SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2002.05969 · ICLR · 2020

Query2box: Reasoning over Knowledge Graphs in Vector Space using Box Embeddings

Hongyu Ren, Weihua Hu, Jure Leskovec, Canada Bieber

arXiv · PDF · Open in the Atlas

Code that ran

30 of 41 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (13 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 17 have no contract verdict; 12 failures of our instrument; 11 no recorded run), 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/2002.05969.json · badge: /api/ran/2002.05969.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
hyren/query2box canonical (classifier label)
classifier basis: classifier_2025
7 ran · 1 no recorded run
kjh9503/caqr canonical (classifier label)
classifier basis: census
1 ran · 0 no recorded run
francoisluus/KGReasoning community
Papers with Code archive
9 ran · 1 no recorded run
Blidge/KGReasoning community
Papers with Code archive
8 ran · 7 no recorded run
Jacfger/simple-stuffs community
Papers with Code archive
5 ran · 1 no recorded run
LHY-24/KG-Compilation community
Papers with Code archive
0 ran · 1 no recorded run
FunctionStatusWhere it lives
Attention Ran
ran on a synthesized input · constructed an object; no result was computed
hyren/query2box/codes/model.py
code served (permissive licence) · get_code("2575bb19d82fba6d")
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 an instance of its own class, Attention: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'mode_dims': 4,
    'expand_dims': 3,
    'center_use_offset': True,
    'att_type': 'whole',
    'bn': 'no',
    'nat': 2,
    'name': 'Test'
}
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 Attention — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Attention 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
AttentionSet Ran
ran on a synthesized input · constructed an object; no result was computed
hyren/query2box/codes/model.py
code served (permissive licence) · get_code("e519b1988ee37a5a")
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 an instance of its own class, AttentionSet: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# Define ARGS dict with all parameters for AttentionSet
ARGS = {
    'mode_dims': 4,
    'expand_dims': 3,
    'center_use_offset': True,
    'att_reg': 0.1,
    'att_tem': 1.0,
    'att_type': 'whole',
    'bn': 'no',
    'nat': 1,
    'name': 'Real'
}
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 AttentionSet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a AttentionSet 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
BetaIntersection Ran
ran on a synthesized input · constructed an object; no result was computed
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("956e9a56d057daa5")
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 an instance of its own class, BetaIntersection: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# BetaIntersection expects:
# - dim: int, embedding dimension
# - use_att: bool, whether to use attention mechanism
# - forward() receives alpha_embeddings and beta_embeddings (both tensors)

# For the forward pass:
# - alpha_embeddings and beta_embeddings should have shape (num_conj, dim)
# - When use_att=True: concatenated along dim=-1, passed through Linear layers
# - When use_att=False: summed along dim=0 and divided by num_conj

ARGS = {
    "dim": 4,
    "use_att": True,
}
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 BetaIntersection — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a BetaIntersection 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
BetaProjection Ran
ran on a synthesized input · constructed an object; no result was computed
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("c16d861762f8696c")
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 an instance of its own class, BetaProjection: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# Define a simple projection_regularizer that takes a tensor and returns a tensor
class SimpleRegularizer(nn.Module):
    def forward(self, x):
        # Just return the input as-is for testing
        return x

# Set up the arguments
entity_dim = 4
relation_dim = 3
hidden_dim = 5
num_layers = 2
projection_regularizer = SimpleRegularizer()

ARGS = {
    'entity_dim': entity_dim,
    'relation_dim': relation_dim,
    'hidden_dim': hidden_dim,
    'projection_regularizer': projection_regularizer,
    'num_layers': num_layers,
}
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 BetaProjection — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a BetaProjection 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
BoxEstimator Ran
ran on a synthesized input · constructed an object; no result was computed
Jacfger/simple-stuffs/fol/estimator_box.py
code served (permissive licence) · get_code("2a211e2e73b1cbde")
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 an instance of its own class, BoxEstimator: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "n_entity": 10,
    "n_relation": 5,
    "gamma": 12.0,
    "entity_dim": 4,
    "relation_dim": 3,
    "offset_activation": "relu",
    "center_reg": 0.1,
    "negative_sample_size": 2,
    "device": "cpu"
}
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 BoxEstimator — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a BoxEstimator 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
BoxOffsetIntersection Ran
ran on a synthesized input
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("47268ecee02ceb4f")
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 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 torch.nn.functional as F

class BoxOffsetIntersection(nn.Module):

    def __init__(self, dim):
        super(BoxOffsetIntersection, self).__init__()
        self.dim = dim
        self.layer1 = nn.Linear(self.dim, self.dim)
        self.layer2 = nn.Linear(self.dim, self.dim)
        nn.init.xavier_uniform_(self.layer1.weight)
        nn.init.xavier_uniform_(self.layer2.weight)

    def forward(self, embeddings):
        layer1_act = F.relu(self.layer1(embeddings))
        layer1_mean = torch.mean(layer1_act, dim=0)
        gate = torch.sigmoid(self.layer2(layer1_mean))
        offset, _ = torch.min(embeddings, dim=0)
        return offset * gate


