SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2110.02544 · NeurIPS · 2021

Learning to Iteratively Solve Routing Problems with Dual-Aspect Collaborative Transformer

Yining Ma, Jingwen Li, Zhiguang Cao, Wen Song, Le Zhang, Zhenghua Chen, Jing Tang, Dac Encoder, Dac Decoder, ×𝑳 Decoder

arXiv · PDF · Open in the Atlas

Code that ran

9 of 16 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 9 have no contract verdict; 0 failures of our instrument; 7 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/2110.02544.json · badge: /api/ran/2110.02544.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
yining043/VRP-DACT canonical (classifier label)
classifier basis: classifier_2025
7 ran · 4 no recorded run
yining043/TSP-improve community
Papers with Code archive
2 ran · 3 no recorded run
FunctionStatusWhere it lives
EmbeddingNet Ran
ran on a synthesized input
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("d32eca8fe0b18c43")
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 tuple. 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

# INIT_ARGS for EmbeddingNet.__init__(node_dim, embedding_dim, seq_length)
INIT_ARGS = {
    'node_dim': 5,
    'embedding_dim': 8,
    'seq_length': 6
}

# ARGS for EmbeddingNet.forward(x, solutions, visited_time=None)
batch_size = 2
seq_length = 6  # Must match INIT_ARGS['seq_length']
node_dim = 5    # Must match INIT_ARGS['node_dim']

# x: node features, shape (batch_size, seq_length, node_dim)
x = torch.randn(batch_size, seq_length, node_dim)

# solutions: permutation indices, shape (batch_size, seq_length)
# Each row should be a permutation of indices 0 to seq_length-1
solutions = torch.tensor([
    [0, 1, 2, 3, 4, 5],
    [2, 0, 4, 1, 5, 3]
], dtype=torch.long)

# visited_time: optional, can be None or shape (batch_size, seq_length)
visited_time = None

ARGS = {
    'x': x,
    'solutions': solutions,
    'visited_time': visited_time
}
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 tuple — a tuple output has no array shape of its own; none is 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
FFandNormsubLayer Ran
ran on a synthesized input
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("850da12d3368c65a")
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 tuple. 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
import math

class Normalization(nn.Module):

    def __init__(self, embed_dim, normalization='batch'):
        super(Normalization, self).__init__()
        normalizer_class = {'batch': nn.BatchNorm1d, 'instance': nn.InstanceNorm1d}.get(normalization, None)
        self.normalization = normalization
        if not self.normalization == 'layer':
            self.normalizer = normalizer_class(embed_dim, affine=True)

    def init_parameters(self):
        for name, param in self.named_parameters():
            stdv = 1.0 / math.sqrt(param.size(-1))
            param.data.uniform_(-stdv, stdv)

    def forward(self, input):
        if self.normalization == 'layer':
            return (input - input.mean((1, 2)).view(-1, 1, 1)) / torch.sqrt(input.var((1, 2)).view(-1, 1, 1) + 1e-05)
        if isinstance(self.normalizer, nn.BatchNorm1d):
            return self.normalizer(input.view(-1, input.size(-1))).view(*input.size())
        elif isinstance(self.normalizer, nn.InstanceNorm1d):
            return self.normalizer(input.permute(0, 2, 1)).permute(0, 2, 1)
        else:
            assert self.normalizer is None, 'Unknown normalizer type'
            return input

class FFandNormsubLayer(nn.Module):

    def __init__(self, n_heads, embed_dim, feed_forward_hidden, normalization='layer'):
        super(FFandNormsubLayer, self).__init__()
        self.FF1 = nn.Sequential(nn.Linear(embed_dim, feed_forward_hidden), nn.ReLU(inplace=True), nn.Linear(feed_forward_hidden, embed_dim)) if feed_forward_hidden > 0 else nn.Linear(embed_dim, embed_dim)
        self.FF2 = nn.Sequential(nn.Linear(embed_dim, feed_forward_hidden), nn.ReLU(inplace=True), nn.Linear(feed_forward_hidden, embed_dim)) if feed_forward_hidden > 0 else nn.Linear(embed_dim, embed_dim)
        self.Norm = Normalization(embed_dim, normalization)

    def forward(self, input1, input2):
        out1 = self.FF1(input1)
        out2 = self.FF2(input2)
        return (self.Norm(out1 + input1), self.Norm(out2 + input2))


