SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2306.02208 · NeurIPS · 2023

Tight Regret Bounds for Single-pass Streaming Multi-armed Bandits

Chen Wang

arXiv · PDF · Open in the Atlas

Code that ran

3 of 3 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; 2 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/2306.02208.json · badge: /api/ran/2306.02208.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
jhwjhw0123/streaming-regret-minimization-mabs canonical (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
FunctionStatusWhere it lives
best_among_remaining Ran
ran, honoured the contract we drafted
jhwjhw0123/streaming-regret-minimization-mabs/algorithms.py
code served (permissive licence) · get_code("737d6fccaf1f934b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

class Arm:
    def __init__(self, p):
        self.p = p
    
    def batch_pull(self, num_pull):
        # Simulate pulling the arm num_pull times and return empirical reward
        pulls = np.random.binomial(1, self.p, num_pull)
        return np.mean(pulls)

class ArmStream:
    def __init__(self):
        pass
    
    def regret_eval(self, p, num_pull):
        # Evaluate regret based on arm probability and number of pulls
        pass

# Create arm_stream
arm_stream = ArmStream()

# Create arm_list with 3 levels
# Each level contains a list with a single arm or empty list
arm_list = [
    [Arm(0.3)],  # Level 0 with one arm
    [Arm(0.7)],  # Level 1 with one arm
    [Arm(0.5)]   # Level 2 with one arm
]

# Number of pulls per level
num_pull = 10

ARGS = {
    'arm_stream': arm_stream,
    'arm_list': arm_list,
    'num_pull': num_pull
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
naive_eps_best_algorithm Ran
ran; our contract draft was wrong, not the code
jhwjhw0123/streaming-regret-minimization-mabs/algorithms.py
code served (permissive licence) · get_code("d511e690f9e5720b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) 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

class Arm:
    def __init__(self, p):
        self.p = p
        self.rng = np.random.RandomState(42)
    
    def batch_pull(self, batch_size):
        # Simulate pulling the arm batch_size times and return average reward
        rewards = self.rng.binomial(1, self.p, int(batch_size))
        return np.mean(rewards)

class ArmStream:
    def __init__(self, n, arm_probs):
        self.n = n
        self.arm_probs = arm_probs
        self.current_index = 0
        self.regret_counter = 0.0
        self.best_p = max(arm_probs)
    
    def read_next_arm(self):
        if self.current_index >= self.n:
            return None
        arm = Arm(self.arm_probs[self.current_index])
        self.current_index += 1
        return arm
    
    def regret_eval(self, p, num_pulls):
        # Calculate regret: difference between best arm and current arm
        regret = (self.best_p - p) * num_pulls
        self.regret_counter += regret

# Create a simple arm stream with 5 arms
n_arms = 5
arm_probabilities = [0.3, 0.5, 0.7, 0.4, 0.6]
arm_stream = ArmStream(n_arms, arm_probabilities)

T = 1000

ARGS = {
    'arm_stream': arm_stream,
    'T': T,
    'verbose': False
}
Setup block drafted by
a language model, Claude Sonnet 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 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
uniform_exploration_algorithm Ran
ran; our contract draft was wrong, not the code
jhwjhw0123/streaming-regret-minimization-mabs/algorithms.py
code served (permissive licence) · get_code("57a40d97fadbe7b3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) 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

class Arm:
    def __init__(self, p):
        self.p = p
        self.rng = np.random.RandomState(42)
    
    def batch_pull(self, batch_size):
        batch_size = int(batch_size)
        pulls = self.rng.binomial(1, self.p, batch_size)
        return np.mean(pulls)

class ArmStream:
    def __init__(self, n, arm_probs):
        self.n = n
        self.arm_probs = arm_probs
        self.current_idx = 0
        self.regret_counter = 0.0
        self.best_p = max(arm_probs)
    
    def read_next_arm(self):
        if self.current_idx >= self.n:
            return None
        arm = Arm(self.arm_probs[self.current_idx])
        self.current_idx += 1
        return arm
    
    def regret_eval(self, p, num_pulls):
        regret = (self.best_p - p) * num_pulls
        self.regret_counter += regret

arm_probs = [0.3, 0.5, 0.7, 0.4, 0.6]
arm_stream = ArmStream(n=5, arm_probs=arm_probs)
T = 100

ARGS = {
    'arm_stream': arm_stream,
    'T': T,
    'verbose': False
}
Setup block drafted by
a language model, Claude Sonnet 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 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

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 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 3 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

Regret minimization in streaming multi-armed bandits (MABs) has been studied extensively in recent years. In the single-pass setting with K arms and T trials, a regret lower bound of Ω(T 2/3 ) has been proved for any algorithm with o(K) memory (Maiti et al. [NeurIPS'21]; Agarwal at al. [COLT'22]). On the other hand, however, the previous best regret upper bound is still O K 1/3 T 2/3 log 1/3 (T ) , which is achieved by the streaming implementation of the simple uniform exploration. The O(K 1/3 log 1/3 (T )) gap leaves the open question of the tight regret bound in the single-pass MABs with sublinear arm memory. In this paper, we answer this open problem and complete the picture of regret minimization in single-pass streaming MABs. We first improve the regret lower bound to Ω(K 1/3 T 2/3 ) for algorithms with o(K) memory, which matches the uniform exploration regret up to a logarithm factor in T . We then show that the log 1/3 (T ) factor is not necessary, and we can achieve O(K 1/3 T 2/3 ) regret by finding an ε-best arm and committing to it in the rest of the trials. For regret minimization with high constant probability, we can apply the single-memory ε-best arm algorithms in Jin et al. [ICML'21] to obtain the optimal bound. Furthermore, for the expected regret minimization, we design an algorithm with a single-arm memory that achieves O(K 1/3 T 2/3 log(K)) regret, and an algorithm with O(log * (n))-memory with the optimal O(K 1/3 T 2/3 ) regret following the ε-best arm algorithm in Assadi and Wang [STOC'20]. We further tested the empirical performances of our algorithms on simulated MABs instances. The simulation results show that the proposed algorithms consistently outperform the benchmark uniform exploration algorithm by a large margin, and on occasions reduce the regret by up to 70% (i.e. 30% of the regret produced by uniform exploration).

For agents

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

get_harvested_code_for_paper("2306.02208")
get_code_for_paper("2306.02208")
have("2306.02208")

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

curl https://syntology.ai/api/ran/2306.02208.json

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

[![Syntology run record](https://syntology.ai/api/ran/2306.02208.svg)](https://syntology.ai/paper/2306.02208)

Connect an agent — have() is free.