SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.14399 · ICML · 2024

Benchmarking Physics-Informed Time-Series Models for Operational Global Station Weather Forecasting

Lei Bai, Tao Han, Song Guo, Zhenghao Chen, Zhibin Wen, Dazhao Du

arXiv · PDF · Open in the Atlas

Code that ran

16 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (15 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 16 have no contract verdict; 0 failures of our instrument; 1 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/2406.14399.json · badge: /api/ran/2406.14399.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
taohan10200/weather-5k application (classifier label)
classifier basis: classifier_2025
16 ran · 1 no recorded run
FunctionStatusWhere it lives
AutoCorrelation Ran
ran on a synthesized input
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("c937754010c54537")
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 torch.nn as nn
import math

INIT_ARGS = {
    'mask_flag': True,
    'factor': 1,
    'scale': None,
    'attention_dropout': 0.1,
    'output_attention': False
}

ARGS = {
    'queries': torch.randn(2, 8, 4, 3),
    'keys': torch.randn(2, 8, 4, 3),
    'values': torch.randn(2, 8, 4, 3),
    'attn_mask': None
}

if __name__ == '__main__':
    from torch import nn
    
    class AutoCorrelation(nn.Module):

        def __init__(self, mask_flag=True, factor=1, scale=None, attention_dropout=0.1, output_attention=False):
            super(AutoCorrelation, self).__init__()
            self.factor = factor
            self.scale = scale
            self.mask_flag = mask_flag
            self.output_attention = output_attention
            self.dropout = nn.Dropout(attention_dropout)

        def values_delay_full(self, values, corr):
            batch = values.shape[0]
            head = values.shape[1]
            channel = values.shape[2]
            length = values.shape[3]
            init_index = torch.arange(length).unsqueeze(0).unsqueeze(0).unsqueeze(0).repeat(batch, head, channel, 1).to(values[0].device)
            top_k = int(self.factor * math.log(length))
            weights = torch.topk(corr, top_k, dim=-1)[0]
            delay = torch.topk(corr, top_k, dim=-1)[1]
            delay = length - delay
            tmp_corr = torch.softmax(weights, dim=-1)
            tmp_values = values.repeat(1, 1, 1, 2)
            delays_agg = torch.zeros_like(values).float()
            for i in range(top_k):
                tmp_delay = init_index + delay[..., i].unsqueeze(-1)
                pattern = torch.gather(tmp_values, dim=-1, index=tmp_delay)
                delays_agg = delays_agg + pattern * tmp_corr[..., i].unsqueeze(-1)
            return delays_agg

        def values_delay_channel(self, values, corr):
            batch = values.shape[0]
            head = values.shape[1]
            channel = values.shape[2]
            length = values.shape[3]
            init_index = torch.arange(length).unsqueeze(0).unsqueeze(0).unsqueeze(0).repeat(batch, head, channel, 1).to(values[0].device)
            top_k = int(self.factor * math.log(length))
            mean_value = torch.mean(corr, dim=2)
            weights = torch.topk(mean_value, top_k, dim=-1)[0]
            delay = torch.topk(mean_value, top_k, dim=-1)[1]
            tmp_corr = torch.softmax(weights, dim=-1)
            tmp_values = values.repeat(1, 1, 1, 2)
            delays_agg = torch.zeros_like(values).float()
            for i in range(top_k):
                tmp_delay = init_index + delay[..., i].unsqueeze(-1).unsqueeze(-1).repeat(1, 1, channel, length)
                pattern = torch.gather(tmp_values, dim=-1, index=tmp_delay)
                delays_agg = delays_agg + pattern * tmp_corr[..., i].unsqueeze(-1).unsqueeze(-1).repeat(1, 1, channel, length)
            return delays_agg

        def forward(self, queries, keys, values, attn_mask):
            B, L, H, E = queries.shape
            _, S, _, D = values.shape
            if L > S:
                zeros = torch.zeros_like(queries[:, :L - S, :]).float()
                values = torch.cat([values, zeros], dim=1)
                keys = torch.cat([keys, zeros], dim=1)
            else:
                values = values[:, :L, :, :]
                keys = keys[:, :L, :, :]
            q_fft = torch.fft.rfft(queries.permute(0, 2, 3, 1).contiguous(), dim=-1)
            k_fft = torch.fft.rfft(keys.permute(0, 2, 3, 1).contiguous(), dim=-1)
            res = q_fft * torch.conj(k_fft)
            corr = torch.fft.irfft(res, dim=-1)
            if self.training:
                V = self.values_delay_channel(values.permute(0, 2, 3, 1).contiguous(), corr).permute(0, 3, 1, 2)
            else:
                V = self.values_delay_channel(values.permute(0, 2, 3, 1).contiguous(), corr).permute(0, 3, 1, 2)
            if self.output_attention:
                return (V.contiguous(), corr.permute(0, 3, 1, 2))
            else:
                return (V.contiguous(), None)

    model = AutoCorrelation(**INIT_ARGS)
    model.eval()
    output = model(**ARGS)
    print("Output shapes:", output[0].shape, output[1])
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)
CausalBlock Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("2a1cd6852143de94")
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, CausalBlock: 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
from torch.nn.utils import weight_norm

torch.manual_seed(0)