INIT_ARGS = {
    'n_heads': 2,
    'embed_dim': 4,
    'feed_forward_hidden': 8,
    'normalization': 'layer'
}

ARGS = {
    'input1': torch.randn(2, 3, 4),
    'input2': torch.randn(2, 3, 4)
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MLP Ran
ran on a synthesized input
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("355814dd5ff90a73")
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 1, dtype family float, shape (2,) 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
import math

class MLP(torch.nn.Module):

    def __init__(self, input_dim=128, feed_forward_dim=64, embedding_dim=64, output_dim=1):
        super(MLP, self).__init__()
        self.fc1 = torch.nn.Linear(input_dim, feed_forward_dim)
        self.fc2 = torch.nn.Linear(feed_forward_dim, embedding_dim)
        self.fc3 = torch.nn.Linear(embedding_dim, output_dim)
        self.dropout = torch.nn.Dropout(p=0.05)
        self.ReLU = nn.ReLU(inplace=True)
        self.init_parameters()

    def init_parameters(self):
        for param in self.parameters():
            stdv = 1.0 / math.sqrt(param.size(-1))
            param.data.uniform_(-stdv, stdv)

    def forward(self, in_):
        result = self.ReLU(self.fc1(in_))
        result = self.dropout(result)
        result = self.ReLU(self.fc2(result))
        result = self.fc3(result).squeeze(-1)
        return result


INIT_ARGS = {
    'input_dim': 4,
    'feed_forward_dim': 3,
    'embedding_dim': 2,
    'output_dim': 1
}

ARGS = {
    'in_': torch.randn(2, 4)
}

if __name__ == '__main__':
    model = MLP(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MultiHeadAttention Ran
ran on a synthesized input
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("fa54d99ecac04864")
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 tuple. 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

INIT_ARGS = {
    'n_heads': 2,
    'input_dim': 4,
    'embed_dim': 8,
}

# Create dummy inputs matching INIT_ARGS dimensions
batch_size = 2
graph_size = 3

# h_node_in: (batch_size, graph_size, input_dim)
h_node_in = torch.randn(batch_size, graph_size, INIT_ARGS['input_dim'])

# h_pos_in: (batch_size, graph_size, input_dim)
h_pos_in = torch.randn(batch_size, graph_size, INIT_ARGS['input_dim'])

ARGS = {
    'h_node_in': h_node_in,
    'h_pos_in': h_pos_in,
}
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 tuple — a tuple output has no array shape of its own; none is 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MultiHeadAttention Ran
ran on a synthesized input
yining043/TSP-improve/nets/attention_model.py
code served (permissive licence) · get_code("3e9c7f43e77628ea")
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, shape (2, 3, 16) 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
from torch import nn
import torch.nn.functional as F
import numpy as np
import math

class MultiHeadAttention(nn.Module):

    def __init__(self, n_heads, input_dim, embed_dim=None, val_dim=None, key_dim=None):
        super(MultiHeadAttention, self).__init__()
        if val_dim is None:
            assert embed_dim is not None, 'Provide either embed_dim or val_dim'
            val_dim = embed_dim // n_heads
        if key_dim is None:
            key_dim = val_dim
        self.n_heads = n_heads
        self.input_dim = input_dim
        self.embed_dim = embed_dim
        self.val_dim = val_dim
        self.key_dim = key_dim
        self.norm_factor = 1 / math.sqrt(key_dim)
        self.W_query = nn.Parameter(torch.Tensor(n_heads, input_dim, key_dim))
        self.W_key = nn.Parameter(torch.Tensor(n_heads, input_dim, key_dim))
        self.W_val = nn.Parameter(torch.Tensor(n_heads, input_dim, val_dim))
        if embed_dim is not None:
            self.W_out = nn.Parameter(torch.Tensor(n_heads, key_dim, embed_dim))
        self.init_parameters()

    def init_parameters(self):
        for param in self.parameters():
            stdv = 1.0 / math.sqrt(param.size(-1))
            param.data.uniform_(-stdv, stdv)

    def forward(self, q, h=None, mask=None):
        """

