SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2110.01548 · NeurIPS · 2021

Uncertainty-Based Offline Reinforcement Learning with Diversified Q-Ensemble

Jang-Hyun Kim, Hyun Song, Gaon An, Seungyong Moon

arXiv · PDF · Open in the Atlas

Code that ran

13 of 21 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (11 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 12 have no contract verdict; 1 failure of our instrument; 8 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/2110.01548.json · badge: /api/ran/2110.01548.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
zzmtsvv/rl_task community
Papers with Code archive
5 ran · 1 no recorded run
corl-team/CORL community
Papers with Code archive
3 ran · 2 no recorded run
snu-mllab/edac community
Papers with Code archive
2 ran · 4 no recorded run
howuhh/sac-n-jax community
Papers with Code archive
2 ran · 1 no recorded run
yihaosun1124/OfflineRL-Kit community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
Actor Ran
ran on a synthesized input · constructed an object; no result was computed
corl-team/CORL/algorithms/offline/edac.py
code served (permissive licence) · get_code("7f11d5ce17556f04")
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 an instance of its own class, Actor: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
import numpy as np

torch.manual_seed(0)

ARGS = {
    'state_dim': 4,
    'action_dim': 3,
    'hidden_dim': 8,
    'max_action': 1.0,
}
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 Actor — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Actor 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
DeterministicActor Ran
ran on a synthesized input · constructed an object; no result was computed
zzmtsvv/rl_task/doge/doge.py
pointer only (licence: NONE) · get_code("241911ed91897454")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, zzmtsvv/rl_task when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Distance Ran
ran on a synthesized input · constructed an object; no result was computed
zzmtsvv/rl_task/doge/doge.py
pointer only (licence: NONE) · get_code("ab927c7eea834816")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, zzmtsvv/rl_task when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

EDACPolicy Ran
ran on a synthesized input · constructed an object; no result was computed
yihaosun1124/OfflineRL-Kit/offlinerlkit/policy/model_free/edac.py
code served (permissive licence) · get_code("f5dc603591d69cbd")
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 an instance of its own class, EDACPolicy: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
import numpy as np
from copy import deepcopy

torch.manual_seed(0)

# Create a minimal actor network that outputs a distribution
class SimpleActor(nn.Module):
    def __init__(self, obs_dim=4, action_dim=3):
        super().__init__()
        self.fc = nn.Linear(obs_dim, action_dim * 2)
    
    def forward(self, obs):
        out = self.fc(obs)
        # Return a simple distribution-like object
        return SimpleDistribution(out)

class SimpleDistribution:
    def __init__(self, params):
        self.params = params
        self.mean = params[..., :params.shape[-1]//2]
        self.logstd = params[..., params.shape[-1]//2:]
    
    def mode(self):
        squashed = torch.tanh(self.mean)
        return squashed, self.mean
    
    def rsample(self):
        noise = torch.randn_like(self.mean)
        raw_action = self.mean + torch.exp(self.logstd) * noise
        squashed = torch.tanh(raw_action)
        return squashed, raw_action
    
    def log_prob(self, squashed_action, raw_action):
        # Simple log prob calculation
        return torch.sum(-0.5 * raw_action ** 2, dim=-1, keepdim=True)

# Create a minimal ensemble critics network
class SimpleCritics(nn.Module):
    def __init__(self, obs_dim=4, action_dim=3, num_critics=2):
        super().__init__()
        self._num_ensemble = num_critics
        self.critics = nn.ModuleList([
            nn.Sequential(
                nn.Linear(obs_dim + action_dim, 8),
                nn.ReLU(),
                nn.Linear(8, 1)
            )
            for _ in range(num_critics)
        ])
    
    def forward(self, obs, actions):
        # obs: (batch, obs_dim), actions: (batch, action_dim)
        x = torch.cat([obs, actions], dim=-1)
        qs = torch.stack([critic(x) for critic in self.critics], dim=0)
        return qs

# Create minimal actor and critics
obs_dim = 4
action_dim = 3
num_critics = 2

actor = SimpleActor(obs_dim=obs_dim, action_dim=action_dim)
critics = SimpleCritics(obs_dim=obs_dim, action_dim=action_dim, num_critics=num_critics)

# Create optimizers
actor_optim = torch.optim.Adam(actor.parameters(), lr=0.001)
critics_optim = torch.optim.Adam(critics.parameters(), lr=0.001)

# Create auto-alpha tuple (for learnable alpha)
target_entropy = torch.tensor(-action_dim, dtype=torch.float32)
log_alpha = torch.tensor(0.0, requires_grad=True)
alpha_optim = torch.optim.Adam([log_alpha], lr=0.001)

