SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.01685 · 2024

SMUTF: Schema Matching Using Generative Tags and Hybrid Features

arXiv · PDF · Open in the Atlas

Code that ran

10 of 10 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; 1 honoured the contract we drafted, 2 violated it and 0 have no contract verdict; 7 failures of our instrument; 0 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/2402.01685.json · badge: /api/ran/2402.01685.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
fireindark707/python-schema-matching canonical (classifier label)
classifier basis: census
10 ran · 0 no recorded run
FunctionStatusWhere it lives
create_similarity_matrix Ran
ran; our contract draft was wrong, not the code
fireindark707/python-schema-matching/cal_column_similarity.py
code served (permissive licence) · get_code("454cca7259d54965")
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 numpy as np
import pandas as pd

# Create small dataframes for table1 and table2
table1_df = pd.DataFrame(
    np.random.rand(2, 3),
    columns=['col1', 'col2', 'col3']
)

table2_df = pd.DataFrame(
    np.random.rand(2, 4),
    columns=['col_a', 'col_b', 'col_c', 'col_d']
)

# Create predictions: need to be reshaped to (len(df1_cols), len(df2_cols))
# df1_cols = 3, df2_cols = 4, so total = 12 predictions
preds = [
    np.random.rand(12),
    np.random.rand(12)
]

# Create prediction labels: same shape as preds
pred_labels_list = [
    np.random.rand(12),
    np.random.rand(12)
]

ARGS = {
    'table1_df': table1_df,
    'table2_df': table2_df,
    'preds': preds,
    'pred_labels_list': pred_labels_list,
    'strategy': 'many-to-many'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — 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
find_all_keys_values Ran
ran; our contract draft was wrong, not the code
fireindark707/python-schema-matching/utils.py
code served (permissive licence) · get_code("d0a1767b4e776e23")
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 defaultdict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from collections import defaultdict

# Construct a valid json_data argument with nested dicts, lists, and scalar values
json_data = {
    "user": {
        "name": "Alice",
        "age": 30,
        "email": "alice@example.com"
    },
    "posts": [
        {
            "id": 1,
            "title": "First Post",
            "likes": 42
        },
        {
            "id": 2,
            "title": "Second Post",
            "likes": 18
        }
    ],
    "tags": ["python", "data", "science"],
    "metadata": {
        "created": "2023-01-15",
        "updated": "2023-02-20",
        "stats": {
            "views": 1000,
            "shares": 50
        }
    }
}

# The parent_key for the top level
parent_key = ""

ARGS = {
    "json_data": json_data,
    "parent_key": parent_key
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type defaultdict — a defaultdict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a defaultdict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_table Ran
ran; our contract draft was wrong, not the code
fireindark707/python-schema-matching/self_features.py
code served (permissive licence) · get_code("411731e041b8ad68")
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 DataFrame. 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 pandas as pd
import tempfile
import os

# Create a temporary CSV file
temp_dir = tempfile.mkdtemp()
temp_filepath = os.path.join(temp_dir, "test_data.csv")

# Write a simple CSV file
data = {
    'col1': [1, 2, 3],
    'col2': [4, 5, 6],
    'col3': ['a', 'b', 'c']
}
df = pd.DataFrame(data)
df.to_csv(temp_filepath, index=False)

ARGS = {
    'filepath': temp_filepath
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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
mainly_numeric Ran
ran, violated the contract we drafted
fireindark707/python-schema-matching/self_features.py
code served (permissive licence) · get_code("6cc49f29f989d839")
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 bool, ndim 0, dtype family float, 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 re

unit_dict = {'万': 10000, '亿': 100000000, '萬': 10000, '億': 100000000, 'K+': 1000, 'M+': 1000000, 'B+': 1000000000}

def mainly_numeric(data_list):
    """
    Checks if the given data list is mostly numeric.
    """
    cnt = 0
    for data in data_list:
        data = str(data)
        data = data.replace(',', '')
        for unit in unit_dict.keys():
            data = data.replace(unit, '')
        numeric_part = re.findall('\\d+', data)
        if len(numeric_part) > 0 and sum((len(x) for x in numeric_part)) >= 0.5 * len(data):
            cnt += 1
    if cnt >= 0.9 * len(data_list):
        return True
    return False


# Setup test data
data_list_numeric = [
    "100",
    "1,000",
    "50万",
    "123",
    "456K+",
    "789",
    "999",
    "2000",
    "3,500",
    "100M+"
]

