SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2305.04477 · ICML · 2023

Behavior Contrastive Learning for Unsupervised Skill Discovery

Peng Liu, Hongyi Guo, Chenjia Bai, Siyuan Li, Zhen Wang, Bin Zhao, Rushuai Yang, Xuelong Li

arXiv · PDF · Open in the Atlas

Code that ran

7 of 9 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 3 failures of our instrument; 2 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/2305.04477.json · badge: /api/ran/2305.04477.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
Rooshy-yang/BeCL canonical (classifier label)
classifier basis: census
6 ran · 1 no recorded run
rooshy-yang/becl canonical (classifier label)
classifier basis: classifier_2025
1 ran · 1 no recorded run
FunctionStatusWhere it lives
BECL Ran
ran on a synthesized input · constructed an object; no result was computed
rooshy-yang/becl/agent/becl.py
code served (permissive licence) · get_code("53554f444b3a77db")
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, BECL: 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 types as _types

torch.manual_seed(0)

def weight_init(m):
    """Custom weight init for Conv2D and Linear layers."""
    if isinstance(m, nn.Linear):
        nn.init.orthogonal_(m.weight.data)
        if hasattr(m.bias, 'data'):
            m.bias.data.fill_(0.0)
    elif isinstance(m, nn.Conv2d) or isinstance(m, nn.ConvTranspose2d):
        gain = nn.init.calculate_gain('relu')
        nn.init.orthogonal_(m.weight.data, gain)
        if hasattr(m.bias, 'data'):
            m.bias.data.fill_(0.0)

class DummyUtil:
    pass

utils = _types.SimpleNamespace(weight_init=weight_init)

class BECL(nn.Module):

    def __init__(self, tau_dim, feature_dim, hidden_dim):
        super().__init__()
        self.embed = nn.Sequential(nn.Linear(tau_dim, hidden_dim), nn.ReLU(), nn.Linear(hidden_dim, hidden_dim), nn.ReLU(), nn.Linear(hidden_dim, feature_dim))
        self.project_head = nn.Sequential(nn.Linear(feature_dim, hidden_dim), nn.ReLU(), nn.Linear(hidden_dim, feature_dim))
        self.apply(utils.weight_init)

    def forward(self, tau):
        features = self.embed(tau)
        features = self.project_head(features)
        return features

ARGS = {
    'tau_dim': 5,
    'feature_dim': 7,
    'hidden_dim': 9
}
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 BECL — 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 BECL 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
episode_len Ran
ran, honoured the contract we drafted
Rooshy-yang/BeCL/replay_buffer.py
code served (permissive licence) · get_code("9f840a22d31a89bc")
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 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

torch.manual_seed(0)

# The function accesses episode.values() and calls .shape[0] on the result
# So episode should be a dict where values are tensors (or array-like with shape)
# The function returns shape[0] - 1, so we need at least 1 element in dimension 0

episode = {
    "observations": torch.randn(5, 4),
    "actions": torch.randn(5, 3),
    "rewards": torch.randn(5, 2),
}

ARGS = {
    "episode": episode
}
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 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
grad_norm Ran
ran, honoured the contract we drafted
Rooshy-yang/BeCL/utils.py
code served (permissive licence) · get_code("2478bb95b38ef24f")
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 float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a simple model with parameters that have gradients
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
    
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        return x

model = SimpleModel()

# Create some dummy input and target
x = torch.randn(2, 4)
target = torch.randn(2, 2)

# Forward pass to generate gradients
output = model(x)
loss = torch.nn.functional.mse_loss(output, target)
loss.backward()

# Extract parameters with gradients
params = [p for p in model.parameters() if p.grad is not None]

ARGS = {
    "params": params,
    "norm_type": 2.0
}
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 float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_episode Ran
ran; our contract draft was wrong, not the code
Rooshy-yang/BeCL/replay_buffer.py
code served (permissive licence) · get_code("184876e5f4ab6562")
How it ran

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

import numpy as np
from pathlib import Path
import tempfile
import os

# Create a temporary file to serve as the episode file
temp_dir = tempfile.mkdtemp()
temp_file_path = Path(temp_dir) / "episode.npz"

# Create a sample episode dictionary with numpy arrays
episode_data = {
    'observations': np.random.rand(5, 4).astype(np.float32),
    'actions': np.random.randint(0, 3, size=(5,)).astype(np.int32),
    'rewards': np.random.rand(5).astype(np.float32),
    'dones': np.random.randint(0, 2, size=(5,)).astype(bool),
}

# Save as npz file
np.savez(temp_file_path, **episode_data)

# Define ARGS with the file path
ARGS = {
    'fn': temp_file_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
make_replay_loader Ran
ran; our contract draft was wrong, not the code
Rooshy-yang/BeCL/replay_buffer.py
code served (permissive licence) · get_code("717d02cf517c8633")
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 DataLoader. 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
import random
from pathlib import Path
from unittest.mock import Mock, MagicMock
import tempfile
import os

torch.manual_seed(0)
np.random.seed(0)
random.seed(0)

# Create a mock storage object with required attributes
class MockStorage:
    def __init__(self):
        # Create a temporary directory for replay files
        self._temp_dir = tempfile.mkdtemp()
        self._replay_dir = Path(self._temp_dir)
        
        # Create mock meta specs
        self._meta_specs = []
        
        # Create a dummy episode file
        self._create_dummy_episode()
    
    def _create_dummy_episode(self):
        # Create a minimal valid episode file
        episode_data = {
            'observation': np.random.randn(10, 4).astype(np.float32),
            'action': np.random.randn(10, 3).astype(np.float32),
            'reward': np.random.randn(10, 1).astype(np.float32),
            'discount': np.ones((10, 1), dtype=np.float32),
        }
        