        :param q: queries (batch_size, n_query, input_dim)
        :param h: data (batch_size, graph_size, input_dim)
        :param mask: mask (batch_size, n_query, graph_size) or viewable as that (i.e. can be 2 dim if n_query == 1)
        Mask should contain 1 if attention is not possible (i.e. mask is negative adjacency)
        :return:
        """
        if h is None:
            h = q
        batch_size, graph_size, input_dim = h.size()
        n_query = q.size(1)
        assert q.size(0) == batch_size
        assert q.size(2) == input_dim
        assert input_dim == self.input_dim, 'Wrong embedding dimension of input'
        hflat = h.contiguous().view(-1, input_dim)
        qflat = q.contiguous().view(-1, input_dim)
        shp = (self.n_heads, batch_size, graph_size, -1)
        shp_q = (self.n_heads, batch_size, n_query, -1)
        Q = torch.matmul(qflat, self.W_query).view(shp_q)
        K = torch.matmul(hflat, self.W_key).view(shp)
        V = torch.matmul(hflat, self.W_val).view(shp)
        compatibility = self.norm_factor * torch.matmul(Q, K.transpose(2, 3))
        if mask is not None:
            mask = mask.view(1, batch_size, n_query, graph_size).expand_as(compatibility)
            compatibility[mask] = -np.inf
        attn = F.softmax(compatibility, dim=-1)
        if mask is not None:
            attnc = attn.clone()
            attnc[mask] = 0
            attn = attnc
        heads = torch.matmul(attn, V)
        out = torch.mm(heads.permute(1, 2, 0, 3).contiguous().view(-1, self.n_heads * self.val_dim), self.W_out.view(-1, self.embed_dim)).view(batch_size, n_query, self.embed_dim)
        return out


INIT_ARGS = {
    'n_heads': 2,
    'input_dim': 8,
    'embed_dim': 16,
    'val_dim': None,
    'key_dim': None,
}

ARGS = {
    'q': torch.randn(2, 3, 8),
    'h': torch.randn(2, 5, 8),
    'mask': None,
}

if __name__ == '__main__':
    model = MultiHeadAttention(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 3, 16)")
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, shape (2, 3, 16) — shape (2, 3, 16) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MultiHeadCompat Ran
ran on a synthesized input
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("211b65805e3d00f5")
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 4, dtype family float, shape (2, 2, 3, 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 torch
from torch import nn
import math

class MultiHeadCompat(nn.Module):

    def __init__(self, n_heads, input_dim, embed_dim=None, val_dim=None, key_dim=None):
        super(MultiHeadCompat, self).__init__()
        if val_dim is None:
            val_dim = embed_dim // n_heads
        if key_dim is None:
            key_dim = val_dim
        self.n_heads = n_heads
        self.input_dim = input_dim
        self.embed_dim = embed_dim
        self.val_dim = val_dim
        self.key_dim = key_dim
        self.norm_factor = 1 / math.sqrt(1 * key_dim)
        self.W_query = nn.Parameter(torch.Tensor(n_heads, input_dim, key_dim))
        self.W_key = nn.Parameter(torch.Tensor(n_heads, input_dim, key_dim))
        self.init_parameters()

    def init_parameters(self):
        for param in self.parameters():
            stdv = 1.0 / math.sqrt(param.size(-1))
            param.data.uniform_(-stdv, stdv)

    def forward(self, q, h=None, mask=None):
        if h is None:
            h = q
        batch_size, graph_size, input_dim = h.size()
        n_query = q.size(1)
        hflat = h.contiguous().view(-1, input_dim)
        qflat = q.contiguous().view(-1, input_dim)
        shp = (self.n_heads, batch_size, graph_size, -1)
        shp_q = (self.n_heads, batch_size, n_query, -1)
        Q = torch.matmul(qflat, self.W_query).view(shp_q)
        K = torch.matmul(hflat, self.W_key).view(shp)
        compatibility = torch.matmul(Q, K.transpose(2, 3))
        return self.norm_factor * compatibility


INIT_ARGS = {
    'n_heads': 2,
    'input_dim': 4,
    'embed_dim': 8,
    'val_dim': 4,
    'key_dim': 4,
}

ARGS = {
    'q': torch.randn(2, 3, 4),
    'h': torch.randn(2, 5, 4),
    'mask': None,
}

if __name__ == '__main__':
    model = MultiHeadCompat(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 2, 3, 5) — shape (2, 2, 3, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MultiHeadDecoder Ran
ran on a synthesized input
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("30b971afd000a522")
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 torch
import numpy as np
from torch import nn

# INIT_ARGS for MultiHeadDecoder
INIT_ARGS = {
    'n_heads': 2,
    'input_dim': 4,
    'embed_dim': 8,
    'val_dim': None,
    'key_dim': 4
}