INIT_ARGS = {
    'dim': 4
}

ARGS = {
    'embeddings': torch.randn(2, 4)
}

if __name__ == '__main__':
    model = BoxOffsetIntersection(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shape:", output.shape)
    print("Output:", output)
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 1, 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)
BoxOffsetIntersection Ran
ran on a synthesized input · constructed an object; no result was computed
Jacfger/simple-stuffs/fol/estimator_box.py
code served (permissive licence) · get_code("c839f4122028b419")
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 an instance of its own class, BoxOffsetIntersection: the run constructed the object and did not call it, so no result was computed. 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
from torch import nn

torch.manual_seed(0)

# The BoxOffsetIntersection module takes:
# - dim: an integer specifying the embedding dimension

# Looking at the forward method:
# - It expects embeddings: a tensor of shape (batch_size, dim)
# - It applies linear layers and operations that work with this shape

dim = 4
batch_size = 2

# Create a small embedding tensor for testing
embeddings = torch.randn(batch_size, dim)

ARGS = {
    "dim": dim
}
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 BoxOffsetIntersection — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a BoxOffsetIntersection 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
CenterIntersection Ran
ran on a synthesized input · constructed an object; no result was computed
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("a2ccec473702f3c7")
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 an instance of its own class, CenterIntersection: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# The CenterIntersection module takes embeddings as input
# The forward method expects embeddings to be a 2D tensor where:
# - It applies linear transformations
# - Uses softmax on dim=0
# - Computes weighted sum over dim=0
# So embeddings should have shape (batch_size, dim)

dim = 4
batch_size = 2

# Create sample embeddings tensor
embeddings = torch.randn(batch_size, dim)

ARGS = {
    "dim": dim,
}
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 CenterIntersection — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CenterIntersection 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
CenterIntersection Ran
ran on a synthesized input
Jacfger/simple-stuffs/fol/estimator_box.py
code served (permissive licence) · get_code("5c301bc69f133009")
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 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

# INIT_ARGS: parameters for CenterIntersection.__init__
INIT_ARGS = {
    'dim': 8
}

# ARGS: parameters for CenterIntersection.forward
# embeddings should be a 2D tensor where:
# - first dimension is the number of embeddings to intersect
# - second dimension is the embedding dimension (must match 'dim' from INIT_ARGS)
ARGS = {
    'embeddings': torch.randn(5, 8)  # 5 embeddings, each of dimension 8
}
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 Tensor, ndim 1, 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)
CenterSet Ran
ran on a synthesized input · constructed an object; no result was computed
hyren/query2box/codes/model.py
code served (permissive licence) · get_code("e6cd914f1f66ce76")
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 an instance of its own class, CenterSet: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# Define the arguments for CenterSet
mode_dims = 4
expand_dims = 3
center_use_offset = True
agg_func = torch.min
bn = 'no'
nat = 1
name = 'Real_center'

ARGS = {
    'mode_dims': mode_dims,
    'expand_dims': expand_dims,
    'center_use_offset': center_use_offset,
    'agg_func': agg_func,
    'bn': bn,
    'nat': nat,
    'name': name,
}
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 CenterSet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CenterSet 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
Identity Ran
ran, honoured the contract we drafted
kjh9503/caqr/CaQR/betae/models.py
pointer only (licence: NONE); the stored copy get_code serves, from akirato/perm-gaussiankg, is pointer only (licence: GPL-3.0) · get_code("f44985272d18127a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, kjh9503/caqr when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

InductiveOffsetSet Ran
ran on a synthesized input · constructed an object; no result was computed
hyren/query2box/codes/model.py
code served (permissive licence) · get_code("23039ea2b760df60")
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 an instance of its own class, InductiveOffsetSet: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# Define parameters for InductiveOffsetSet
mode_dims = 4
expand_dims = 3
offset_use_center = True
off_reg = 0.1
agg_func = torch.min
name = 'Real_offset'

# Create small tensors for forward pass
batch_size = 2

# embeds1, embeds2, embeds3: base embeddings of size [batch_size, mode_dims]
embeds1 = torch.randn(batch_size, mode_dims)
embeds2 = torch.randn(batch_size, mode_dims)
embeds3 = torch.randn(batch_size, mode_dims)

# embeds1_o, embeds2_o, embeds3_o: offset embeddings of size [batch_size, mode_dims]
embeds1_o = torch.randn(batch_size, mode_dims)
embeds2_o = torch.randn(batch_size, mode_dims)
embeds3_o = torch.randn(batch_size, mode_dims)

ARGS = {
    'mode_dims': mode_dims,
    'expand_dims': expand_dims,
    'offset_use_center': offset_use_center,
    'off_reg': off_reg,
    'agg_func': agg_func,
    'name': name
}
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 InductiveOffsetSet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a InductiveOffsetSet 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
LogicIntersection Ran
ran on a synthesized input · constructed an object; no result was computed
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("47248dd2646c39bd")
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 an instance of its own class, LogicIntersection: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# Build arguments for LogicIntersection
ARGS = {
    'dim': 4,
    'tnorm': 'mins',
    'bounded': True,
    'use_att': True,
    'use_gtrans': False
}
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 LogicIntersection — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a LogicIntersection 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
LogicProjection Ran
ran on a synthesized input · constructed an object; no result was computed
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("25d6c06688daacdd")
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 an instance of its own class, LogicProjection: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'entity_dim': 4,
    'relation_dim': 3,
    'hidden_dim': 8,
    'num_layers': 2,
    'bounded': True
}
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 LogicProjection — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a LogicProjection 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
MeanSet Ran
ran on a synthesized input
hyren/query2box/codes/model.py
code served (permissive licence) · get_code("c14351660e82d6b2")
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
import torch.nn as nn

