SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2002.09403 · 2020

Inexact Tensor Methods with Dynamic Accuracies

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; 0 honoured the contract we drafted, 0 violated it and 9 have no contract verdict; 0 failures of our instrument; 0 no recorded run), measured 2026-11-21; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2002.09403.json · badge: /api/ran/2002.09403.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
doikov/dynamic-accuracies community (classifier label)
classifier basis not recorded on the link
9 ran · 0 no recorded run
FunctionStatusWhere it lives
create_log_reg_oracle Ran
ran on a synthesized input
doikov/dynamic-accuracies/oracles.py
code served (permissive licence) · get_code("ce60b31344d4a637")
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 LogRegL2Oracle. 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 scipy.sparse

# Create small feature matrix A (m samples x n features)
# Using distinct dimensions: 5 samples, 3 features
m, n = 5, 3
A = np.random.randn(m, n)

# Create binary labels b (m samples,)
b = np.array([1.0, -1.0, 1.0, -1.0, 1.0])

# Regularization coefficient
regcoef = 0.1

ARGS = {
    "A": A,
    "b": b,
    "regcoef": regcoef
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LogRegL2Oracle — a LogRegL2Oracle output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LogRegL2Oracle 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
create_log_sum_exp_oracle Ran
ran on a synthesized input
doikov/dynamic-accuracies/oracles.py
code served (permissive licence) · get_code("ad1a0bbfea706dda")
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 LogSumExpOracle. 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 scipy.sparse

# Create a small matrix A (m x n where m=5, n=3)
# The function uses A.dot(x), A.T.dot(x), and A.toarray()
m, n = 5, 3
A = np.random.randn(m, n)

# Create vector b (m x 1) to match the rows of A
b = np.random.randn(m)

# Create a scalar mu for temperature parameter
mu = 0.5

ARGS = {
    'A': A,
    'b': b,
    'mu': mu
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LogSumExpOracle — a LogSumExpOracle output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LogSumExpOracle 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
create_log_sum_exp_zero_oracle Ran
ran on a synthesized input
doikov/dynamic-accuracies/oracles.py
code served (permissive licence) · get_code("cfded9389cb1cc10")
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 LogSumExpOracle. 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 scipy.sparse

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

# Create a small matrix A (m x n) where m=5, n=4
# This will be used for matvec operations
m, n = 5, 4
A = np.random.randn(m, n)

# Create vector b (m,) to match the number of rows in A
b = np.random.randn(m)

# Create a small positive scalar mu for the temperature parameter
mu = 0.1

# Construct the arguments dictionary
ARGS = {
    'A': A,
    'b': b,
    'mu': mu
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LogSumExpOracle — a LogSumExpOracle output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LogSumExpOracle 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
cubic_newton_step Ran
ran on a synthesized input
doikov/dynamic-accuracies/newton_methods.py
code served (permissive licence) · get_code("bcca049194949363")
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 scipy
from scipy.linalg import LinAlgError

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

# Define dimension
n = 4

# Create a small positive definite matrix A
# Use a simple symmetric positive definite matrix
A = np.array([
    [4.0, 0.5, 0.2, 0.1],
    [0.5, 3.0, 0.3, 0.2],
    [0.2, 0.3, 2.0, 0.1],
    [0.1, 0.2, 0.1, 1.5]
], dtype=float)

# Create gradient vector g
g = np.array([1.0, -0.5, 0.3, -0.2], dtype=float)

# Create parameter H (controls the cubic term)
H = 0.5

# Create optional matrix B (identity for simplicity, but could be different)
B = np.eye(n, dtype=float)

# Create small epsilon for convergence
eps = 1e-9

ARGS = {
    'g': g,
    'A': A,
    'H': H,
    'B': B,
    'eps': eps
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — 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
get_labels Ran
ran on a synthesized input
doikov/dynamic-accuracies/experimental_tools.py
code served (permissive licence) · get_code("fa6292c3e4ccdb23")
How it ran

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

import torch

class Strategy:
    def __init__(self, label):
        self.label = label

strategies = [
    Strategy(label="strategy_a"),
    Strategy(label="strategy_b"),
    Strategy(label="strategy_c"),
]