# ARGS for MultiHeadDecoder.forward
batch_size = 2
graph_size = 3
dim = INIT_ARGS['embed_dim']

h_em = torch.randn(batch_size, graph_size, dim)
pos_em = torch.randn(batch_size, graph_size, dim)
solving_state_for_net = None  # Not used in forward, so can be None

ARGS = {
    'h_em': h_em,
    'pos_em': pos_em,
    'solving_state_for_net': solving_state_for_net
}
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)
MultiHeadDecoder Ran
ran on a synthesized input
yining043/TSP-improve/nets/attention_model.py
code served (permissive licence) · get_code("5b8ea40ba7cb0c66")
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 tuple. 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
import torch.nn.functional as F
import numpy as np
import math


class MockProblem:
    """Mock problem class that provides get_swap_mask method"""
    def __init__(self, graph_size):
        self.graph_size = graph_size
    
    def get_swap_mask(self, rec):
        """Return a mask for diagonal/invalid swaps"""
        batch_size = rec.size(0)
        mask = torch.zeros(batch_size, self.graph_size, self.graph_size)
        for i in range(batch_size):
            for j in range(self.graph_size):
                mask[i, j, j] = 1.0
        return mask


class MultiHeadDecoder(nn.Module):

    def __init__(self, problem, input_dim, embed_dim=None, val_dim=None, key_dim=None):
        super(MultiHeadDecoder, self).__init__()
        self.problem = problem
        self.n_heads = 1
        if val_dim is None:
            assert embed_dim is not None, 'Provide either embed_dim or val_dim'
            val_dim = embed_dim // self.n_heads
        if key_dim is None:
            key_dim = val_dim
        self.input_dim = input_dim
        self.embed_dim = embed_dim
        self.val_dim = val_dim
        self.key_dim = key_dim
        self.norm_factor = 1 / math.sqrt(key_dim)
        self.W_query = nn.Parameter(torch.Tensor(self.n_heads, input_dim, key_dim))
        self.W_key = nn.Parameter(torch.Tensor(self.n_heads, input_dim, key_dim))
        self.init_parameters()

    def init_parameters(self):
        for param in self.parameters():
            stdv = 1.0 / math.sqrt(param.size(-1))
            param.data.uniform_(-stdv, stdv)

    def forward(self, q, exchange, rec):
        """

        :param q: queries (batch_size, n_query, input_dim)
        :param exchange: (batch_size, 2)
        :return:
        """
        h = q
        batch_size, graph_size, input_dim = h.size()
        n_query = q.size(1)
        assert q.size(0) == batch_size
        assert q.size(2) == input_dim
        assert input_dim == self.input_dim, 'Wrong embedding dimension of input'
        hflat = h.contiguous().view(-1, input_dim)
        qflat = q.contiguous().view(-1, input_dim)
        shp = (self.n_heads, batch_size, graph_size, -1)
        shp_q = (self.n_heads, batch_size, n_query, -1)
        Q = torch.matmul(qflat, self.W_query).view(shp_q)
        K = torch.matmul(hflat, self.W_key).view(shp)
        compatibility_raw = self.norm_factor * torch.matmul(Q, K.transpose(2, 3))
        compatibility = torch.tanh(compatibility_raw[0]) * 10.0
        mask_dia = self.problem.get_swap_mask(rec).expand_as(compatibility)
        compatibility[mask_dia.bool()] = -np.inf
        if exchange is not None:
            compatibility[torch.arange(batch_size), exchange[:, 0], exchange[:, 1]] = -np.inf
            compatibility[torch.arange(batch_size), exchange[:, 1], exchange[:, 0]] = -np.inf
        im = compatibility.view(batch_size, -1)
        im_log_soft = F.log_softmax(im, dim=-1)
        im_soft = F.softmax(im, dim=-1)
        return (im_log_soft, im_soft)


# Setup configuration
input_dim = 4
embed_dim = 8
graph_size = 3
batch_size = 2

problem = MockProblem(graph_size)

INIT_ARGS = {
    'problem': problem,
    'input_dim': input_dim,
    'embed_dim': embed_dim,
    'val_dim': None,
    'key_dim': None,
}