# INIT_ARGS for MeanSet.__init__
INIT_ARGS = {}

# ARGS for MeanSet.forward
# The forward method expects:
# - embeds1: tensor
# - embeds1_o: tensor (not used in computation but required parameter)
# - embeds2: tensor
# - embeds2_o: tensor (not used in computation but required parameter)
# - embeds3: list or tensor (optional, default=[])
# - embeds3_o: list or tensor (optional, default=[])

# Create tensors with distinct dimensions
batch_size = 2
embed_dim = 4

embeds1 = torch.randn(batch_size, embed_dim)
embeds1_o = torch.randn(batch_size, embed_dim)
embeds2 = torch.randn(batch_size, embed_dim)
embeds2_o = torch.randn(batch_size, embed_dim)

# Test case 1: without embeds3 (will use default empty list)
ARGS = {
    'embeds1': embeds1,
    'embeds1_o': embeds1_o,
    'embeds2': embeds2,
    'embeds2_o': embeds2_o,
    'embeds3': [],
    'embeds3_o': []
}
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)
MinSet Ran
ran on a synthesized input
hyren/query2box/codes/model.py
code served (permissive licence) · get_code("060a24b37fb8ba59")
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
import torch.nn as nn

# INIT_ARGS for MinSet.__init__()
INIT_ARGS = {}

# ARGS for MinSet.forward()
batch_size = 2
feature_dim = 4

embeds1 = torch.randn(batch_size, feature_dim)
embeds1_o = torch.randn(batch_size, feature_dim)
embeds2 = torch.randn(batch_size, feature_dim)
embeds2_o = torch.randn(batch_size, feature_dim)
embeds3 = torch.randn(batch_size, feature_dim)
embeds3_o = torch.randn(batch_size, feature_dim)

ARGS = {
    'embeds1': embeds1,
    'embeds1_o': embeds1_o,
    'embeds2': embeds2,
    'embeds2_o': embeds2_o,
    'embeds3': embeds3,
    'embeds3_o': embeds3_o
}
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)
OffsetSet Ran
ran on a synthesized input · constructed an object; no result was computed
hyren/query2box/codes/model.py
code served (permissive licence) · get_code("11bb0cf115a0240f")
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 an instance of its own class, OffsetSet: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn

torch.manual_seed(0)

# OffsetSet parameters:
# - mode_dims: int, dimension of mode embeddings
# - expand_dims: int, dimension to expand to
# - offset_use_center: bool, whether to use center in offset
# - agg_func: callable, aggregation function (default torch.min)
# - name: str, name for parameter registration (default 'Real_offset')

mode_dims = 4
expand_dims = 3
offset_use_center = True
agg_func = torch.min
name = 'test_offset'

# Create example inputs for forward pass
batch_size = 2

# embeds1 and embeds2 are required (mode_dims dimensional)
embeds1 = torch.randn(batch_size, mode_dims)
embeds2 = torch.randn(batch_size, mode_dims)

# embeds1_o and embeds2_o are offsets (expand_dims dimensional)
embeds1_o = torch.randn(batch_size, expand_dims)
embeds2_o = torch.randn(batch_size, expand_dims)

# embeds3 and embeds3_o are optional
embeds3 = torch.randn(batch_size, mode_dims)
embeds3_o = torch.randn(batch_size, expand_dims)

ARGS = {
    'mode_dims': mode_dims,
    'expand_dims': expand_dims,
    'offset_use_center': offset_use_center,
    'agg_func': agg_func,
    'name': name
}
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 OffsetSet — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a OffsetSet 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
SizePredict Ran
ran on a synthesized input · constructed an object; no result was computed
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("2a2b3c074a6a00b5")
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 an instance of its own class, SizePredict: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
import torch.nn.functional as F

class SizePredict(nn.Module):

    def __init__(self, entity_dim):
        super(SizePredict, self).__init__()
        self.layer2 = nn.Linear(entity_dim, entity_dim // 4)
        self.layer1 = nn.Linear(entity_dim // 4, entity_dim // 16)
        self.layer0 = nn.Linear(entity_dim // 16, 1)
        nn.init.xavier_uniform_(self.layer2.weight)
        nn.init.xavier_uniform_(self.layer1.weight)
        nn.init.xavier_uniform_(self.layer0.weight)

    def forward(self, entropy_embedding):
        x = self.layer2(entropy_embedding)
        x = F.relu(x)
        x = self.layer1(x)
        x = F.relu(x)
        x = self.layer0(x)
        x = torch.sigmoid(x)
        return x.squeeze()

torch.manual_seed(0)

entity_dim = 16
entropy_embedding = torch.randn(2, entity_dim)

