SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2504.04798 · 2025

TabRep: Training Tabular Diffusion Models with a Simple and Effective Continuous Representation

arXiv · PDF · Open in the Atlas

Code that ran

18 of 19 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; 10 failures of our instrument; 1 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/2504.04798.json · badge: /api/ran/2504.04798.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
jacobyhsi/TabRep canonical (classifier label)
classifier basis not recorded on the link
18 ran · 1 no recorded run
FunctionStatusWhere it lives
approx_standard_normal_cdf Ran
ran, honoured the contract we drafted
jacobyhsi/TabRep/tabrep_ddpm/models/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from tsinghua-fib-lab/Spatio-temporal-Diffusion-Point-Processes, is pointer only (licence: NONE) · get_code("9deab2a47adbf3b7")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from tsinghua-fib-lab/Spatio-temporal-Diffusion-Point-Processes, and GitHub detected no licence in that repository 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.

betas_for_alpha_bar Ran
ran, honoured the contract we drafted
jacobyhsi/TabRep/tabrep_ddpm/models/gaussian_multinomial_distribution.py
pointer only (licence: MIT); the stored copy get_code serves, from blt2114/twisted_diffusion_sampler, is pointer only (licence: NONE) · get_code("6cdc8772f5db6f48")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from blt2114/twisted_diffusion_sampler, and GitHub detected no licence in that repository 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.

bits_needed Ran
ran on a synthesized input
jacobyhsi/TabRep/tabrep_ddpm/sample.py
code served (permissive licence) · get_code("12ed451084a39969")
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 ndarray, ndim 1, dtype family int, shape (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 numpy as np

# The function takes 'categories' as input and returns 2 * np.ones_like(categories)
# This means categories should be a numpy array (or array-like) that supports np.ones_like()

# Create a small categories array
categories = np.array([1, 2, 3, 4])

ARGS = {
    "categories": categories
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (4,) — shape (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
cat2psk Ran
ran on a synthesized input
jacobyhsi/TabRep/tabrep_ddpm/models/gaussian_multinomial_distribution.py
code served (permissive licence) · get_code("f013f5243dee7c34")
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 math

torch.manual_seed(0)

# Create a small tensor with shape (2, 4) - batch size 2, 4 categories
tensor = torch.randint(0, 5, (2, 4), dtype=torch.float32)

# categories should be a list/tensor of category counts
# The function divides 2*pi by each category count
categories = [3, 4, 5, 6]

ARGS = {
    "tensor": tensor,
    "categories": categories
}
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
convert_psk Ran
ran on a synthesized input
jacobyhsi/TabRep/tabrep_flow/psk_utils.py
code served (permissive licence) · get_code("0730a0bcb1bd3048")
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 math

torch.manual_seed(0)

# Analyze the function:
# - tensor: a torch tensor that gets converted to int, has .device attribute, supports indexing with dim=-1
# - categories: passed to torch.tensor(), so should be a list/array of integers
# 
# The function:
# 1. Gets device from tensor
# 2. Converts tensor to int
# 3. Creates cat_spacings from categories (list of integers)
# 4. Multiplies tensor by cat_spacings (broadcasting)
# 5. Applies sin and cos
# 6. Concatenates along dim=-1
#
# So tensor should be a small multi-dimensional float tensor
# categories should be a list of integers

# Create a small tensor: batch_size=2, some internal dims that will work with categories
# Let's make it shape (2, 4) - batch of 2, with 4 features
tensor = torch.randn(2, 4, dtype=torch.float32)

# categories should be a list of integers that broadcasts with tensor
# If tensor is (2, 4), categories could be a scalar or shape (4,)
# Looking at the code: cat_spacings = 2 * math.pi / torch.tensor(categories).to(device)
# Then: phases = tensor * cat_spacings
# This suggests categories should have compatible shape for broadcasting
# Let's use a list of 4 integers (one per feature dimension)
categories = [2, 3, 4, 5]

ARGS = {
    "tensor": tensor,
    "categories": categories,
}
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
discretized_gaussian_log_likelihood Ran
ran; our contract draft was wrong, not the code
jacobyhsi/TabRep/tabrep_ddpm/models/utils.py
code served (permissive licence) · get_code("6b6f39a3cb3edfa0")
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
import numpy as np

torch.manual_seed(0)