ARGS = {
    'actor': actor,
    'critics': critics,
    'actor_optim': actor_optim,
    'critics_optim': critics_optim,
    'tau': 0.005,
    'gamma': 0.99,
    'alpha': (target_entropy, log_alpha, alpha_optim),
    'max_q_backup': False,
    'deterministic_backup': True,
    'eta': 1.0,
}
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 EDACPolicy — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a EDACPolicy 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
EnsembleCritic Ran
ran on a synthesized input · constructed an object; no result was computed
howuhh/sac-n-jax/sac_n_jax_flax.py
code served (permissive licence) · get_code("b4a9de39db5c3e24")
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 an instance of its own class, EnsembleCritic: the run constructed the object and did not call it, so no result was computed. 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 jax
import jax.numpy as jnp
import flax.linen as nn

def uniform_init(bound: float):
    def _init(key, shape, dtype):
        return jax.random.uniform(key, shape=shape, minval=-bound, maxval=bound, dtype=dtype)
    return _init

class Critic(nn.Module):
    hidden_dim: int = 256

    @nn.compact
    def __call__(self, state, action):
        network = nn.Sequential([nn.Dense(self.hidden_dim, kernel_init=nn.initializers.he_uniform(), bias_init=nn.initializers.constant(0.1)), nn.relu, nn.Dense(self.hidden_dim, kernel_init=nn.initializers.he_uniform(), bias_init=nn.initializers.constant(0.1)), nn.relu, nn.Dense(self.hidden_dim, kernel_init=nn.initializers.he_uniform(), bias_init=nn.initializers.constant(0.1)), nn.relu, nn.Dense(1, kernel_init=uniform_init(0.003), bias_init=uniform_init(0.003))])
        state_action = jnp.hstack([state, action])
        out = network(state_action).squeeze(-1)
        return out

class EnsembleCritic(nn.Module):
    hidden_dim: int = 256
    num_critics: int = 10

    @nn.compact
    def __call__(self, state, action):
        ensemble = nn.vmap(target=Critic, in_axes=None, out_axes=0, variable_axes={'params': 0}, split_rngs={'params': True}, axis_size=self.num_critics)
        q_values = ensemble(self.hidden_dim)(state, action)
        return q_values

state = jnp.ones((2, 4))
action = jnp.ones((2, 3))

ARGS = {
    'hidden_dim': 256,
    'num_critics': 10,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
flax, jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EnsembleCritic — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a EnsembleCritic 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
EnsembledCritic Ran
ran on a synthesized input · constructed an object; no result was computed
zzmtsvv/rl_task/doge/doge.py
pointer only (licence: NONE) · get_code("31807722c6d0ebca")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, zzmtsvv/rl_task when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

EnsembledLinear Ran
ran on a synthesized input · constructed an object; no result was computed
zzmtsvv/rl_task/doge/doge.py
pointer only (licence: NONE) · get_code("5e8059b5878ab4e3")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, zzmtsvv/rl_task when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ParallelizedEnsemble Ran
ran on a synthesized input
snu-mllab/edac/lifelong_rl/models/networks.py
code served (permissive licence) · get_code("97f6ff89661513fb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, 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
import torch
from torch import nn as nn
from torch.nn import functional as F
import types as _types

device = None

def from_numpy(*args, **kwargs):
    return torch.from_numpy(*args, **kwargs).float().to(device)

def identity(x):
    return x

def ones(*sizes, torch_device=None, **kwargs):
    if torch_device is None:
        torch_device = device
    return torch.ones(*sizes, **kwargs, device=torch_device)

def randn(*args, torch_device=None, **kwargs):
    if torch_device is None:
        torch_device = device
    return torch.randn(*args, **kwargs, device=torch_device)

def zeros(*sizes, torch_device=None, **kwargs):
    if torch_device is None:
        torch_device = device
    return torch.zeros(*sizes, **kwargs, device=torch_device)

ptu = _types.SimpleNamespace(
    device=device,
    from_numpy=from_numpy,
    randn=randn,
    zeros=zeros,
    ones=ones,
    identity=identity
)

class ParallelizedLayer(nn.Module):

    def __init__(self, ensemble_size, input_dim, output_dim, w_std_value=1.0, b_init_value=0.0):
        super().__init__()
        w_init = ptu.randn((ensemble_size, input_dim, output_dim))
        w_init = torch.fmod(w_init, 2) * w_std_value
        self.W = nn.Parameter(w_init, requires_grad=True)
        b_init = ptu.zeros((ensemble_size, 1, output_dim)).float()
        b_init += b_init_value
        self.b = nn.Parameter(b_init, requires_grad=True)