ARGS = {
    "entity_dim": entity_dim
}
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 SizePredict — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a SizePredict 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
find_optimal_batch Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Jacfger/simple-stuffs/fol/estimator_box.py
code served (permissive licence) · get_code("d8b0a4e3912306de")
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 int, ndim 0, dtype family int, shape () 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
import torch.nn as nn

torch.manual_seed(0)

# Create tiny tensors with distinct dimensions
answer_emb = torch.randn(2, 4)  # batch_size=2, embedding_dim=4
query_dist = torch.randn(3, 5)  # distinct shape for query_dist

# Define a simple compute_logit function that takes batched answer embeddings and query_dist
def compute_logit(batch_answer_emb, query_dist):
    # The function calls this with either:
    # - batch_answer_emb.unsqueeze(0) when union=False (adds 1 dim)
    # - batch_answer_emb.unsqueeze(0).unsqueeze(0) when union=True (adds 2 dims)
    # So batch_answer_emb will have shape (..., 4) and we need to return a logit tensor
    # This simple implementation just returns a scalar logit
    return torch.tensor(1.0)

ARGS = {
    "answer_emb": answer_emb,
    "query_dist": query_dist,
    "compute_logit": compute_logit,
    "union": False
}
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 int, ndim 0, dtype family int, shape () — shape () 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
flatten_structure Ran
ran; our contract draft was wrong, not the code
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("d7c6b03fd8f6a3bc")
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 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 flatten_structure recursively flattens a nested structure
# It takes query_structure as input which can be either:
# - a string (base case)
# - an iterable of nested structures (recursive case)

# Test case 1: Simple string
simple_string = "query"

# Test case 2: Nested list structure with strings
nested_structure = [
    "entity",
    ["relation", "entity"],
    [["relation", "entity"], "entity"]
]

# Test case 3: Mixed depth structure
mixed_structure = [
    "start",
    ["middle1", "middle2"],
    [["deep1", "deep2"], ["deep3", "deep4"]]
]

ARGS = {
    "query_structure": nested_structure
}
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 1, 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
inclusion_sampling Ran
ran; our contract draft was wrong, not the code
Jacfger/simple-stuffs/fol/estimator_box.py
code served (permissive licence) · get_code("c423c8497887b2c3")
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. 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
import random

torch.manual_seed(0)
np.random.seed(0)
random.seed(0)

# answer_set: List[IntList] - a list of lists of integers
# Each inner list represents the set of valid answers for a particular query
answer_set = [
    [0, 1, 2],      # First query has answers {0, 1, 2}
    [3, 4],         # Second query has answers {3, 4}
]

# negative_size: int - number of negative samples to draw per query
negative_size = 5

# entity_num: int - total number of entities (upper bound for negative sampling)
entity_num = 10

# k: int - number of answers to sample from each answer set (default=1)
k = 1

# base_num: int - base number for subsampling weight calculation (default=4)
base_num = 4

ARGS = {
    "answer_set": answer_set,
    "negative_size": negative_size,
    "entity_num": entity_num,
    "k": k,
    "base_num": base_num,
}
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 tuple — a tuple output has no array shape of its own; none is 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
order_bounds Ran
ran; our contract draft was wrong, not the code
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("6df8d7def6d8b856")
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 2, dtype family float, shape (2, 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

torch.manual_seed(0)

# Create a small embedding tensor
# Shape: (batch_size=2, embedding_dim=4)
# The function expects the last dimension to be split into lower and upper bounds
embedding = torch.rand(2, 4)

ARGS = {
    "embedding": embedding
}
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 2, dtype family float, shape (2, 4) — shape (2, 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: none by the harness; any seed is in the setup block
query_2p Ran
ran; our contract draft was wrong, not the code
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("c347b52e0b2d0f36")
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 2, dtype family float, shape (2, 8) 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
import torch.nn as nn
from typing import Tuple

torch.manual_seed(0)

# Define a simple embedding module that stores embeddings and can be called
class SimpleEmbedding(nn.Module):
    def __init__(self, num_embeddings, embedding_dim):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(num_embeddings, embedding_dim))
    
    def forward(self, indices):
        return torch.nn.functional.embedding(indices, self.weight)

# Define entity and predicate embeddings
num_entities = 8
num_predicates = 6
embedding_dim = 4

entity_embeddings = SimpleEmbedding(num_entities, embedding_dim)
predicate_embeddings = SimpleEmbedding(num_predicates, embedding_dim)

# Define queries: shape (batch_size=2, 3) with indices
# queries[:, 0] = subject entity indices
# queries[:, 1] = first predicate indices
# queries[:, 2] = second predicate indices
queries = torch.tensor([
    [0, 1, 2],
    [3, 4, 5]
], dtype=torch.long)

# Define a simple scoring function that takes (s_emb, p_emb, candidates_emb) and returns scores
def scoring_function(s_emb: torch.Tensor, p_emb: torch.Tensor, candidates_emb: torch.Tensor) -> torch.Tensor:
    # s_emb: (batch_size, embedding_dim)
    # p_emb: (batch_size, embedding_dim)
    # candidates_emb: (num_entities, embedding_dim)
    # Returns: (batch_size, num_entities)
    batch_size = s_emb.shape[0]
    num_candidates = candidates_emb.shape[0]
    