ARGS = {
    "strategies": strategies,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_tolerance Ran
ran on a synthesized input
doikov/dynamic-accuracies/utils.py
code served (permissive licence) · get_code("5a1a81a9e115a3c8")
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 Tolerance. 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)

# Test case 1: None (uses default Tolerance)
tolerance_options_none = None

# Test case 2: Tolerance object (returns as-is)
tolerance_obj = type('Tolerance', (), {})()
tolerance_obj.tolerance = 1e-08
tolerance_obj.criterion = 'grad'
tolerance_obj.relative = False

# Test case 3: Dict with 'func' criterion
tolerance_dict_func = {
    'tolerance': 1e-06,
    'criterion': 'func',
    'f_star': 0.5,
    'relative': True,
    'f_0': 10.0
}

# Test case 4: Dict with 'grad' criterion
tolerance_dict_grad = {
    'tolerance': 1e-05,
    'criterion': 'grad',
    'relative': False
}

# Test case 5: Dict with 'grad_uniform_convex' criterion
tolerance_dict_grad_uc = {
    'tolerance': 1e-04,
    'criterion': 'grad_uniform_convex',
    'p': 2.5,
    'sigma': 0.8
}

# Test case 6: Dict with 'grad_norm_bound' criterion
tolerance_dict_grad_norm_bound = {
    'tolerance': 1e-03,
    'criterion': 'grad_norm_bound',
    'c': 0.01
}

# Test case 7: Dict with 'grad_norm_lambda_bound' criterion
tolerance_dict_grad_norm_lambda = {
    'tolerance': 1e-02,
    'criterion': 'grad_norm_lambda_bound',
    'c': 0.25,
    'lambda_bound': lambda x: np.sum(x * x) if isinstance(x, np.ndarray) else x.dot(x)
}

# Define ARGS - mapping parameter names to values
ARGS = {
    'tolerance_options': tolerance_dict_grad
}
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 Tolerance — a Tolerance output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Tolerance 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_tolerance_strategy Ran
ran on a synthesized input
doikov/dynamic-accuracies/utils.py
code served (permissive licence) · get_code("66bf6110477f1099")
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 ToleranceStrategy. 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

class ToleranceStrategy(object):
    """
    Choosing the inner tolerance.
    """

    def __init__(self, strategy='constant', **kwargs):
        self.strategy = strategy
        if strategy == 'constant':
            self.delta = kwargs.get('delta', 1e-08)
            self.label = kwargs.get('label', '%f' % self.delta)
        elif strategy == 'power':
            self.alpha = kwargs.get('alpha', 2)
            self.c = kwargs.get('c', 1.0)
            self.label = kwargs.get('label', '1/k^%d' % self.alpha)
        elif strategy == 'adaptive':
            self.alpha = kwargs.get('alpha', 1)
            self.c = kwargs.get('c', 1.0)
            self.label = kwargs.get('label', 'adaptive')
        else:
            raise ValueError('Unknown tolerance strategy {}'.format(strategy))

    @classmethod
    def from_dict(cls, dict):
        return cls(**dict)

    def to_dict(self):
        return self.__dict__

    def get_tolerance(self, k, f_prev, f_cur):
        if self.strategy == 'constant':
            return self.delta
        elif self.strategy == 'power':
            return self.c / (k + 1) ** self.alpha
        elif self.strategy == 'adaptive':
            if f_prev is None or f_prev <= f_cur + 1e-09:
                return 1e-09
            else:
                return self.c * (f_prev - f_cur) ** self.alpha
        else:
            raise ValueError('Unknown tolerance strategy {}'.format(self.strategy))

def get_tolerance_strategy(tolerance_strategy_options=None):
    """
    Helper to construct ToleranceStrategy.
    """
    if tolerance_strategy_options:
        if type(tolerance_strategy_options) is ToleranceStrategy:
            return tolerance_strategy_options
        else:
            return ToleranceStrategy.from_dict(tolerance_strategy_options)
    else:
        return ToleranceStrategy()

