SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.08555 · 2020

Pipeline PSRO: A Scalable Approach for Finding Approximate Nash Equilibria in Large Games

arXiv · PDF · Open in the Atlas

Code that ran

7 of 7 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 7 have no contract verdict; 0 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/2006.08555.json · badge: /api/ran/2006.08555.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
JBLanier/pipeline-psro canonical (classifier label)
classifier basis: census
7 ran · 0 no recorded run
FunctionStatusWhere it lives
create_reference_policy_update_callback_for_self_play_eval Ran
ran on a synthesized input
JBLanier/pipeline-psro/multiplayer-rl/mprl/rl/common/custom_eval_callbacks.py
code served (permissive licence) · get_code("52754f30e667f367")
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 function. 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 logging
from unittest.mock import Mock, MagicMock

logger = logging.getLogger(__name__)

# Create a mock trainer object with necessary methods and attributes
class MockTrainer:
    def __init__(self):
        self.config = {
            'new_policy_accept_score': 0.5
        }
        self.policy_version = 0
        self._weights = {
            'trainable_policy': {'param1': 0.1, 'param2': 0.2},
            'reference_policy': {'param1': 0.05, 'param2': 0.15}
        }
    
    def get_weights(self, policy_keys):
        result = {}
        for key in policy_keys:
            if key in self._weights:
                result[key] = self._weights[key].copy()
        return result
    
    def set_weights(self, weights_dict):
        for key, val in weights_dict.items():
            self._weights[key] = val.copy() if isinstance(val, dict) else val

# Create eval_metrics dict with nested structure
eval_metrics = {
    ('trainable_policy', 'win_percentage'): 0.7,
    'current_reference_policy_version': 0
}

# Create a custom accept callback
def accept_callback_fn(trainer):
    logger.info('Custom accept callback executed')

ARGS = {
    'trainable_policy_key': 'trainable_policy',
    'reference_policy_key': 'reference_policy',
    'new_policy_accept_score': 0.5,
    'accept_metric': ('trainable_policy', 'win_percentage'),
    'accept_mode': 'gte',
    'accept_callback': accept_callback_fn,
    'reset_trainable_policy_weights_on_failure': False,
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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_worker_process_pool_with_ids Ran
ran on a synthesized input
JBLanier/pipeline-psro/multiplayer-rl/mprl/utils.py
code served (permissive licence) · get_code("07bb49d778f2619a")
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 Pool. 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 multiprocessing as mp

def create_worker_process_pool_with_ids(num_workers):
    ids = list(range(num_workers))
    manager = mp.Manager()
    id_queue = manager.Queue()
    for i in ids:
        id_queue.put(i)
    return mp.Pool(processes=num_workers, initializer=worker_index_init, initargs=(id_queue,))

def worker_index_init(queue=None, index_val=None):
    global pool_worker_index
    if queue is not None:
        pool_worker_index = queue.get()
    else:
        pool_worker_index = index_val

ARGS = {
    "num_workers": 4
}
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 Pool — a Pool output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Pool 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_training_data_from_file Ran
ran on a synthesized input
JBLanier/pipeline-psro/multiplayer-rl/mprl/utils.py
code served (permissive licence) · get_code("da0e4c55ea1a63f8")
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 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 numpy as np
import tempfile
import os

# Create a temporary directory and a temporary .npy file
temp_dir = tempfile.mkdtemp()
temp_file_path = os.path.join(temp_dir, 'test_data.npy')

# Create sample training data and save it
sample_data = np.random.randn(2, 4).astype(np.float32)
np.save(temp_file_path, sample_data)

# Extract the name without the .npy extension
name = temp_file_path[:-4]

# Change to the temp directory so the function can find the file
original_cwd = os.getcwd()
os.chdir(temp_dir)

ARGS = {
    'name': 'test_data'
}
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 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
policy_with_dirichlet_noise Ran
ran on a synthesized input
JBLanier/pipeline-psro/multiplayer-rl/mprl/utils.py
code served (permissive licence) · get_code("49507ebdfbe4ea29")
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 float, shape (5,) 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 random seed for reproducibility
np.random.seed(42)

# Create a valid policy distribution over 5 actions
policy = np.array([0.2, 0.3, 0.2, 0.15, 0.15])

# Create a valid_actions_mask with 4 valid actions (True) and 1 invalid (False)
valid_actions_mask = np.array([True, True, False, True, True])