    # Simple scoring: dot product of (s_emb + p_emb) with each candidate
    combined = s_emb + p_emb  # (batch_size, embedding_dim)
    scores = torch.matmul(combined, candidates_emb.t())  # (batch_size, num_entities)
    return scores

# Define t_norm (triangular norm) - using minimum as a simple t-norm
def t_norm(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
    # Element-wise minimum
    return torch.min(a, b)

k = 3

ARGS = {
    "entity_embeddings": entity_embeddings,
    "predicate_embeddings": predicate_embeddings,
    "queries": queries,
    "scoring_function": scoring_function,
    "k": k,
    "t_norm": t_norm
}
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 2, dtype family float, shape (2, 8) — shape (2, 8) 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
query_3i Ran
ran; our contract draft was wrong, not the code
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("52f5e28fee4638ab")
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 2, dtype family float, shape (2, 6) 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
import torch.nn as nn
from typing import Callable, Tuple, Optional

torch.manual_seed(0)

# Create entity embeddings module
class EntityEmbeddings(nn.Module):
    def __init__(self, num_entities: int, embedding_dim: int):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(num_entities, embedding_dim))
    
    def forward(self, indices: torch.Tensor) -> torch.Tensor:
        return torch.nn.functional.embedding(indices, self.weight)

# Create predicate embeddings module
class PredicateEmbeddings(nn.Module):
    def __init__(self, num_predicates: int, embedding_dim: int):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(num_predicates, embedding_dim))
    
    def forward(self, indices: torch.Tensor) -> torch.Tensor:
        return torch.nn.functional.embedding(indices, self.weight)

# Create a simple scoring function
def scoring_function(s_emb: torch.Tensor, p_emb: torch.Tensor, candidates_emb: torch.Tensor) -> torch.Tensor:
    # Simple dot product scoring: (batch_size, num_candidates)
    # s_emb: (batch_size, embedding_dim)
    # p_emb: (batch_size, embedding_dim)
    # candidates_emb: (num_candidates, embedding_dim)
    combined = s_emb + p_emb  # (batch_size, embedding_dim)
    scores = torch.mm(combined, candidates_emb.t())  # (batch_size, num_candidates)
    return scores

# Create a simple t-norm function (min operator)
def t_norm(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
    return torch.min(x, y)

# Initialize embeddings
num_entities = 6
num_predicates = 5
embedding_dim = 4

entity_embeddings = EntityEmbeddings(num_entities, embedding_dim)
predicate_embeddings = PredicateEmbeddings(num_predicates, embedding_dim)

# Create queries for query_3i
# query_3i expects queries with shape (batch_size, 6) where each pair represents (subject, predicate)
# Format: [s1, p1, s2, p2, s3, p3]
batch_size = 2
queries = torch.tensor([
    [0, 0, 1, 1, 2, 2],
    [3, 1, 4, 2, 5, 3]
], dtype=torch.long)

ARGS = {
    "entity_embeddings": entity_embeddings,
    "predicate_embeddings": predicate_embeddings,
    "queries": queries,
    "scoring_function": scoring_function,
    "t_norm": t_norm
}
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 2, dtype family float, shape (2, 6) — shape (2, 6) 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
query_3p Ran
ran; our contract draft was wrong, not the code
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("805d2b1bc7c0dc90")
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 2, dtype family float, shape (2, 10) 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
import torch.nn as nn

torch.manual_seed(0)

# Define a simple embedding module
class SimpleEmbedding(nn.Module):
    def __init__(self, num_embeddings, embedding_dim):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(num_embeddings, embedding_dim))
    
    def forward(self, indices):
        return torch.nn.functional.embedding(indices, self.weight)

# Define a simple scoring function
def simple_scoring_function(s_emb, p_emb, candidates_emb):
    # s_emb: [batch_size, emb_dim]
    # p_emb: [batch_size, emb_dim]
    # candidates_emb: [num_candidates, emb_dim]
    # Output: [batch_size, num_candidates]
    s_expanded = s_emb.unsqueeze(2)  # [batch_size, emb_dim, 1]
    p_expanded = p_emb.unsqueeze(2)  # [batch_size, emb_dim, 1]
    combined = s_expanded + p_expanded  # [batch_size, emb_dim, 1]
    scores = torch.matmul(combined.squeeze(2), candidates_emb.t())  # [batch_size, num_candidates]
    return scores

# Define a simple t-norm (min operation)
def simple_t_norm(a, b):
    return torch.min(a, b)

# Create instances
num_entities = 10
num_predicates = 5
embedding_dim = 4
batch_size = 2
k = 3

entity_embeddings = SimpleEmbedding(num_entities, embedding_dim)
predicate_embeddings = SimpleEmbedding(num_predicates, embedding_dim)

# Create queries: [batch_size, 4] where each row is [subject, pred1, pred2, pred3]
queries = torch.randint(0, num_entities, (batch_size, 1), dtype=torch.long)
queries = torch.cat([
    queries,
    torch.randint(0, num_predicates, (batch_size, 3), dtype=torch.long)
], dim=1)

