SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2105.03153 · 2021

Pairwise Fairness for Ordinal Regression

arXiv · PDF · Open in the Atlas

Code that ran

9 of 9 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, 0 violated it and 0 have no contract verdict; 8 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/2105.03153.json · badge: /api/ran/2105.03153.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
amazon-research/fair-ordinal-regression canonical (classifier label)
classifier basis: census
9 ran · 0 no recorded run
FunctionStatusWhere it lives
compute_cost Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
amazon-research/fair-ordinal-regression/learn_fair_threshold_model/evaluation.py
code served (permissive licence) · get_code("77331f760d7f305b")
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 float32, 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 numpy as np

# Create small ground-truth labels (indices into cost_matrix rows)
true = np.array([0, 1, 0, 2])

# Create small predictions (indices into cost_matrix columns)
pred = np.array([1, 1, 2, 0])

# Create a cost matrix where cost_matrix[i, j] is the cost of 
# misclassifying ground-truth label i as prediction j
# Shape: (num_classes, num_classes) = (3, 3)
cost_matrix = np.array([
    [0, 5, 10],
    [3, 0, 7],
    [8, 2, 0]
], dtype=np.float32)

ARGS = {
    "true": true,
    "pred": pred,
    "cost_matrix": cost_matrix
}
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 float32, 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
evaluate_DP_exact Ran
ran; our contract draft was wrong, not the code
amazon-research/fair-ordinal-regression/learn_fair_threshold_model/evaluation.py
code served (permissive licence) · get_code("84b7d112ccd1070e")
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 float64, ndim 0, 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 numpy as np

# Small example with 2 protected groups, 3 label classes, 8 samples
labels = np.array([0, 1, 2, 0, 1, 2, 0, 1], dtype=np.int64)
prot_attribute = np.array([0, 0, 0, 0, 1, 1, 1, 1], dtype=np.int64)
nr_groups = 2

ARGS = {
    "labels": labels,
    "prot_attribute": prot_attribute,
    "nr_groups": nr_groups,
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
evaluate_EO_exact Ran
ran; our contract draft was wrong, not the code
amazon-research/fair-ordinal-regression/learn_fair_threshold_model/evaluation.py
code served (permissive licence) · get_code("83a63654701028c4")
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 float64, ndim 0, 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 numpy as np

# Create small test data for the function
np.random.seed(0)

# Number of samples
n = 10

# Number of classes (k)
k = 3

# Number of protected groups
nr_groups = 2

# Generate true labels in {0, 1, 2}
true_labels = np.random.randint(0, k, size=n)

# Generate predicted labels in {0, 1, 2}
predicted_labels = np.random.randint(0, k, size=n)

# Generate protected attributes in {0, 1}
prot_attribute = np.random.randint(0, nr_groups, size=n)

ARGS = {
    "true_labels": true_labels,
    "predicted_labels": predicted_labels,
    "prot_attribute": prot_attribute,
    "nr_groups": nr_groups
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
find_best_left_shift Ran
ran; our contract draft was wrong, not the code
amazon-research/fair-ordinal-regression/learn_fair_threshold_model/local_search_demographic_parity.py
code served (permissive licence) · get_code("f6a02bddcd7a19b0")
How it ran

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

import numpy as np

np.random.seed(0)

# Analyze what the function needs:
# - n: int (number of items)
# - n_thresh: int (number of thresholds)
# - nr_groups: int (number of protected groups)
# - labels: array of shape (n,) with values in range [0, nr_groups)
# - pred: array (appears unused in function body, but is a parameter)
# - prot_attribute: array of shape (n,) with values in range [0, nr_groups)
# - threshold_indices: array of shape (n_thresh,) with values in range [0, n]
# - helper_table_T: 2D array of shape (n_thresh+1, nr_groups)
# - helper_table_P: 2D array of shape (nr_groups, nr_groups)
# - el_per_group: array of shape (nr_groups,) with positive values
# - cost: scalar float
# - cost_matrix: 2D array of shape (nr_groups, n_thresh+1)
# - lambd: scalar float
# - scores: array of shape (n,) with float values

n = 10
n_thresh = 4
nr_groups = 3

# Construct labels: array of integers in [0, nr_groups)
labels = np.array([0, 1, 2, 0, 1, 2, 0, 1, 2, 0])

# Construct pred (unused but required)
pred = np.random.rand(n)

# Construct prot_attribute: array of integers in [0, nr_groups)
prot_attribute = np.array([0, 1, 2, 0, 1, 2, 0, 1, 2, 0])