    def forward(self, x):
        return x @ self.W + self.b

class ParallelizedEnsemble(nn.Module):

    def __init__(self, ensemble_size, hidden_sizes, input_size, output_size, init_w=0.003, hidden_activation=F.relu, output_activation=identity, b_init_value=0.0, layer_norm=False, layer_norm_kwargs=None, spectral_norm=False):
        super().__init__()
        self.ensemble_size = ensemble_size
        self.input_size = input_size
        self.output_size = output_size
        self.elites = [i for i in range(self.ensemble_size)]
        self.hidden_activation = hidden_activation
        self.output_activation = output_activation
        self.input_mu = nn.Parameter(ptu.zeros(input_size), requires_grad=False).float()
        self.input_std = nn.Parameter(ptu.ones(input_size), requires_grad=False).float()
        self.fcs = []
        in_size = input_size
        for i, next_size in enumerate(hidden_sizes):
            layer_size = (ensemble_size, in_size, next_size)
            fc = ParallelizedLayer(ensemble_size, in_size, next_size, w_std_value=1 / (2 * np.sqrt(in_size)), b_init_value=b_init_value)
            if spectral_norm:
                fc = nn.utils.spectral_norm(fc, name='W')
            self.__setattr__('fc%d' % i, fc)
            self.fcs.append(fc)
            in_size = next_size
        self.last_fc = ParallelizedLayer(ensemble_size, in_size, output_size, w_std_value=1 / (2 * np.sqrt(in_size)), b_init_value=b_init_value)

    def forward(self, input):
        dim = len(input.shape)
        h = (input - self.input_mu) / self.input_std
        if dim < 3:
            h = h.unsqueeze(0)
            if dim == 1:
                h = h.unsqueeze(0)
            h = h.repeat(self.ensemble_size, 1, 1)
        for i, fc in enumerate(self.fcs):
            h = fc(h)
            h = self.hidden_activation(h)
        preactivation = self.last_fc(h)
        output = self.output_activation(preactivation)
        if dim == 1:
            output = output.squeeze(1)
        return output

    def sample(self, input):
        preds = self.forward(input)
        inds = torch.randint(0, len(self.elites), input.shape[:-1])
        inds = inds.unsqueeze(dim=-1).to(device=ptu.device)
        inds = inds.repeat(1, preds.shape[2])
        samples = (inds == 0).float() * preds[self.elites[0]]
        for i in range(1, len(self.elites)):
            samples += (inds == i).float() * preds[self.elites[i]]
        return samples

    def fit_input_stats(self, data, mask=None):
        mean = np.mean(data, axis=0, keepdims=True)
        std = np.std(data, axis=0, keepdims=True)
        std[std < 1e-12] = 1.0
        if mask is not None:
            mean *= mask
            std *= mask
        self.input_mu.data = ptu.from_numpy(mean)
        self.input_std.data = ptu.from_numpy(std)


INIT_ARGS = {
    'ensemble_size': 3,
    'hidden_sizes': [8, 6],
    'input_size': 4,
    'output_size': 2,
    'init_w': 0.003,
    'hidden_activation': F.relu,
    'output_activation': identity,
    'b_init_value': 0.0,
    'layer_norm': False,
    'layer_norm_kwargs': None,
    'spectral_norm': False,
}

ARGS = {
    'input': torch.randn(2, 4),
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ParallelizedLayer Ran
ran on a synthesized input · constructed an object; no result was computed
snu-mllab/edac/lifelong_rl/models/networks.py
code served (permissive licence) · get_code("765f2c4c8cb57eab")
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 an instance of its own class, ParallelizedLayer: the run constructed the object and did not call it, so no result was computed. 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
from torch import nn as nn
import types as _types

torch.manual_seed(0)

# Set up the device module functions
device = None

def randn(*args, torch_device=None, **kwargs):
    if torch_device is None:
        torch_device = device
    return torch.randn(*args, **kwargs, device=torch_device)

def zeros(*sizes, torch_device=None, **kwargs):
    if torch_device is None:
        torch_device = device
    return torch.zeros(*sizes, **kwargs, device=torch_device)

# Create the ptu namespace with the necessary functions
ptu = _types.SimpleNamespace(
    device=device,
    randn=randn,
    zeros=zeros
)