ARGS = {
    "entity_embeddings": entity_embeddings,
    "predicate_embeddings": predicate_embeddings,
    "queries": queries,
    "scoring_function": simple_scoring_function,
    "k": k,
    "t_norm": simple_t_norm,
}
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 2, dtype family float, shape (2, 10) — shape (2, 10) 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
query_pi Ran
ran; our contract draft was wrong, not the code
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("51a131fbee1009cf")
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 2, dtype family float, shape (2, 10) 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
import torch.nn as nn
from typing import Callable

torch.manual_seed(0)

# Define entity embeddings module
class EntityEmbeddings(nn.Module):
    def __init__(self, num_entities: int, embedding_dim: int):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(num_entities, embedding_dim))
    
    def forward(self, indices: torch.Tensor) -> torch.Tensor:
        return self.weight[indices]

# Define predicate embeddings module
class PredicateEmbeddings(nn.Module):
    def __init__(self, num_predicates: int, embedding_dim: int):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(num_predicates, embedding_dim))
    
    def forward(self, indices: torch.Tensor) -> torch.Tensor:
        return self.weight[indices]

# Create embedding modules
num_entities = 10
num_predicates = 5
embedding_dim = 4

entity_embeddings = EntityEmbeddings(num_entities, embedding_dim)
predicate_embeddings = PredicateEmbeddings(num_predicates, embedding_dim)

# Define a simple scoring function
def scoring_function(s_emb: torch.Tensor, p_emb: torch.Tensor, candidates_emb: torch.Tensor) -> torch.Tensor:
    # Compute scores: batch_size x num_candidates
    # s_emb: (batch_size, embedding_dim)
    # p_emb: (batch_size, embedding_dim)
    # candidates_emb: (num_candidates, embedding_dim)
    s_expanded = s_emb.unsqueeze(2)  # (batch_size, embedding_dim, 1)
    p_expanded = p_emb.unsqueeze(2)  # (batch_size, embedding_dim, 1)
    combined = s_expanded + p_expanded  # (batch_size, embedding_dim, 1)
    scores = torch.matmul(combined.squeeze(2), candidates_emb.t())  # (batch_size, num_candidates)
    return scores

# Define a t-norm function (element-wise minimum)
def t_norm(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:
    return torch.minimum(x, y)

# Create queries tensor for query_pi
# query_pi expects queries with shape (batch_size, 5):
# - queries[:, 0:3] for query_2p (subject, predicate1, predicate2)
# - queries[:, 3:5] for query_1p (subject, predicate)
batch_size = 2
queries = torch.randint(0, num_entities, (batch_size, 5))
# Make sure first column (subject) is valid
queries[:, 0] = torch.randint(0, num_entities, (batch_size,))
# Make sure predicate indices are valid
queries[:, 1] = torch.randint(0, num_predicates, (batch_size,))
queries[:, 2] = torch.randint(0, num_predicates, (batch_size,))
queries[:, 3] = torch.randint(0, num_entities, (batch_size,))
queries[:, 4] = torch.randint(0, num_predicates, (batch_size,))

k = 3

ARGS = {
    'entity_embeddings': entity_embeddings,
    'predicate_embeddings': predicate_embeddings,
    'queries': queries,
    'scoring_function': scoring_function,
    'k': k,
    't_norm': t_norm
}
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 2, dtype family float, shape (2, 10) — shape (2, 10) 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
query_to_atoms Ran
ran; our contract draft was wrong, not the code
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("bc5f262d753a9eb4")
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. 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)

# Simple query structure examples
# The function expects either a string or a nested list of strings
# Strings can be 'r' (relation), 'e' (entity), 'u' (union), 'n' (negation)

# Example: a simple query structure like [['e', 'r', 'e'], 'r', 'e']
query_structure = [['e', 'r', 'e'], 'r', 'e']

# flat_ids should be a tensor of shape (batch_size, query_length)
# where query_length matches the flattened structure length
# The flattened structure of above is: ['e', 'r', 'e', 'r', 'e'] - length 5
batch_size = 2
query_length = 5

# Create flat_ids with entity and relation IDs
# Entities: 0, 1, 2, 3...
# Relations: 100, 101, 102...
flat_ids = torch.tensor([
    [0, 100, 1, 101, 2],
    [3, 102, 4, 103, 5]
], dtype=torch.long)

ARGS = {
    "query_structure": query_structure,
    "flat_ids": flat_ids
}
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 — a tuple output has no array shape of its own; none is 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
query_up_dnf Ran
ran; our contract draft was wrong, not the code
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("02f4d333d76fc417")
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 2, dtype family float, shape (2, 8) 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
import torch.nn as nn
from typing import Callable, Tuple, Optional

torch.manual_seed(0)

# Define entity embeddings module
class EntityEmbeddings(nn.Module):
    def __init__(self, num_entities: int, embedding_dim: int):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(num_entities, embedding_dim))
    
    def forward(self, indices: torch.Tensor) -> torch.Tensor:
        return torch.nn.functional.embedding(indices, self.weight)

# Define predicate embeddings module
class PredicateEmbeddings(nn.Module):
    def __init__(self, num_predicates: int, embedding_dim: int):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(num_predicates, embedding_dim))
    
    def forward(self, indices: torch.Tensor) -> torch.Tensor:
        return torch.nn.functional.embedding(indices, self.weight)