# Construct threshold_indices: array of shape (n_thresh,) with values in range [1, n]
# Make sure not all are 0, and some are greater than previous
threshold_indices = np.array([2, 4, 6, 8])

# Construct helper_table_T: shape (n_thresh+1, nr_groups)
helper_table_T = np.random.rand(n_thresh + 1, nr_groups)

# Construct helper_table_P: shape (nr_groups, nr_groups)
helper_table_P = np.random.rand(nr_groups, nr_groups)

# Construct el_per_group: shape (nr_groups,) with positive values
el_per_group = np.array([3.0, 3.0, 4.0])

# Construct cost: scalar
cost = 5.0

# Construct cost_matrix: shape (nr_groups, n_thresh+1)
cost_matrix = np.random.rand(nr_groups, n_thresh + 1)

# Construct lambd: scalar
lambd = 0.5

# Construct scores: array of shape (n,)
scores = np.linspace(0.0, 1.0, n)

ARGS = {
    'n': n,
    'n_thresh': n_thresh,
    'nr_groups': nr_groups,
    'labels': labels,
    'pred': pred,
    'prot_attribute': prot_attribute,
    'threshold_indices': threshold_indices,
    'helper_table_T': helper_table_T,
    'helper_table_P': helper_table_P,
    'el_per_group': el_per_group,
    'cost': cost,
    'cost_matrix': cost_matrix,
    'lambd': lambd,
    'scores': scores,
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
find_best_left_shift Ran
ran; our contract draft was wrong, not the code
amazon-research/fair-ordinal-regression/learn_fair_threshold_model/local_search_equal_opportunity.py
code served (permissive licence) · get_code("78c7c4822d5efb84")
How it ran

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

import numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Define parameters based on function signature and usage
n = 10  # number of samples
n_thresh = 5  # number of thresholds
nr_groups = 3  # number of protected attribute groups

# labels: array of class labels for each sample, used as indices into cost_matrix
# Range should be [0, num_classes)
labels = np.array([0, 1, 0, 2, 1, 0, 1, 2, 0, 1], dtype=int)

# pred: predictions (not directly used in function body, but required parameter)
pred = np.array([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95], dtype=float)

# prot_attribute: protected attribute group for each sample, indices into nr_groups
prot_attribute = np.array([0, 1, 2, 0, 1, 2, 0, 1, 2, 0], dtype=int)

# threshold_indices: indices where thresholds are placed (must have length n_thresh)
# Start with some valid indices that aren't all 0
threshold_indices = np.array([2, 4, 6, 8, 10], dtype=int)

# scores: score for each sample, used for comparison with MY_EPS tolerance
scores = np.array([0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.95], dtype=float)

# helper_table_T: shape (num_classes, n_thresh, nr_groups)
# Used as: helper_table_T[:labels[ell], thresh, :]
num_classes = 3
helper_table_T = np.random.rand(num_classes, n_thresh + 1, nr_groups) * 5

# helper_table_P: shape (nr_groups, nr_groups)
# Used for computing fairness metrics via transpose
helper_table_P = np.random.rand(nr_groups, nr_groups) * 10

# el_per_group_that_are_bigger: shape (nr_groups,)
# Used for normalization: helper_table_P / el_per_group_that_are_bigger
el_per_group_that_are_bigger = np.array([2.0, 3.0, 2.5])

# cost: scalar cost value
cost = 5.0

# cost_matrix: shape (num_classes, n_thresh + 1)
# Accessed as cost_matrix[labels[...], thresh/thresh+1]
cost_matrix = np.random.rand(num_classes, n_thresh + 1) * 10