# Define ARGS with valid parameters for ParallelizedLayer
ARGS = {
    'ensemble_size': 2,
    'input_dim': 4,
    'output_dim': 3,
    'w_std_value': 1.0,
    'b_init_value': 0.0
}
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 ParallelizedLayer — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ParallelizedLayer 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
VectorizedCritic Ran
ran on a synthesized input · constructed an object; no result was computed
corl-team/CORL/algorithms/offline/edac.py
code served (permissive licence) · get_code("832ae1f434765b20")
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 an instance of its own class, VectorizedCritic: the run constructed the object and did not call it, so no result was computed. 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 torch.nn as nn
import math

torch.manual_seed(0)

state_dim = 4
action_dim = 3
hidden_dim = 5
num_critics = 2

ARGS = {
    "state_dim": state_dim,
    "action_dim": action_dim,
    "hidden_dim": hidden_dim,
    "num_critics": num_critics,
}
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 VectorizedCritic — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a VectorizedCritic 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
VectorizedLinear Ran
ran on a synthesized input · constructed an object; no result was computed
corl-team/CORL/algorithms/offline/edac.py
code served (permissive licence) · get_code("0f7fd096ae2490f9")
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 an instance of its own class, VectorizedLinear: the run constructed the object and did not call it, so no result was computed. 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 math
import torch
import torch.nn as nn

torch.manual_seed(0)

class VectorizedLinear(nn.Module):

    def __init__(self, in_features: int, out_features: int, ensemble_size: int):
        super().__init__()
        self.in_features = in_features
        self.out_features = out_features
        self.ensemble_size = ensemble_size
        self.weight = nn.Parameter(torch.empty(ensemble_size, in_features, out_features))
        self.bias = nn.Parameter(torch.empty(ensemble_size, 1, out_features))
        self.reset_parameters()

    def reset_parameters(self):
        for layer in range(self.ensemble_size):
            nn.init.kaiming_uniform_(self.weight[layer], a=math.sqrt(5))
        fan_in, _ = nn.init._calculate_fan_in_and_fan_out(self.weight[0])
        bound = 1 / math.sqrt(fan_in) if fan_in > 0 else 0
        nn.init.uniform_(self.bias, -bound, bound)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        return x @ self.weight + self.bias


ARGS = {
    "in_features": 4,
    "out_features": 3,
    "ensemble_size": 2
}
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 VectorizedLinear — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a VectorizedLinear 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
doge_config Ran
ran on a synthesized input · constructed an object; no result was computed
zzmtsvv/rl_task/doge/doge.py
pointer only (licence: NONE) · get_code("527759337db08777")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, zzmtsvv/rl_task when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

uniform_init Ran
ran; our contract draft was wrong, not the code
howuhh/sac-n-jax/sac_n_jax_flax.py
code served (permissive licence) · get_code("b5491b8fa060cb74")
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 jax
import jax.numpy as jnp

def uniform_init(bound: float):
    def _init(key, shape, dtype):
        return jax.random.uniform(key, shape=shape, minval=-bound, maxval=bound, dtype=dtype)
    return _init

# Test the function to understand what it does
# It takes a bound parameter and returns an initialization function
# The initialization function takes (key, shape, dtype) and returns random uniform values

ARGS = {
    "bound": 0.5
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — 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
Critic Not yet run
unverified (harvested, no recorded run)
howuhh/sac-n-jax/sac_n_jax_flax.py
code served (permissive licence) · get_code("5794e64ed17e773b")
How it ran

no invocation recorded.

DOGE Not yet run
unverified (harvested, no recorded run)
zzmtsvv/rl_task/doge/doge.py
pointer only (licence: NONE) · get_code("deb27526958aa790")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, zzmtsvv/rl_task when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

EDAC Not yet run
unverified (harvested, no recorded run)
corl-team/CORL/algorithms/offline/edac.py
code served (permissive licence) · get_code("0f2060e11b06334a")
How it ran

no invocation recorded.

from_numpy Not yet run
unverified (harvested, no recorded run)
snu-mllab/edac/lifelong_rl/models/networks.py
code served (permissive licence) · get_code("1d1f97e963885eb0")
How it ran

no invocation recorded.

ones Not yet run
unverified (harvested, no recorded run)
snu-mllab/edac/lifelong_rl/models/networks.py
code served (permissive licence) · get_code("78a13fda3d5531a2")
How it ran

no invocation recorded.

randn Not yet run
unverified (harvested, no recorded run)
snu-mllab/edac/lifelong_rl/models/networks.py
code served (permissive licence) · get_code("826e9009d90a358b")
How it ran

no invocation recorded.

soft_update Not yet run
unverified (harvested, no recorded run)
corl-team/CORL/algorithms/offline/edac.py
code served (permissive licence) · get_code("12fd41e242a13169")
How it ran

no invocation recorded.

zeros Not yet run
unverified (harvested, no recorded run)
snu-mllab/edac/lifelong_rl/models/networks.py
code served (permissive licence) · get_code("662ecb7bbf9644d8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 21 samples lifted from this paper's linked repositories, 21 carry a function identity and 0 do not. Of the 21 distinct functions with an identity, 9 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 2 the same month, 6 later) and 12 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.