# Create a noise proportion between 0 and 1
noise_proportion = 0.25

ARGS = {
    "policy": policy,
    "valid_actions_mask": valid_actions_mask,
    "noise_proportion": noise_proportion,
}
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 float, shape (5,) — shape (5,) 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
pretty_print Ran
ran on a synthesized input
JBLanier/pipeline-psro/multiplayer-rl/mprl/utility_services/utils.py
code served (permissive licence) · get_code("f48bf6417161da6f")
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 json
import numbers
import numpy as np
import yaml

class _SafeFallbackEncoder(json.JSONEncoder):

    def __init__(self, nan_str='null', **kwargs):
        super(_SafeFallbackEncoder, self).__init__(**kwargs)
        self.nan_str = nan_str

    def default(self, value):
        try:
            if np.isnan(value):
                return self.nan_str
            if type(value).__module__ == np.__name__ and isinstance(value, np.ndarray):
                return value.tolist()
            if issubclass(type(value), numbers.Integral):
                return int(value)
            if issubclass(type(value), numbers.Number):
                return float(value)
            return super(_SafeFallbackEncoder, self).default(value)
        except Exception:
            return str(value)

def pretty_print(result):
    result = result.copy()
    result.update(config=None)
    out = {}
    for k, v in result.items():
        if v is not None:
            out[k] = v
    cleaned = json.dumps(out, cls=_SafeFallbackEncoder)
    return yaml.safe_dump(json.loads(cleaned), default_flow_style=False)

# Setup code to construct valid arguments
ARGS = {
    'result': {
        'accuracy': 0.95,
        'loss': 0.123,
        'epoch': 42,
        'iterations': np.int64(1000),
        'learning_rate': np.float32(0.001),
        'metrics': {
            'precision': 0.92,
            'recall': 0.89
        },
        'array_data': np.array([1.0, 2.0, 3.0]),
        'nan_value': float('nan'),
        'none_field': None,
        'description': 'test_run',
        'config': {'hidden_dim': 256, 'num_layers': 3}
    }
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, yaml — 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
seconds_to_text Ran
ran on a synthesized input
JBLanier/pipeline-psro/multiplayer-rl/mprl/utility_services/utils.py
code served (permissive licence) · get_code("42371142597f2af5")
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 torch
import numpy as np

# seconds_to_text takes a single parameter 'secs' which should be a numeric value
# representing seconds. The function converts it to int and performs arithmetic operations.
# We'll test with a value that exercises multiple time units.

ARGS = {
    'secs': 90061  # 1 day, 1 hour, 1 minute, 1 second
}
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 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
with_base_config Ran
ran on a synthesized input
JBLanier/pipeline-psro/multiplayer-rl/mprl/utility_services/utils.py
code served (permissive licence) · get_code("d7263ae5e7760caa")
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 copy

def with_base_config(base_config, extra_config):
    """Returns dict with updates applied, helpful for one-liners"""
    config = copy.deepcopy(base_config)
    config.update(extra_config)
    return config

# Setup code
base_config = {
    "learning_rate": 0.001,
    "batch_size": 32,
    "num_epochs": 10,
    "model_name": "bert",
    "hidden_size": 768
}

extra_config = {
    "learning_rate": 0.0001,
    "dropout": 0.1,
    "num_layers": 12
}

ARGS = {
    "base_config": base_config,
    "extra_config": extra_config
}
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 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

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 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 7 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

Finding approximate Nash equilibria in zero-sum imperfect-information games is challenging when the number of information states is large. Policy Space Response Oracles (PSRO) is a deep reinforcement learning algorithm grounded in game theory that is guaranteed to converge to an approximate Nash equilibrium. However, PSRO requires training a reinforcement learning policy at each iteration, making it too slow for large games. We show through counterexamples and experiments that DCH and Rectified PSRO, two existing approaches to scaling up PSRO, fail to converge even in small games. We introduce Pipeline PSRO (P2SRO), the first scalable general method for finding approximate Nash equilibria in large zero-sum imperfect-information games. P2SRO is able to parallelize PSRO with convergence guarantees by maintaining a hierarchical pipeline of reinforcement learning workers, each training against the policies generated by lower levels in the hierarchy. We show that unlike existing methods, P2SRO converges to an approximate Nash equilibrium, and does so faster as the number of parallel workers increases, across a variety of imperfect information games. We also introduce an open-source environment for Barrage Stratego, a variant of Stratego with an approximate game tree complexity of $10^{50}$. P2SRO is able to achieve state-of-the-art performance on Barrage Stratego and beats all existing bots. Experiment code is available athttps://github.com/JBLanier/pipeline-psro.

For agents

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

get_harvested_code_for_paper("2006.08555")
get_code_for_paper("2006.08555")
have("2006.08555")

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

curl https://syntology.ai/api/ran/2006.08555.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.08555.svg)](https://syntology.ai/paper/2006.08555)

Connect an agent — have() is free.