# Create instances
num_entities = 8
num_predicates = 6
embedding_dim = 4

entity_embeddings = EntityEmbeddings(num_entities, embedding_dim)
predicate_embeddings = PredicateEmbeddings(num_predicates, embedding_dim)

# Create queries tensor: shape (batch_size, 6)
# Each query has 6 indices: [s1, p1, s2, p2, p_unused, p5]
batch_size = 2
queries = torch.tensor([
    [0, 1, 2, 3, 4, 5],
    [1, 2, 3, 4, 5, 0]
], dtype=torch.long)

# Define scoring function
def scoring_function(s_emb: torch.Tensor, p_emb: torch.Tensor, candidates_emb: torch.Tensor) -> torch.Tensor:
    """Compute scores between subject+predicate and candidates"""
    batch_size = s_emb.shape[0]
    num_candidates = candidates_emb.shape[0]
    
    # Simple scoring: dot product of (s_emb + p_emb) with each candidate
    s_p = s_emb + p_emb  # (batch_size, embedding_dim)
    scores = torch.mm(s_p, candidates_emb.t())  # (batch_size, num_candidates)
    return scores

# Define t_norm function (min operation for fuzzy logic)
def t_norm(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
    return torch.minimum(a, b)

# Define t_conorm function (max operation for fuzzy logic)
def t_conorm(a: torch.Tensor, b: torch.Tensor) -> torch.Tensor:
    return torch.maximum(a, b)

k = 3

ARGS = {
    "entity_embeddings": entity_embeddings,
    "predicate_embeddings": predicate_embeddings,
    "queries": queries,
    "scoring_function": scoring_function,
    "k": k,
    "t_norm": t_norm,
    "t_conorm": t_conorm,
}
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 2, dtype family float, shape (2, 8) — shape (2, 8) 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
score_candidates Ran
ran; our contract draft was wrong, not the code
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("b97e0b8fe06b9171")
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. 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 torch.nn as nn
from typing import Tuple, Optional

torch.manual_seed(0)

# Create a simple entity embedding module that takes indices and returns embeddings
class SimpleEntityEmbedding(nn.Module):
    def __init__(self, num_entities: int, embedding_dim: int):
        super().__init__()
        self.embedding = nn.Embedding(num_entities, embedding_dim)
    
    def forward(self, indices: torch.Tensor) -> torch.Tensor:
        return self.embedding(indices)

# Create a simple scoring function
def simple_scoring_function(s_emb: torch.Tensor, p_emb: torch.Tensor, candidates_emb: torch.Tensor) -> torch.Tensor:
    # s_emb: (batch_size, embedding_size)
    # p_emb: (batch_size, embedding_size)
    # candidates_emb: (num_candidates, embedding_size)
    # Returns: (batch_size, num_candidates)
    batch_size = s_emb.shape[0]
    num_candidates = candidates_emb.shape[0]
    
    # Simple scoring: dot product of (s_emb + p_emb) with each candidate
    combined = s_emb + p_emb  # (batch_size, embedding_size)
    scores = torch.matmul(combined, candidates_emb.t())  # (batch_size, num_candidates)
    return scores

# Set up parameters
batch_size = 2
embedding_size = 4
num_candidates = 7
num_entities = 10
k = 3

# Create tensors
s_emb = torch.randn(batch_size, embedding_size)
p_emb = torch.randn(batch_size, embedding_size)
candidates_emb = torch.randn(num_candidates, embedding_size)

# Create entity embeddings module
entity_embeddings = SimpleEntityEmbedding(num_entities=num_entities, embedding_dim=embedding_size)

# Build the arguments dictionary
ARGS = {
    's_emb': s_emb,
    'p_emb': p_emb,
    'candidates_emb': candidates_emb,
    'k': k,
    'entity_embeddings': entity_embeddings,
    'scoring_function': simple_scoring_function
}
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 — a tuple output has no array shape of its own; none is 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
valclamp Ran
ran; our contract draft was wrong, not the code
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("51d9d0e0413cd046")
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 2, dtype family float, shape (2, 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

torch.manual_seed(0)

# Create a small tensor input
x = torch.randn(2, 4)

# Define parameters
a = 1
b = 6
lo = 0
hi = 1

# Construct ARGS dictionary with all parameters
ARGS = {
    "x": x,
    "a": a,
    "b": b,
    "lo": lo,
    "hi": hi
}
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 2, dtype family float, shape (2, 4) — shape (2, 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: none by the harness; any seed is in the setup block
AppFOQEstimator Not yet run
unverified (harvested, no recorded run)
Jacfger/simple-stuffs/fol/estimator_box.py
code served (permissive licence) · get_code("b8322293c1e3a4ac")
How it ran

no invocation recorded.

CQD Not yet run
unverified (harvested, no recorded run)
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("9b8c70940dfe67a9")
How it ran

no invocation recorded.

KGReasoning Not yet run
unverified (harvested, no recorded run)
francoisluus/KGReasoning/models.py
code served (permissive licence) · get_code("87b741616782eced")
How it ran

no invocation recorded.

KGReasoning Not yet run
unverified (harvested, no recorded run)
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("61e00b75af63ddec")
How it ran

no invocation recorded.