# lambd: regularization parameter for fairness
lambd = 0.5

ARGS = {
    'n': n,
    'n_thresh': n_thresh,
    'nr_groups': nr_groups,
    'labels': labels,
    'pred': pred,
    'prot_attribute': prot_attribute,
    'threshold_indices': threshold_indices,
    'helper_table_T': helper_table_T,
    'helper_table_P': helper_table_P,
    'el_per_group_that_are_bigger': el_per_group_that_are_bigger,
    'cost': cost,
    'cost_matrix': cost_matrix,
    'lambd': lambd,
    'scores': scores,
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
find_best_right_shift Ran
ran; our contract draft was wrong, not the code
amazon-research/fair-ordinal-regression/learn_fair_threshold_model/local_search_demographic_parity.py
code served (permissive licence) · get_code("527f5c42a18a3689")
How it ran

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

import numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Parameters based on function signature and usage
n = 10  # number of elements
n_thresh = 3  # number of thresholds
nr_groups = 2  # number of protected attribute groups
num_classes = 4  # for cost_matrix rows

# labels: array of class labels for each element, indexed by threshold_indices
labels = np.array([0, 1, 2, 3, 1, 0, 2, 1, 3, 2])

# pred: predictions (used indirectly through indexing)
pred = np.array([0, 1, 1, 3, 1, 0, 2, 1, 3, 2])

# prot_attribute: protected attribute values (group membership for each element)
# 1-indexed access, so we need n elements
prot_attribute = np.array([0, 1, 0, 1, 0, 1, 0, 1, 0, 1])

# threshold_indices: indices of thresholds (must be sorted and <= n)
threshold_indices = np.array([2, 5, 10])

# helper_table_T: confusion matrix for thresholds
# shape: (n_thresh, nr_groups) - tracks transitions between thresholds
helper_table_T = np.array([[3, 2], [2, 3], [2, 3]], dtype=float)

# helper_table_P: parity matrix
# shape: (nr_groups, nr_groups) - tracks protected attribute distribution
helper_table_P = np.array([[5, 5], [5, 5]], dtype=float)

# el_per_group: number of elements per group
el_per_group = np.array([5.0, 5.0])

# cost: scalar cost value
cost = 2.5

# cost_matrix: cost for each (class, threshold) combination
# shape: (num_classes, n_thresh + 1)
cost_matrix = np.array([
    [0.1, 0.2, 0.3, 0.4],
    [0.15, 0.25, 0.35, 0.45],
    [0.12, 0.22, 0.32, 0.42],
    [0.18, 0.28, 0.38, 0.48]
], dtype=float)

# lambd: regularization parameter
lambd = 0.5

# scores: scores for ranking/ordering elements
scores = np.array([0.1, 0.2, 0.25, 0.3, 0.35, 0.4, 0.5, 0.6, 0.7, 0.8])

ARGS = {
    'n': n,
    'n_thresh': n_thresh,
    'nr_groups': nr_groups,
    'labels': labels,
    'pred': pred,
    'prot_attribute': prot_attribute,
    'threshold_indices': threshold_indices,
    'helper_table_T': helper_table_T,
    'helper_table_P': helper_table_P,
    'el_per_group': el_per_group,
    'cost': cost,
    'cost_matrix': cost_matrix,
    'lambd': lambd,
    'scores': scores
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
find_best_right_shift Ran
ran; our contract draft was wrong, not the code
amazon-research/fair-ordinal-regression/learn_fair_threshold_model/local_search_equal_opportunity.py
code served (permissive licence) · get_code("3c276f90716a64a3")
How it ran

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

import numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Define parameters based on function usage
n = 10  # total number of elements
n_thresh = 3  # number of thresholds
nr_groups = 2  # number of protected attribute groups

# labels: array of class labels for each element
labels = np.array([0, 1, 0, 1, 0, 1, 0, 1, 0, 1])

# pred: predictions (used indirectly through other parameters)
pred = np.array([0.1, 0.9, 0.2, 0.8, 0.3, 0.7, 0.4, 0.6, 0.5, 0.55])

