SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2405.14769 · ICML · 2024

Pragmatic Feature Preferences: Learning Reward-Relevant Preferences from Human Input

David Abel, Tianmin Shu, Andi Peng, Yuying Sun

arXiv · PDF · Open in the Atlas

Code that ran

4 of 5 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; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 2 failures of our instrument; 1 no recorded run), measured 2026-11-22; 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/2405.14769.json · badge: /api/ran/2405.14769.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
jlin816/rewards-from-language canonical (classifier label)
classifier basis: census
4 ran · 1 no recorded run
FunctionStatusWhere it lives
expand_categorical_var Ran
ran; our contract draft was wrong, not the code
jlin816/rewards-from-language/rewards/utils.py
code served (permissive licence) · get_code("489c96b934a1b5a6")
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 torch

def expand_categorical_var(feat_name, categories, selected_category):
    opt = {f'{feat_name}={cat}': 0 for cat in categories}
    opt[f'{feat_name}={selected_category}'] = 1
    return opt

ARGS = {
    'feat_name': 'color',
    'categories': ['red', 'blue', 'green', 'yellow'],
    'selected_category': 'blue'
}
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 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
extremal_indicators Ran
ran; our contract draft was wrong, not the code
jlin816/rewards-from-language/rewards/utils.py
code served (permissive licence) · get_code("11d2efcd245927eb")
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 int, shape (4, 3) 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

# Create a small 2D array with shape (4, 3)
# axis=0 represents samples, axis=1 represents features
x = np.array([
    [1.0, 5.0, 3.0],
    [2.0, 5.0, 2.0],
    [3.0, 2.0, 3.0],
    [0.0, 1.0, 1.0]
], dtype=np.float32)

# Define the function to use (default is np.max)
fn = np.max

# Define the unique parameter
unique = False

ARGS = {
    "x": x,
    "fn": fn,
    "unique": unique
}
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 int, shape (4, 3) — shape (4, 3) 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
validate_embedding_speaker Ran
ran, honoured the contract we drafted
jlin816/rewards-from-language/rewards/trainer.py
code served (permissive licence) · get_code("0161a3d0444d6c55")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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

torch.manual_seed(0)

class MockEmbeddingModel(nn.Module):
    def __init__(self, latent_type=None):
        super().__init__()
        self.latent_type = latent_type
        if latent_type is not None:
            self.latent_reward_weights = torch.tensor([0.5, 0.3, 0.2])
            self.latent_reward_prior_logits = torch.tensor([1.0, 0.5, 0.2])
    
    def add_negatives_and_score_batch(self, options, reward_weights, text, hard_negatives):
        batch_size = reward_weights.size(0)
        num_options = 5
        scores = torch.randn(num_options, batch_size, 3)
        extras = {}
        return scores, extras
    
    def get_reward_only_logits(self, scores):
        return torch.randn(scores.size(1))
    
    def marginalize_latents(self, scores):
        return torch.randn(scores.size(1), scores.size(2))
    
    def train(self, mode=True):
        super().train(mode)
        return self
    
    def eval(self):
        super().eval()
        return self

model = MockEmbeddingModel(latent_type="discrete")

def create_batch():
    batch_size = 2
    batch = {
        'utterance': torch.randn(batch_size, 4),
        'reward_weights': torch.randn(batch_size, 3),
        'options': torch.randn(batch_size, 5, 4),
        'hard_negatives': torch.randn(batch_size, 2, 4),
    }
    return batch

val_dl = [create_batch() for _ in range(2)]

ARGS = {
    'model': model,
    'val_dl': val_dl,
    'cuda': False,
    'topk': 3,
    'validate_as_listener': False,
}
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 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
validate_listener Ran
ran, honoured the contract we drafted
jlin816/rewards-from-language/rewards/trainer_listener.py
code served (permissive licence) · get_code("726f82ecbfc24254")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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
import torch.nn.functional as F

torch.manual_seed(0)

# Define a minimal model that satisfies the function's usage
class ListenerModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)  # 3 options
    
    def forward(self, text, options):
        # text and options are used, but we just need to return logits
        # matching the batch size and number of options
        batch_size = text.shape[0]
        return self.linear(text)

# Create model instance
model = ListenerModel()

# Create a minimal validation dataloader
# Each batch should be a dict with: utterance, reward_weights, options, optimal_index
val_dl = [
    {
        'utterance': torch.randn(2, 4),           # batch_size=2, embedding_dim=4
        'reward_weights': torch.randn(2, 3),     # batch_size=2, num_options=3
        'options': torch.randn(2, 3, 4),         # batch_size=2, num_options=3, dim=4
        'optimal_index': torch.tensor([0, 2])    # batch_size=2, indices in [0,1,2]
    },
    {
        'utterance': torch.randn(2, 4),
        'reward_weights': torch.randn(2, 3),
        'options': torch.randn(2, 3, 4),
        'optimal_index': torch.tensor([1, 0])
    }
]

ARGS = {
    'model': model,
    'val_dl': val_dl,
    'cuda': False
}
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 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
collate_rewards_and_options Not yet run
unverified (harvested, no recorded run)
jlin816/rewards-from-language/rewards/data.py
code served (permissive licence) · get_code("5597528d7319f714")
How it ran

no invocation recorded.

Also present in other papers' code

Of 5 samples lifted from this paper's linked repositories, 5 carry a function identity and 0 do not. Of the 5 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 5 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

Humans use social context to specify preferences over behaviors, i.e. their reward functions. Yet, algorithms for inferring reward models from preference data do not take this social learning view into account. Inspired by pragmatic human communication, we study how to extract fine-grained data regarding why an example is preferred that is useful for learning more accurate reward models. We propose to enrich binary preference queries to ask both (1) which features of a given example are preferable in addition to (2) comparisons between examples themselves. We derive an approach for learning from these feature-level preferences, both for cases where users specify which features are reward-relevant, and when users do not. We evaluate our approach on linear bandit settings in both vision-and language-based domains. Results support the efficiency of our approach in quickly converging to accurate rewards with fewer comparisons vs. example-only labels. Finally, we validate the real-world applicability with a behavioral experiment on a mushroom foraging task. Our findings suggest that incorporating pragmatic feature preferences is a promising approach for more efficient user-aligned reward learning.

For agents

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

get_harvested_code_for_paper("2405.14769")
get_code_for_paper("2405.14769")
have("2405.14769")

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

curl https://syntology.ai/api/ran/2405.14769.json

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

[![Syntology run record](https://syntology.ai/api/ran/2405.14769.svg)](https://syntology.ai/paper/2405.14769)

Connect an agent — have() is free.