ARGS = {
    'data_list': data_list_numeric
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, 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
make_csv_from_json Ran
ran, honoured the contract we drafted
fireindark707/python-schema-matching/utils.py
code served (permissive licence) · get_code("e8961ea558445614")
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 DataFrame. 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 pandas as pd
import json
import tempfile
import os

# Create a temporary JSON file for testing
temp_dir = tempfile.mkdtemp()
json_file_path = os.path.join(temp_dir, 'test_data.json')

# Create sample JSON data
sample_data = {
    'user1': {
        'name': 'Alice',
        'age': 30,
        'scores': [85, 90, 88]
    },
    'user2': {
        'name': 'Bob',
        'age': 25,
        'scores': [92, 88]
    }
}

# Write the JSON file
with open(json_file_path, 'w', encoding='utf-8') as f:
    json.dump(sample_data, f)

ARGS = {
    'file_path': json_file_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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
preprocess_text Ran
ran; our contract draft was wrong, not the code
fireindark707/python-schema-matching/relation_features.py
code served (permissive licence) · get_code("c30d94617013b74c")
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 str. 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 re

def preprocess_text(text):
    text = text.lower()
    text = re.split('[\\s\\_\\.]', text)
    text = ' '.join(text).strip()
    return text

ARGS = {
    "text": "Hello World_This.Is A Test"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
read_mapping Ran
ran; our contract draft was wrong, not the code
fireindark707/python-schema-matching/relation_features.py
code served (permissive licence) · get_code("77835b36787efed5")
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 set. 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 os
import tempfile

# Create a temporary mapping file for testing
temp_dir = tempfile.mkdtemp()
mapping_file = os.path.join(temp_dir, "mapping.txt")

# Write sample mapping data to the file
with open(mapping_file, 'w') as f:
    f.write("<key1>, <value1>\n")
    f.write("<key2>, <value2>\n")
    f.write("key3, value3\n")

ARGS = {
    "mapping_file": mapping_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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
schema_matching Ran
ran; our contract draft was wrong, not the code
fireindark707/python-schema-matching/cal_column_similarity.py
code served (permissive licence) · get_code("501865ce0c7fa3b7")
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 numpy as np
import pandas as pd
import xgboost as xgb
import tempfile
import os
from pathlib import Path

# Create temporary directories and files for the model
temp_dir = tempfile.mkdtemp()
model_dir = os.path.join(temp_dir, 'model', '2022-04-12-12-06-32')
os.makedirs(model_dir, exist_ok=True)

# Create minimal XGBoost models and thresholds
for i in range(2):
    # Create a minimal booster
    dtrain = xgb.DMatrix(np.random.rand(10, 5), label=np.random.rand(10))
    params = {'max_depth': 2, 'eta': 0.1, 'objective': 'reg:squarederror'}
    bst = xgb.train(params, dtrain, num_boost_round=1)
    bst.save_model(os.path.join(model_dir, f'{i}.model'))
    
    # Create threshold file
    with open(os.path.join(model_dir, f'{i}.threshold'), 'w') as f:
        f.write('0.5')

# Create sample CSV files
temp_csv1 = os.path.join(temp_dir, 'table1.csv')
temp_csv2 = os.path.join(temp_dir, 'table2.csv')

df1 = pd.DataFrame({
    'col_a': [1, 2, 3],
    'col_b': [4, 5, 6],
    'col_c': [7, 8, 9]
})
df1.to_csv(temp_csv1, index=False)

df2 = pd.DataFrame({
    'col_x': [10, 11, 12],
    'col_y': [13, 14, 15]
})
df2.to_csv(temp_csv2, index=False)

# Mock the helper functions that are called
def make_csv_from_json(path):
    return pd.DataFrame({'col': [1, 2, 3]})

def table_column_filter(df):
    return df

def make_data_from(table1_df, table2_df, type='test'):
    n_cols1 = len(table1_df.columns)
    n_cols2 = len(table2_df.columns)
    n_pairs = n_cols1 * n_cols2
    features = np.random.rand(n_pairs, 5)
    return features, None

def test(bst, threshold, features, test_labels=None, type='inference'):
    n_samples = len(features)
    pred = np.random.rand(n_samples)
    pred_labels = np.where(pred > threshold, 1, 0)
    return pred, pred_labels