# Create small tensors with distinct sizes for different dimensions
batch_size = 2
dim = 4

# x: target images rescaled to [-1, 1]
x = torch.randn(batch_size, dim) * 0.5

# means: Gaussian mean, same shape as x
means = torch.randn(batch_size, dim) * 0.3

# log_scales: Gaussian log stddev, same shape as x
log_scales = torch.randn(batch_size, dim) * 0.1

ARGS = {
    "x": x,
    "means": means,
    "log_scales": log_scales,
}
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 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
geglu Ran
ran; our contract draft was wrong, not the code
jacobyhsi/TabRep/tabrep_ddpm/models/modules.py
pointer only (licence: MIT); the stored copy get_code serves, from fangzy96/tabcutmix, is pointer only (licence: NONE) · get_code("49f8fe0655f00ee5")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from fangzy96/tabcutmix, and GitHub detected no licence in that repository 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.

geglu Ran
ran; our contract draft was wrong, not the code
jacobyhsi/TabRep/tabrep_flow/models/modules.py
code served (permissive licence) · get_code("3971992cf9c5dcfc")
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, 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
from torch import Tensor

torch.manual_seed(0)

# geglu expects a Tensor with last dimension being even
# It chunks along the last dimension, applies GELU to one half and multiplies with the other
# Create a small tensor: batch_size=2, feature_dim=4 (must be even)
x = torch.randn(2, 4)

ARGS = {
    "x": x
}
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, 2) — shape (2, 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
get_named_beta_schedule Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jacobyhsi/TabRep/tabrep_ddpm/models/gaussian_multinomial_distribution.py
code served (permissive licence) · get_code("141937afc98f8d1e")
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 ndarray, ndim 1, dtype family float, shape (100,) 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 math
import numpy as np

# Test with both schedule types
ARGS = {
    'schedule_name': 'cosine',
    'num_diffusion_timesteps': 100
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (100,) — shape (100,) 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
normal_kl Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jacobyhsi/TabRep/tabrep_ddpm/models/utils.py
code served (permissive licence) · get_code("4707efa98ac6b989")
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 small tensors for KL divergence computation
# The function expects mean and logvar parameters for two Gaussians
mean1 = torch.randn(2, 4)
logvar1 = torch.randn(2, 4)
mean2 = torch.randn(2, 4)
logvar2 = torch.randn(2, 4)

ARGS = {
    "mean1": mean1,
    "logvar1": logvar1,
    "mean2": mean2,
    "logvar2": logvar2,
}
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
pad_t_like_x Ran
ran; our contract draft was wrong, not the code
jacobyhsi/TabRep/tabrep_flow/models/flow_matching.py
code served (permissive licence) · get_code("eeb5b8effffc6cdd")
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 4, dtype family float, shape (2, 1, 1, 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: t is a tensor, x is 4D (batch, channels, height, width)
t = torch.randn(2)  # shape (bs,)
x = torch.randn(2, 3, 4, 5)  # shape (bs, C, W, H)

ARGS = {
    "t": t,
    "x": x,
}
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 4, dtype family float, shape (2, 1, 1, 1) — shape (2, 1, 1, 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
psk2cat Ran
ran on a synthesized input
jacobyhsi/TabRep/tabrep_flow/psk_utils.py
code served (permissive licence) · get_code("6aa6248e94eb7cb2")
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 int, 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
import math

torch.manual_seed(0)

# psk_tensor: should be a 2D tensor with shape (batch_size, 2*cat_len)
# The function splits it into sin_phases and cos_phases, each of shape (batch_size, cat_len)
batch_size = 2
cat_len = 4
psk_tensor = torch.randn(batch_size, 2 * cat_len)

# categories: should be a tensor or array-like with shape (cat_len,)
# The function uses categories.shape[0] to get cat_len
# It also compares feature_cat values against each category value
categories = torch.tensor([2, 3, 2, 4])

ARGS = {
    "psk_tensor": psk_tensor,
    "categories": categories,
}
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 int, 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
reglu Ran
ran; our contract draft was wrong, not the code
jacobyhsi/TabRep/tabrep_ddpm/models/modules.py
pointer only (licence: MIT); the stored copy get_code serves, from fangzy96/tabcutmix, is pointer only (licence: NONE) · get_code("58bd3831c7fb6729")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from fangzy96/tabcutmix, and GitHub detected no licence in that repository 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.

reglu Ran
ran; our contract draft was wrong, not the code
jacobyhsi/TabRep/tabrep_flow/models/modules.py
code served (permissive licence) · get_code("7944b6aa2da08a07")
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, 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
import torch.nn.functional as F

torch.manual_seed(0)

# Create a small tensor with even last dimension for reglu
# reglu splits the last dimension in half, so we need shape[-1] to be even
# Using batch_size=2, feature_dim=4 (which is even)
x = torch.randn(2, 4)

ARGS = {
    "x": x
}
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, 2) — shape (2, 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
split_num_cat_target Ran
ran on a synthesized input
jacobyhsi/TabRep/tabrep_ddpm/sample.py
code served (permissive licence) · get_code("b5a210f358403b55")
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

torch.manual_seed(0)