# prot_attribute: protected attribute group for each element (0 or 1)
prot_attribute = np.array([0, 1, 0, 1, 0, 1, 0, 1, 0, 1])

# threshold_indices: indices where thresholds are applied
threshold_indices = np.array([2, 5, 8])

# helper_table_T: 3D table tracking label/threshold/protected_attribute combinations
# Shape: (n_classes=2, n_thresholds+1=4, nr_groups=2)
helper_table_T = np.random.randint(0, 5, size=(2, 4, 2))

# helper_table_P: 2D table for protected attribute comparisons
# Shape: (nr_groups=2, nr_groups=2)
helper_table_P = np.random.randint(0, 10, size=(2, 2))

# el_per_group_that_are_bigger: elements per group that are bigger than threshold
# Shape: (nr_groups=2,)
el_per_group_that_are_bigger = np.array([5, 5], dtype=float)

# cost: current misclassification cost
cost = 10.0

# cost_matrix: cost matrix for labels and thresholds
# Shape: (n_classes=2, n_thresholds+1=4)
cost_matrix = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [2.0, 1.0, 2.0, 3.0]
])

# lambd: fairness penalty parameter
lambd = 0.5

# scores: prediction scores for each element
scores = np.array([0.1, 0.9, 0.2, 0.8, 0.3, 0.7, 0.4, 0.6, 0.5, 0.55])

ARGS = {
    'n': n,
    'n_thresh': n_thresh,
    'nr_groups': nr_groups,
    'labels': labels,
    'pred': pred,
    'prot_attribute': prot_attribute,
    'threshold_indices': threshold_indices,
    'helper_table_T': helper_table_T,
    'helper_table_P': helper_table_P,
    'el_per_group_that_are_bigger': el_per_group_that_are_bigger,
    'cost': cost,
    'cost_matrix': cost_matrix,
    'lambd': lambd,
    'scores': scores
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_best_constant_predictor_for_MAE Ran
ran, honoured the contract we drafted
amazon-research/fair-ordinal-regression/plot_utils.py
code served (permissive licence) · get_code("bac89d70dd9a10fe")
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 int64, 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 numpy as np

# Create sample training labels - the function computes median and rounds
train_labels = np.array([1.2, 2.5, 3.7, 4.1, 5.3, 6.8, 7.2, 8.9])

ARGS = {
    "train_labels": train_labels
}
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 int64, 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
make_predictions_with_model Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
amazon-research/fair-ordinal-regression/learn_fair_threshold_model/learn_fair_model.py
code served (permissive licence) · get_code("5f8b1816e0edbfab")
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 (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 numpy as np

# Set seed for reproducibility
np.random.seed(0)

# Create X_test: shape (2, 4) - 2 samples, 4 features
X_test = np.random.randn(2, 4).astype(np.float32)

# Create w: shape (4,) - weight vector matching feature dimension
w = np.random.randn(4).astype(np.float32)

# Create thresholds: sorted array of threshold values for searchsorted
# searchsorted requires sorted thresholds
thresholds = np.array([-1.0, 0.0, 1.0, 2.0], dtype=np.float32)

ARGS = {
    "X_test": X_test,
    "w": w,
    "thresholds": thresholds,
}
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 (2,) — shape (2,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 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 9 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 initiate the study of fairness for ordinal regression. We adapt two fairness notions previously considered in fair ranking and propose a strategy for training a predictor that is approximately fair according to either notion. Our predictor has the form of a threshold model, composed of a scoring function and a set of thresholds, and our strategy is based on a reduction to fair binary classification for learning the scoring function and local search for choosing the thresholds. We provide generalization guarantees on the error and fairness violation of our predictor, and we illustrate the effectiveness of our approach in extensive experiments.

For agents

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

get_harvested_code_for_paper("2105.03153")
get_code_for_paper("2105.03153")
have("2105.03153")

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

curl https://syntology.ai/api/ran/2105.03153.json

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

[![Syntology run record](https://syntology.ai/api/ran/2105.03153.svg)](https://syntology.ai/paper/2105.03153)

Connect an agent — have() is free.