# Test case 1: None (default behavior)
ARGS_case1 = {
    'tolerance_strategy_options': None
}

# Test case 2: Dictionary with constant strategy
ARGS_case2 = {
    'tolerance_strategy_options': {
        'strategy': 'constant',
        'delta': 1e-06,
        'label': 'constant_1e-06'
    }
}

# Test case 3: Dictionary with power strategy
ARGS_case3 = {
    'tolerance_strategy_options': {
        'strategy': 'power',
        'alpha': 3,
        'c': 0.5,
        'label': '1/k^3'
    }
}

# Test case 4: Dictionary with adaptive strategy
ARGS_case4 = {
    'tolerance_strategy_options': {
        'strategy': 'adaptive',
        'alpha': 2,
        'c': 0.1,
        'label': 'adaptive_custom'
    }
}

# Test case 5: ToleranceStrategy object directly
ARGS_case5 = {
    'tolerance_strategy_options': ToleranceStrategy(strategy='constant', delta=1e-07)
}

ARGS = ARGS_case1
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 ToleranceStrategy — a ToleranceStrategy output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ToleranceStrategy 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
norms_init Ran
ran on a synthesized input
doikov/dynamic-accuracies/utils.py
code served (permissive licence) · get_code("ebf2bf5314576eb7")
How it ran

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

import numpy as np

# Construct test arguments for norms_init function

# Case 1: B and Binv both None (identity case)
ARGS_case1 = {
    'B': None,
    'Binv': None
}

# Case 2: B provided, Binv computed internally
B_matrix = np.array([
    [2.0, 0.5],
    [0.5, 1.5]
], dtype=np.float64)

ARGS_case2 = {
    'B': B_matrix,
    'Binv': None
}

# Case 3: Both B and Binv provided
Binv_matrix = np.linalg.inv(B_matrix)

ARGS_case3 = {
    'B': B_matrix,
    'Binv': Binv_matrix
}

# Use case 2 as the main test case (B provided, Binv auto-computed)
ARGS = ARGS_case2
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 tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
run_method Ran
ran on a synthesized input
doikov/dynamic-accuracies/experimental_tools.py
code served (permissive licence) · get_code("b1f26ae9f0525c55")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 2, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from datetime import datetime

torch.manual_seed(0)

# Define a simple method that matches the expected signature
def simple_method(param):
    """
    Expected to return a tuple of (result, status, history)
    param is a dict with some configuration
    """
    # Simulate some computation
    history = [1.0, 2.0, 3.0]
    status = "success"
    result = param.get("value", 0) * 2
    return result, status, history

# Define parameters - list of dicts to pass to the method
parameters = [
    {"value": 1, "name": "config_a"},
    {"value": 2, "name": "config_b"},
    {"value": 3, "name": "config_c"},
]

# Define labels for each parameter
labels = [
    "strategy_one",
    "strategy_two",
    "strategy_three",
]

# Construct ARGS dict with all required parameters
ARGS = {
    "method": simple_method,
    "parameters": parameters,
    "labels": labels,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 2, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

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 2026-11-21.

Repositories linked to this paper

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

Abstract

In this paper, we study inexact high-order Tensor Methods for solving convex optimization problems with composite objective. At every step of such methods, we use approximate solution of the auxiliary problem, defined by the bound for the residual in function value. We propose two dynamic strategies for choosing the inner accuracy: the first one is decreasing as $1/k^{p + 1}$, where $p \geq 1$ is the order of the method and $k$ is the iteration counter, and the second approach is using for the inner accuracy the last progress in the target objective. We show that inexact Tensor Methods with these strategies achieve the same global convergence rate as in the error-free case. For the second approach we also establish local superlinear rates (for $p \geq 2$), and propose the accelerated scheme. Lastly, we present computational results on a variety of machine learning problems for several methods and different accuracy policies.

For agents

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

get_harvested_code_for_paper("2002.09403")
get_code_for_paper("2002.09403")
have("2002.09403")

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

curl https://syntology.ai/api/ran/2002.09403.json

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

[![Syntology run record](https://syntology.ai/api/ran/2002.09403.svg)](https://syntology.ai/paper/2002.09403)

Connect an agent — have() is free.