# num_inverse: callable that takes ndarray and returns ndarray
def num_inverse_fn(x):
    return x * 2.0

# cat_inverse: callable that takes tensor and returns tensor
def cat_inverse_fn(x):
    return x.float() * 3.0

# Build synthetic data: 2 samples, 7 features total
# Let's say: 3 numeric features + 2 categorical features + 2 target columns
# For regression: target is numeric (2 cols), so n_num_feat = 3 + 2 = 5
# categorical features: 2 cols, so n_cat_feat = 2
# Total columns: 5 + 2 = 7

syn_data = np.array([
    [1.0, 2.0, 3.0, 0.5, 0.7, 1, 2],  # sample 1
    [4.0, 5.0, 6.0, 0.9, 0.3, 0, 1],  # sample 2
], dtype=np.float32)

info = {
    'task_type': 'regression',  # regression task
    'num_col_idx': [0, 1, 2],    # 3 numeric feature columns
    'cat_col_idx': [3, 4],        # 2 categorical feature columns
    'target_col_idx': [5, 6],     # 2 target columns (numeric for regression)
}

ARGS = {
    'syn_data': syn_data,
    'info': info,
    'num_inverse': num_inverse_fn,
    'cat_inverse': cat_inverse_fn,
}
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
split_num_cat_target Ran
ran on a synthesized input
jacobyhsi/TabRep/tabrep_flow/sample.py
code served (permissive licence) · get_code("b1d5215baceb33a4")
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

torch.manual_seed(0)

# Helper functions for inverse transforms
def num_inverse(x):
    """Convert normalized numeric data back to original scale"""
    # Simple inverse: multiply by 100 and add 10
    return x * 100 + 10

def cat_inverse(x):
    """Convert encoded categorical data back to original categories"""
    # Simple inverse: round and convert to int, then map to category strings
    result = np.round(x).astype(int)
    # Map 0->cat_a, 1->cat_b, 2->cat_c
    categories = ['cat_a', 'cat_b', 'cat_c']
    return np.array([[categories[int(val) % 3] for val in row] for row in result])

# Construct synthetic data
# For regression task: 2 numeric features + 1 numeric target, 3 categorical features
# Total: 3 numeric columns, 3 categorical columns
# Shape: (batch_size=2, total_features=6)
syn_data = np.array([
    [0.1, 0.2, 0.3, 0.4, 0.5, 0.6],
    [0.15, 0.25, 0.35, 0.45, 0.55, 0.65]
], dtype=np.float32)

# Info dict for regression task
info_regression = {
    'task_type': 'regression',
    'num_col_idx': [0, 1],      # 2 numeric feature indices
    'cat_col_idx': [0, 1, 2],   # 3 categorical feature indices
    'target_col_idx': [0],      # 1 numeric target index
}

# Info dict for classification task
info_classification = {
    'task_type': 'classification',
    'num_col_idx': [0, 1],      # 2 numeric feature indices
    'cat_col_idx': [0, 1],      # 2 categorical feature indices
    'target_col_idx': [0],      # 1 categorical target index
}

# Use regression task for the main ARGS
ARGS = {
    'syn_data': syn_data,
    'info': info_regression,
    'num_inverse': num_inverse,
    'cat_inverse': cat_inverse,
}
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
timestep_embedding Ran
ran; our contract draft was wrong, not the code
jacobyhsi/TabRep/tabrep_ddpm/models/modules.py
code served (permissive licence) · get_code("fe2398f0b08da143")
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 (4, 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 math

torch.manual_seed(0)

# Create valid arguments for timestep_embedding function
timesteps = torch.tensor([0, 1, 2, 3], dtype=torch.float32)  # 1-D tensor of batch size 4
dim = 8  # output dimension
max_period = 10000  # default max_period

ARGS = {
    "timesteps": timesteps,
    "dim": dim,
    "max_period": max_period,
}
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 (4, 8) — shape (4, 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
timestep_embedding Ran
ran; our contract draft was wrong, not the code
jacobyhsi/TabRep/tabrep_flow/models/modules.py
code served (permissive licence) · get_code("6c617b25817f8107")
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 (4, 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 math
import torch

torch.manual_seed(0)

# Create valid arguments for timestep_embedding function
timesteps = torch.tensor([0, 5, 10, 15], dtype=torch.long)
dim = 8
max_period = 10000