        # Save as npz file with proper naming: episode_{idx}_{length}.npz
        episode_path = self._replay_dir / 'episode_0_9.npz'
        np.savez(episode_path, **episode_data)

storage = MockStorage()

ARGS = {
    'storage': storage,
    'max_size': 1000,
    'batch_size': 2,
    'num_workers': 0,
    'save_snapshot': False,
    'nstep': 2,
    'discount': 0.99,
}
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 DataLoader — a DataLoader output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataLoader 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
param_norm Ran
ran, honoured the contract we drafted
Rooshy-yang/BeCL/utils.py
code served (permissive licence) · get_code("8708aaca52006e47")
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 float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a small neural network with a few parameters
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
    
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        return x

model = SimpleModel()

# Extract parameters from the model
params = list(model.parameters())

# Define ARGS dict with all parameters
ARGS = {
    'params': params,
    'norm_type': 2.0
}
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 float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
to_torch Ran
ran; our contract draft was wrong, not the code
Rooshy-yang/BeCL/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from premiertaco/premier-taco, is pointer only (licence: NONE) · get_code("6244e9922f4e0610")
How it ran

The body of this sample is withheld: MIT, as GitHub detects it on this paper's repository, Rooshy-yang/BeCL, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

make_agent Not yet run
unverified (harvested, no recorded run)
Rooshy-yang/BeCL/finetune.py
code served (permissive licence) · get_code("2a76dfe317adeb03")
How it ran

no invocation recorded.

weight_init Not yet run
unverified (harvested, no recorded run)
rooshy-yang/becl/agent/becl.py
code served (permissive licence) · get_code("2d7fab6117dbc6a8")
How it ran

no invocation recorded.

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, 8 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 0 the same month, 8 later) and 1 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
to_torch 23 other papers
11 dated earlier, 12 later · 7 only as an assembly brick
2103.04551 (2021-03)
Behavior From the Void: Unsupervised Active Pre-Training
2107.09645 (2021-07)
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
2202.10324 (2022-02)
VRL3: A Data-Driven Framework for Visual Deep Reinforcement Learning
and 20 more not named here
weight_init
brick of this paper's assembly: helper
13 other papers
7 dated earlier, 6 later · 12 only as an assembly brick
2004.13649 (2020-04)
Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixels
2103.04551 (2021-03)
Behavior From the Void: Unsupervised Active Pre-Training
2110.14118 (2021-10)
Object-Aware Regularization for Addressing Causal Confusion in Imitation Learning
and 10 more not named here
episode_len 12 other papers
4 dated earlier, 8 later
2107.09645 (2021-07)
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
2211.12740 (2022-11)
Masked Autoencoding for Scalable and Generalizable Decision Making
2302.00965 (2023-02)
Visual Imitation Learning with Patch Rewards
and 9 more not named here
load_episode 10 other papers
3 dated earlier, 7 later
2107.09645 (2021-07)
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
2302.00965 (2023-02)
Visual Imitation Learning with Patch Rewards
2303.01497 (2023-03)
Teach a Robot to FISH: Versatile Imitation from One Minute of Demonstrations
and 7 more not named here
make_agent 9 other papers
4 dated earlier, 5 later
2103.04551 (2021-03)
Behavior From the Void: Unsupervised Active Pre-Training
2110.15191 (2021-10)
URLB: Unsupervised Reinforcement Learning Benchmark
2202.00161 (2022-02)
CIC: Contrastive Intrinsic Control for Unsupervised Skill Discovery
and 6 more not named here
grad_norm 4 other papers
1 dated earlier, 3 later
2211.12740 (2022-11)
Masked Autoencoding for Scalable and Generalizable Decision Making
2405.14073 (2024-05)
PEAC: Unsupervised Pre-training for Cross-Embodiment Reinforcement Learning
2405.16030 (2024-05)
Constrained Ensemble Exploration for Unsupervised Skill Discovery
and 1 more not named here
param_norm 3 other papers
3 later
2405.14073 (2024-05)
PEAC: Unsupervised Pre-training for Cross-Embodiment Reinforcement Learning
2405.16030 (2024-05)
Constrained Ensemble Exploration for Unsupervised Skill Discovery
2406.16255 (2024-06)
Uncertainty-Aware Reward-Free Exploration with General Function Approximation
make_replay_loader 1 other paper
1 later
2405.16030 (2024-05)
Constrained Ensemble Exploration for Unsupervised Skill Discovery

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

In reinforcement learning, unsupervised skill discovery aims to learn diverse skills without extrinsic rewards. Previous methods discover skills by maximizing the mutual information (MI) between states and skills. However, such an MI objective tends to learn simple and static skills and may hinder exploration. In this paper, we propose a novel unsupervised skill discovery method through contrastive learning among behaviors, which makes the agent produce similar behaviors for the same skill and diverse behaviors for different skills. Under mild assumptions, our objective maximizes the MI between different behaviors based on the same skill, which serves as an upper bound of the previous MI objective. Meanwhile, our method implicitly increases the state entropy to obtain better state coverage. We evaluate our method on challenging mazes and continuous control tasks. The results show that our method generates diverse and far-reaching skills, and also obtains competitive performance in downstream tasks compared to the state-of-the-art methods.

For agents

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

get_harvested_code_for_paper("2305.04477")
get_code_for_paper("2305.04477")
have("2305.04477")

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

curl https://syntology.ai/api/ran/2305.04477.json

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

[![Syntology run record](https://syntology.ai/api/ran/2305.04477.svg)](https://syntology.ai/paper/2305.04477)

Connect an agent — have() is free.