': 0.0,
    'HITS1*1': 0,
    'HITS3*3': 0,
    'HITS10*10': 0
}

mrr = 0.85
h1 = 1
h3 = 1
h10 = 1
mul_mrr = 0.85
h1_1 = 1
h3_3 = 3
h10_10 = 10

ARGS = {
    'add_log': add_log,
    'mrr': mrr,
    'h1': h1,
    'h3': h3,
    'h10': h10,
    'mul_mrr': mul_mrr,
    'h1_1': h1_1,
    'h3_3': h3_3,
    'h10_10': h10_10
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ranking2metrics Ran
ran; our contract draft was wrong, not the code
HKUST-KnowComp/FIT/evaluate_lmpnn.py
code served (permissive licence) · get_code("e9bc5bf12edea40c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function expects:
# - ranking: a tensor that can be indexed with lists
# - easy_ans: a set or list of indices
# - hard_ans: a set or list of indices
# - ranking_device: a torch device

# Create a small ranking tensor with scores for different answers
# We need enough elements to index with easy_ans and hard_ans
ranking = torch.randn(10)

# Create disjoint sets of easy and hard answers
easy_ans = {0, 1}
hard_ans = {2, 3, 4}

# Device is CPU
ranking_device = torch.device('cpu')

ARGS = {
    'ranking': ranking,
    'easy_ans': easy_ans,
    'hard_ans': hard_ans,
    'ranking_device': ranking_device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tensorize_batch_entities Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("b048d0b5cfcddddc")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family int, shape (4, 1) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Test case 1: List[int] - batch size = 1
entities_list_int = [1, 2, 3, 4]

# Test case 2: List[List[int]] - multiple samples
entities_list_of_lists = [[1, 2, 3], [4, 5, 6]]

# Test case 3: torch.Tensor with shape [batch_size, num_entities]
entities_tensor = torch.tensor([[10, 11, 12, 13], [14, 15, 16, 17]], dtype=torch.long)

device = torch.device('cpu')

# Use the first test case for ARGS
ARGS = {
    'entities': entities_list_int,
    'device': device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family int, shape (4, 1) — shape (4, 1) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
FIT_finetune Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("106410206f55ee55")
How it ran

no invocation recorded.

KnowledgeGraph Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("c7e631917f0831d7")
How it ran

no invocation recorded.

construct_matrix_dynamic Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("58bd1c00d37189f3")
How it ran

no invocation recorded.

construct_matrix_list Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("e1824a035caa07b0")
How it ran

no invocation recorded.

csp_efo1 Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("de00fcf9dff8f944")
How it ran

no invocation recorded.

cut_node_sub_problem Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("b9fe352ddb40ce7c")
How it ran

no invocation recorded.

evaluate_batch_joint Not yet run
unverified (harvested, no recorded run)
HKUST-KnowComp/FIT/evaluate_lmpnn.py
code served (permissive licence) · get_code("34cfe3992789f9f7")
How it ran

no invocation recorded.

extend_ans Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("2d594b302a995572")
How it ran

no invocation recorded.

find_enumerate_node Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("53b482ea547b3034")
How it ran

no invocation recorded.

find_leaf_node Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("c1c6d4c74680abee")
How it ran

no invocation recorded.

kg_remove_node Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("d4fa4a2e520e91da")
How it ran

no invocation recorded.

neural_cut_node_sub_problem Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("9224a778f5fbd47f")
How it ran

no invocation recorded.

neural_existential_update Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("003464ef621fcac7")
How it ran

no invocation recorded.

neural_extend_ans Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("df6ae0aeda5453e6")
How it ran

no invocation recorded.

neural_solve_EFO1_X Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("2868d57816b54ef1")
How it ran

no invocation recorded.

neural_solve_conjunctive Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("d9353c5d35da90dd")
How it ran

no invocation recorded.

node_filter Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("9c109521a0a3bf1c")
How it ran

no invocation recorded.

node_pair_filtering Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("4968b4d922dd7429")
How it ran

no invocation recorded.

solve_EFO1_new Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("2a1fdf0490126a9e")
How it ran

no invocation recorded.

solve_conjunctive Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("f408248313c609f9")
How it ran

no invocation recorded.

topology_filter Not yet run
unverified (harvested, no recorded run)
hkust-knowcomp/fit/Neural_FIT.py
code served (permissive licence) · get_code("95aa5762a066cf5b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 41 samples lifted from this paper's linked repositories, 41 carry a function identity and 0 do not. Of the 41 distinct functions with an identity, 10 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 10 later) and 31 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
align_entities_relations 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
compute_batch_score_complex
brick of this paper's assembly: helper
1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
compute_batch_score_distmult 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
compute_batch_score_transe 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
evaluate_batch_joint 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
load_beta_model 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
load_model 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
log_add_metric 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
ranking2metrics 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation
read_from_yaml 1 other paper
1 later
2307.13701 (2023-07)
$\text{EFO}_{k}$-CQA: Towards Knowledge Graph Complex Query Answering beyond Set Operation

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Reasoning on knowledge graphs is a challenging task because it utilizes observed information to predict the missing one. Particularly, answering complex queries based on first-order logic is one of the crucial tasks to verify learning to reason abilities for generalization and composition. Recently, the prevailing method is query embedding which learns the embedding of a set of entities and treats logic operations as set operations and has shown great empirical success. Though there has been much research following the same formulation, many of its claims lack a formal and systematic inspection. In this paper, we rethink this formulation and justify many of the previous claims by characterizing the scope of queries investigated previously and precisely identifying the gap between its formulation and its goal, as well as providing complexity analysis for the currently investigated queries. Moreover, we develop a new dataset containing ten new types of queries with features that have never been considered and therefore can provide a thorough investigation of complex queries. Finally, we propose a new neural-symbolic method, Fuzzy Inference with Truth value (FIT), where we equip the neural link predictors with fuzzy logic theory to support end-to-end learning using complex queries with provable reasoning capability. Empirical results show that our method outperforms previous methods significantly in the new dataset and also surpasses previous methods in the existing dataset at the same time.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2304.07063")
get_code_for_paper("2304.07063")
have("2304.07063")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2304.07063.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2304.07063.svg)](https://syntology.ai/paper/2304.07063)

Connect an agent — have() is free.