KGReasoning Not yet run
unverified (harvested, no recorded run)
LHY-24/KG-Compilation/src/models.py
pointer only (licence: NONE) · get_code("823b93b900c73ef8")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, LHY-24/KG-Compilation when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

N3 Not yet run
unverified (harvested, no recorded run)
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("13dc0495d6916564")
How it ran

no invocation recorded.

Query2box Not yet run
unverified (harvested, no recorded run)
hyren/query2box/codes/model.py
code served (permissive licence) · get_code("5cf4606d614eb571")
How it ran

no invocation recorded.

query_1p Not yet run
unverified (harvested, no recorded run)
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("38511adc942a62d6")
How it ran

no invocation recorded.

query_2i Not yet run
unverified (harvested, no recorded run)
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("b821d97259813ae4")
How it ran

no invocation recorded.

query_2u_dnf Not yet run
unverified (harvested, no recorded run)
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("c5eb44c00d8b7df8")
How it ran

no invocation recorded.

query_ip Not yet run
unverified (harvested, no recorded run)
Blidge/KGReasoning/models.py
code served (permissive licence) · get_code("2493b517068163e8")
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 39 distinct functions with an identity, 22 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 22 later) and 17 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
Identity
brick of this paper's assembly: helper
7 other papers
2 dated earlier, 5 later
1806.01445 (2018-06)
Embedding Logical Queries on Knowledge Graphs
1909.05311 (2019-09)
Graph-Based Reasoning over Heterogeneous External Knowledge for Commonsense Question Answering
2006.12862 (2020-06)
Automatic Data Augmentation for Generalization in Deep Reinforcement Learning
and 4 more not named here
BetaProjection
brick of this paper's assembly: model
2 other papers
2 later · 1 only as an assembly brick
2010.11465 (2020-10)
Beta Embeddings for Multi-Hop Logical Reasoning in Knowledge Graphs
2110.13522 (2021-10)
Probabilistic Entity Representation Model for Reasoning over Knowledge Graphs
find_optimal_batch
brick of this paper's assembly: helper
2 other papers
2 later · 1 only as an assembly brick
2010.11465 (2020-10)
Beta Embeddings for Multi-Hop Logical Reasoning in Knowledge Graphs
2109.08925 (2021-09)
Benchmarking the Combinatorial Generalizability of Complex Query Answering on Knowledge Graphs
inclusion_sampling
brick of this paper's assembly: helper
2 other papers
2 later · 1 only as an assembly brick
2010.11465 (2020-10)
Beta Embeddings for Multi-Hop Logical Reasoning in Knowledge Graphs
2109.08925 (2021-09)
Benchmarking the Combinatorial Generalizability of Complex Query Answering on Knowledge Graphs
AppFOQEstimator
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2010.11465 (2020-10)
Beta Embeddings for Multi-Hop Logical Reasoning in Knowledge Graphs
BetaIntersection
brick of this paper's assembly: model
1 other paper
1 later · 1 only as an assembly brick
2010.11465 (2020-10)
Beta Embeddings for Multi-Hop Logical Reasoning in Knowledge Graphs
CQD
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
N3
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
flatten_structure
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
order_bounds
brick of this paper's assembly: helper
1 other paper
1 later
2109.08925 (2021-09)
Benchmarking the Combinatorial Generalizability of Complex Query Answering on Knowledge Graphs
query_1p
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_2i
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_2p
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_2u_dnf
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_3i
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_3p
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_ip
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_pi
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_to_atoms
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors
query_up_dnf
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2011.03459 (2020-11)
Complex Query Answering with Neural Link Predictors

Shown: the 20 functions present in the most other papers. 2 more shared functions are not listed here.

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

Answering complex logical queries on large-scale incomplete knowledge graphs (KGs) is a fundamental yet challenging task. Recently, a promising approach to this problem has been to embed KG entities as well as the query into a vector space such that entities that answer the query are embedded close to the query. However, prior work models queries as single points in the vector space, which is problematic because a complex query represents a potentially large set of its answer entities, but it is unclear how such a set can be represented as a single point. Furthermore, prior work can only handle queries that use conjunctions (∧) and existential quantifiers (∃). Handling queries with logical disjunctions (∨) remains an open problem. Here we propose QUERY2BOX, an embedding-based framework for reasoning over arbitrary queries with ∧, ∨, and ∃ operators in massive and incomplete KGs. Our main insight is that queries can be embedded as boxes (i.e., hyper-rectangles), where a set of points inside the box corresponds to a set of answer entities of the query. We show that conjunctions can be naturally represented as intersections of boxes and also prove a negative result that handling disjunctions would require embedding with dimension proportional to the number of KG entities. However, we show that by transforming queries into a Disjunctive Normal Form, QUERY2BOX is capable of handling arbitrary logical queries with ∧, ∨, ∃ in a scalable manner. We demonstrate the effectiveness of QUERY2BOX on three large KGs and show that QUERY2BOX achieves up to 25% relative improvement over the state of the art.

For agents

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

get_harvested_code_for_paper("2002.05969")
get_code_for_paper("2002.05969")
have("2002.05969")

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

curl https://syntology.ai/api/ran/2002.05969.json

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

[![Syntology run record](https://syntology.ai/api/ran/2002.05969.svg)](https://syntology.ai/paper/2002.05969)

Connect an agent — have() is free.