ARGS = {
    'q': torch.randn(batch_size, graph_size, input_dim),
    'exchange': torch.tensor([[0, 1], [1, 2]]),
    'rec': torch.zeros(batch_size, graph_size),
}
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 tuple — a tuple output has no array shape of its own; none is 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Normalization Ran
ran on a synthesized input
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("08bc67177889a2b3")
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, shape (2, 3, 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 torch
from torch import nn
import math

class Normalization(nn.Module):

    def __init__(self, embed_dim, normalization='batch'):
        super(Normalization, self).__init__()
        normalizer_class = {'batch': nn.BatchNorm1d, 'instance': nn.InstanceNorm1d}.get(normalization, None)
        self.normalization = normalization
        if not self.normalization == 'layer':
            self.normalizer = normalizer_class(embed_dim, affine=True)

    def init_parameters(self):
        for name, param in self.named_parameters():
            stdv = 1.0 / math.sqrt(param.size(-1))
            param.data.uniform_(-stdv, stdv)

    def forward(self, input):
        if self.normalization == 'layer':
            return (input - input.mean((1, 2)).view(-1, 1, 1)) / torch.sqrt(input.var((1, 2)).view(-1, 1, 1) + 1e-05)
        if isinstance(self.normalizer, nn.BatchNorm1d):
            return self.normalizer(input.view(-1, input.size(-1))).view(*input.size())
        elif isinstance(self.normalizer, nn.InstanceNorm1d):
            return self.normalizer(input.permute(0, 2, 1)).permute(0, 2, 1)
        else:
            assert self.normalizer is None, 'Unknown normalizer type'
            return input


INIT_ARGS = {
    'embed_dim': 4,
    'normalization': 'batch'
}

ARGS = {
    'input': torch.randn(2, 3, 4)
}

if __name__ == '__main__':
    model = Normalization(**INIT_ARGS)
    output = model(**ARGS)
    print(output.shape)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
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, shape (2, 3, 4) — shape (2, 3, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Actor Not yet run
unverified (harvested, no recorded run)
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("c62ea025f6578059")
How it ran

no invocation recorded.

AttentionModel Not yet run
unverified (harvested, no recorded run)
yining043/TSP-improve/nets/attention_model.py
code served (permissive licence) · get_code("afcbe0b62b04095b")
How it ran

no invocation recorded.

EmbeddingNet Not yet run
unverified (harvested, no recorded run)
yining043/TSP-improve/nets/attention_model.py
code served (permissive licence) · get_code("5a3c7c6596133237")
How it ran

no invocation recorded.

MultiHeadAttentionLayer Not yet run
unverified (harvested, no recorded run)
yining043/TSP-improve/nets/attention_model.py
code served (permissive licence) · get_code("1905f5c7465accec")
How it ran

no invocation recorded.

MultiHeadAttentionsubLayer Not yet run
unverified (harvested, no recorded run)
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("500395d982e1b1bb")
How it ran

no invocation recorded.

MultiHeadEncoder Not yet run
unverified (harvested, no recorded run)
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("8147e91b3d0fe9f1")
How it ran

no invocation recorded.

mySequential Not yet run
unverified (harvested, no recorded run)
yining043/VRP-DACT/nets/actor_network.py
code served (permissive licence) · get_code("b48a3565db4b5491")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 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 16 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

Recently, Transformer has become a prevailing deep architecture for solving vehicle routing problems (VRPs). However, it is less effective in learning improvement models for VRP because its positional encoding (PE) method is not suitable in representing VRP solutions. This paper presents a novel Dual-Aspect Collaborative Transformer (DACT) to learn embeddings for the node and positional features separately, instead of fusing them together as done in existing ones, so as to avoid potential noises and incompatible correlations. Moreover, the positional features are embedded through a novel cyclic positional encoding (CPE) method to allow Transformer to effectively capture the circularity and symmetry of VRP solutions (i.e., cyclic sequences). We train DACT using Proximal Policy Optimization and design a curriculum learning strategy for better sample efficiency. We apply DACT to solve the traveling salesman problem (TSP) and capacitated vehicle routing problem (CVRP). Results show that our DACT outperforms existing Transformer based improvement models, and exhibits much better generalization performance across different problem sizes on synthetic and benchmark instances, respectively.

For agents

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

get_harvested_code_for_paper("2110.02544")
get_code_for_paper("2110.02544")
have("2110.02544")

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

curl https://syntology.ai/api/ran/2110.02544.json

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

[![Syntology run record](https://syntology.ai/api/ran/2110.02544.svg)](https://syntology.ai/paper/2110.02544)

Connect an agent — have() is free.