FunctionAlso present inEarliest-datedOthers named
soft_update
brick of this paper's assembly: update_rule
6 other papers
2 dated earlier, 1 the same month, 3 later · 5 only as an assembly brick
2006.04779 (2020-06)
Conservative Q-Learning for Offline Reinforcement Learning
2006.09359 (2020-06)
AWAC: Accelerating Online Reinforcement Learning with Offline Datasets
2110.06169 (2021-10)
Offline Reinforcement Learning with Implicit Q-Learning
and 3 more not named here
EnsembledCritic
brick of this paper's assembly: helper
4 other papers
1 dated earlier, 1 the same month, 2 later · 2 only as an assembly brick
1906.00949 (2019-06)
Stabilizing Off-Policy Q-Learning via Bootstrapping Error Reduction
2110.06169 (2021-10)
Offline Reinforcement Learning with Implicit Q-Learning
2301.02328 (2023-01)
Extreme Q-Learning: MaxEnt RL without Entropy
and 1 more not named here
EnsembledLinear
brick of this paper's assembly: helper
2 other papers
1 dated earlier, 1 later · 1 only as an assembly brick
1906.00949 (2019-06)
Stabilizing Off-Policy Q-Learning via Bootstrapping Error Reduction
2306.06569 (2023-06)
Policy Regularization with Dataset Constraint for Offline Reinforcement Learning
VectorizedLinear
brick of this paper's assembly: helper
2 other papers
2 later · 2 only as an assembly brick
2406.16255 (2024-06)
Uncertainty-Aware Reward-Free Exploration with General Function Approximation
2506.12622 (2025-06)
DR-SAC: Distributionally Robust Soft Actor-Critic for Reinforcement Learning under Uncertainty
DeterministicActor
brick of this paper's assembly: helper
1 other paper
1 later
2306.06569 (2023-06)
Policy Regularization with Dataset Constraint for Offline Reinforcement Learning
from_numpy
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2010.08252 (2020-10)
Hyperparameter Auto-Tuning in Self-Supervised Robotic Learning
ones
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2010.08252 (2020-10)
Hyperparameter Auto-Tuning in Self-Supervised Robotic Learning
uniform_init
brick of this paper's assembly: helper
1 other paper
1 later
2512.04341 (2025-12)
Long-Horizon Model-Based Offline Reinforcement Learning Without Explicit Conservatism
zeros
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2010.08252 (2020-10)
Hyperparameter Auto-Tuning in Self-Supervised Robotic Learning

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

Offline reinforcement learning (offline RL), which aims to find an optimal policy from a previously collected static dataset, bears algorithmic difficulties due to function approximation errors from out-of-distribution (OOD) data points. To this end, offline RL algorithms adopt either a constraint or a penalty term that explicitly guides the policy to stay close to the given dataset. However, prior methods typically require accurate estimation of the behavior policy or sampling from OOD data points, which themselves can be a non-trivial problem. Moreover, these methods under-utilize the generalization ability of deep neural networks and often fall into suboptimal solutions too close to the given dataset. In this work, we propose an uncertainty-based offline RL method that takes into account the confidence of the Q-value prediction and does not require any estimation or sampling of the data distribution. We show that the clipped Q-learning, a technique widely used in online RL, can be leveraged to successfully penalize OOD data points with high prediction uncertainties. Surprisingly, we find that it is possible to substantially outperform existing offline RL methods on various tasks by simply increasing the number of Q-networks along with the clipped Q-learning. Based on this observation, we propose an ensemble-diversified actor-critic algorithm that reduces the number of required ensemble networks down to a tenth compared to the naive ensemble while achieving state-of-the-art performance on most of the D4RL benchmarks considered.

For agents

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

get_harvested_code_for_paper("2110.01548")
get_code_for_paper("2110.01548")
have("2110.01548")

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

curl https://syntology.ai/api/ran/2110.01548.json

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

[![Syntology run record](https://syntology.ai/api/ran/2110.01548.svg)](https://syntology.ai/paper/2110.01548)

Connect an agent — have() is free.