# Inject mock functions into the module namespace for schema_matching to use
import sys
current_module = sys.modules[__name__]
current_module.make_csv_from_json = make_csv_from_json
current_module.table_column_filter = table_column_filter
current_module.make_data_from = make_data_from
current_module.test = test

ARGS = {
    'table1_pth': temp_csv1,
    'table2_pth': temp_csv2,
    'threshold': None,
    'strategy': 'many-to-many',
    'model_pth': model_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, xgboost — 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
strict_numeric Ran
ran, violated the contract we drafted
fireindark707/python-schema-matching/self_features.py
code served (permissive licence) · get_code("6841a8dff9b5f051")
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 bool, ndim 0, dtype family float, 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 numpy as np

torch.manual_seed(0)

# The function iterates through data_list and tries to convert each element to float
# It checks if at least 95% of elements are successfully convertible to float
# So we need a list of items, most of which can be converted to float

data_list = [
    "1.5",
    "2.3",
    "4.7",
    "10",
    "-5.2",
    "0.0",
    "100",
    "3.14",
    "not_a_number",  # This one will fail conversion
    "also_not_numeric"  # This one will also fail
]

# This list has 8 numeric out of 10 items = 80%, which is < 95%
# Let's create one that passes the 95% threshold
data_list_passing = [
    "1.5",
    "2.3",
    "4.7",
    "10",
    "-5.2",
    "0.0",
    "100",
    "3.14",
    "999",
    "42",
    "invalid"  # 1 out of 11 = ~91% fail, 90.9% pass > 95%? No, 10/11 = 90.9%
]

# Let me recalculate: need at least 95% to pass
# So if we have 20 items, we need at least 19 to be numeric (95% of 20 = 19)
data_list_passing = [str(float(i)) for i in range(19)] + ["not_numeric"]

verbose = False

ARGS = {
    "data_list": data_list_passing,
    "verbose": verbose
}
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 bool, ndim 0, dtype family float, 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
table_column_filter Ran
ran; our contract draft was wrong, not the code
fireindark707/python-schema-matching/utils.py
code served (permissive licence) · get_code("1cafb7ab99c06935")
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 DataFrame. 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 pandas as pd
import numpy as np

# Create a sample DataFrame with various column types
# Some columns will be filtered out, others will remain
data = {
    'Name': ['Alice', 'Bob', 'Charlie', 'David'],
    'Age': [25, 30, np.nan, 35],
    'City': ['--', '--', '--', '--'],  # All dashes - should be removed
    'Score': [85, np.nan, np.nan, np.nan],  # Only one valid entry - should be removed
    'Status': ['Active', 'Inactive', 'Active', 'Pending'],
    'Unnamed: 5': [1, 2, 3, 4],  # Unnamed column - should be removed
    'Department': ['HR', 'IT', 'Finance', 'IT'],
}

table_df = pd.DataFrame(data)

ARGS = {
    'table_df': table_df
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 10 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.

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

We introduce SMUTF (Schema Matching Using Generative Tags and Hybrid Features), a unique approach for large-scale tabular data schema matching (SM), which assumes that supervised learning does not affect performance in open-domain tasks, thereby enabling effective cross-domain matching. This system uniquely combines rule-based feature engineering, pre-trained language models, and generative large language models. In an innovative adaptation inspired by the Humanitarian Exchange Language, we deploy "generative tags" for each data column, enhancing the effectiveness of SM. SMUTF exhibits extensive versatility, working seamlessly with any pre-existing pre-trained embeddings, classification methods, and generative models. Recognizing the lack of extensive, publicly available datasets for SM, we have created and open-sourced the HDXSM dataset from the public humanitarian data. We believe this to be the most exhaustive SM dataset currently available. In evaluations across various public datasets and the novel HDXSM dataset, SMUTF demonstrated exceptional performance, surpassing existing state-of-the-art models in terms of accuracy and efficiency, and improving the F1 score by 11.84% and the AUC of ROC by 5.08%. Code is available at https://github.com/fireindark707/Python-Schema-Matching.

For agents

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

get_harvested_code_for_paper("2402.01685")
get_code_for_paper("2402.01685")
have("2402.01685")

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

curl https://syntology.ai/api/ran/2402.01685.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.01685.svg)](https://syntology.ai/paper/2402.01685)

Connect an agent — have() is free.