ARGS = {
    "timesteps": timesteps,
    "dim": dim,
    "max_period": max_period,
}
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 (4, 8) — shape (4, 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
recover_data Not yet run
unverified (harvested, no recorded run)
jacobyhsi/TabRep/tabrep_ddpm/sample.py
code served (permissive licence) · get_code("0110658cf954b46a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 10 are also present in other papers whose code we hold (10 with at least one paper dated earlier, 0 the same month, 8 later) and 6 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
betas_for_alpha_bar 92 other papers
88 dated earlier, 3 later, 1 same year, order not determinable · 3 only as an assembly brick
2006.11239 (2020-06)
Denoising Diffusion Probabilistic Models
2010.02502 (2020-10)
Denoising Diffusion Implicit Models
2102.09672 (2021-02)
Improved Denoising Diffusion Probabilistic Models
and 89 more not named here
get_named_beta_schedule 63 other papers
59 dated earlier, 3 later, 1 same year, order not determinable
2010.02502 (2020-10)
Denoising Diffusion Implicit Models
2108.02938 (2021-08)
ILVR: Conditioning Method for Denoising Diffusion Probabilistic Models
2111.11672 (2021-11)
Few-shot Image Generation with Mixup-based Distance Learning
and 60 more not named here
normal_kl 34 other papers
31 dated earlier, 3 later
2006.11239 (2020-06)
Denoising Diffusion Probabilistic Models
2209.15421 (2022-09)
TabDDPM: Modelling Tabular Data with Diffusion Models
2211.13221 (2022-11)
Latent Video Diffusion Models for High-Fidelity Long Video Generation
and 31 more not named here
timestep_embedding 29 other papers
25 dated earlier, 4 later · 9 only as an assembly brick
2006.11239 (2020-06)
Denoising Diffusion Probabilistic Models
2010.11929 (2020-10)
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
2103.11161 (2021-03)
MonteFloor: Extending MCTS for Reconstructing Accurate Large-Scale Floor Plans
and 26 more not named here
approx_standard_normal_cdf 10 other papers
9 dated earlier, 1 later
2006.11239 (2020-06)
Denoising Diffusion Probabilistic Models
2209.15421 (2022-09)
TabDDPM: Modelling Tabular Data with Diffusion Models
and 7 more not named here
discretized_gaussian_log_likelihood 8 other papers
7 dated earlier, 1 later
2006.11239 (2020-06)
Denoising Diffusion Probabilistic Models
2209.15421 (2022-09)
TabDDPM: Modelling Tabular Data with Diffusion Models
and 5 more not named here
geglu 8 other papers
7 dated earlier, 1 later
2102.08921 (2021-02)
How Faithful is your Synthetic Data? Sample-level Metrics for Evaluating and Auditing Generative Models
2207.03208 (2022-07)
Revisiting Pretraining Objectives for Tabular Deep Learning
2209.15421 (2022-09)
TabDDPM: Modelling Tabular Data with Diffusion Models
and 5 more not named here
reglu 8 other papers
7 dated earlier, 1 later
2102.08921 (2021-02)
How Faithful is your Synthetic Data? Sample-level Metrics for Evaluating and Auditing Generative Models
2207.03208 (2022-07)
Revisiting Pretraining Objectives for Tabular Deep Learning
2209.15421 (2022-09)
TabDDPM: Modelling Tabular Data with Diffusion Models
and 5 more not named here
pad_t_like_x 1 other paper
1 dated earlier
2405.16577 (2024-05)
Reflected Flow Matching
recover_data 1 other paper
1 dated earlier
2410.20626 (2024-10)
TabDiff: a Mixed-type Diffusion Model for Tabular Data Generation

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

Diffusion models have been the predominant generative model for tabular data generation. However, they face the conundrum of modeling under a separate versus a unified data representation. The former encounters the challenge of jointly modeling all multi-modal distributions of tabular data in one model. While the latter alleviates this by learning a single representation for all features, it currently leverages sparse suboptimal encoding heuristics and necessitates additional computation costs. In this work, we address the latter by presenting TabRep, a tabular diffusion architecture trained with a unified continuous representation. To motivate the design of our representation, we provide geometric insights into how the data manifold affects diffusion models. The key attributes of our representation are composed of its density, flexibility to provide ample separability for nominal features, and ability to preserve intrinsic relationships. Ultimately, TabRep provides a simple yet effective approach for training tabular diffusion models under a continuous data manifold. Our results showcase that TabRep achieves superior performance across a broad suite of evaluations. It is the first to synthesize tabular data that exceeds the downstream quality of the original datasets while preserving privacy and remaining computationally efficient. Code is available at https://github.com/jacobyhsi/TabRep.

For agents

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

get_harvested_code_for_paper("2504.04798")
get_code_for_paper("2504.04798")
have("2504.04798")

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

curl https://syntology.ai/api/ran/2504.04798.json

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

[![Syntology run record](https://syntology.ai/api/ran/2504.04798.svg)](https://syntology.ai/paper/2504.04798)

Connect an agent — have() is free.