ARGS = {
    "n_inputs": 3,
    "n_outputs": 4,
    "kernel_size": 3,
    "stride": 1,
    "dilation": 1,
    "padding": 1,
    "dropout": 0.2,
}
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 CausalBlock — 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 CausalBlock 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
CausalConv Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("bc4b6d85c64832c9")
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, CausalConv: 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
from torch.nn.utils import weight_norm

torch.manual_seed(0)

class Chomp1d(nn.Module):
    def __init__(self, chomp_size):
        super(Chomp1d, self).__init__()
        self.chomp_size = chomp_size

    def forward(self, x):
        return x[:, :, :-self.chomp_size].contiguous()

class CausalBlock(nn.Module):
    def __init__(self, n_inputs, n_outputs, kernel_size, stride, dilation, padding, dropout=0.2):
        super(CausalBlock, self).__init__()
        self.conv1 = weight_norm(nn.Conv1d(n_inputs, n_outputs, kernel_size, stride=stride, padding=padding, dilation=dilation))
        self.chomp1 = Chomp1d(padding)
        self.relu1 = nn.ReLU()
        self.dropout1 = nn.Dropout(dropout)
        self.conv2 = weight_norm(nn.Conv1d(n_outputs, n_outputs, kernel_size, stride=stride, padding=padding, dilation=dilation))
        self.chomp2 = Chomp1d(padding)
        self.relu2 = nn.ReLU()
        self.dropout2 = nn.Dropout(dropout)
        self.net = nn.Sequential(self.conv1, self.chomp1, self.relu1, self.dropout1, self.conv2, self.chomp2, self.relu2, self.dropout2)
        self.downsample = nn.Conv1d(n_inputs, n_outputs, 1) if n_inputs != n_outputs else None
        self.relu = nn.ReLU()
        self.init_weights()

    def init_weights(self):
        self.conv1.weight.data.normal_(0, 0.01)
        self.conv2.weight.data.normal_(0, 0.01)
        if self.downsample is not None:
            self.downsample.weight.data.normal_(0, 0.01)

    def forward(self, x):
        out = self.net(x)
        res = x if self.downsample is None else self.downsample(x)
        return self.relu(out + res)

class CausalConv(nn.Module):
    def __init__(self, num_inputs, num_channels, kernel_size=2, dropout=0.2):
        super(CausalConv, self).__init__()
        layers = []
        num_levels = len(num_channels)
        for i in range(num_levels):
            dilation_size = 3 ** i
            in_channels = num_inputs if i == 0 else num_channels[i - 1]
            out_channels = num_channels[i]
            layers += [CausalBlock(in_channels, out_channels, kernel_size, stride=1, dilation=dilation_size, padding=(kernel_size - 1) * dilation_size, dropout=dropout)]
        self.network = nn.Sequential(*layers)

    def forward(self, x):
        return self.network(x)

ARGS = {
    "num_inputs": 4,
    "num_channels": [8, 16, 32],
    "kernel_size": 2,
    "dropout": 0.2,
}
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 CausalConv — 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 CausalConv 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
Chomp1d Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("5c2a29db14be6a0a")
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, Chomp1d: 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

torch.manual_seed(0)

ARGS = {
    "chomp_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 Chomp1d — 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 Chomp1d 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
CrossCorrelation Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("55d5a3c23533ecf4")
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, CrossCorrelation: 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

torch.manual_seed(0)

# Create a minimal causal_fusion module that satisfies how it's used in forward()
class MinimalCausalFusion(nn.Module):
    def __init__(self):
        super(MinimalCausalFusion, self).__init__()
        self.linear = nn.Linear(1, 1)
    
    def forward(self, x):
        # x comes in as (B*H, E*L, N)
        # Just return it unchanged to satisfy the interface
        return x

# Build minimal tensors for forward() call
B = 2  # batch
N = 3  # nodes
L = 4  # query length
H = 5  # heads
E = 6  # embedding dimension
S = 4  # key/value length

queries = torch.randn(B, N, L, H, E)
keys = torch.randn(B, N, S, H, E)
values = torch.randn(B, N, S, H, E)
attn_mask = None

ARGS = {
    'causal_fusion': MinimalCausalFusion(),
    'mask_flag': True,
    'factor': 2,
    'scale': None,
    'attention_dropout': 0.1,
    'output_attention': 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 CrossCorrelation — 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 CrossCorrelation 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
CrossCorrelationLayer Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("dbc3b380a430e700")
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, CrossCorrelationLayer: 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

torch.manual_seed(0)

# Define a minimal correlation module that satisfies the interface
class MinimalCorrelation(nn.Module):
    def forward(self, queries, keys, values, attn_mask):
        # queries, keys, values shape: (batch, n_nodes, seq_len, n_heads, d_keys)
        # Return output and attention
        B, N, L, H, D = queries.shape
        out = torch.randn(B, N, L, H, D)
        attn = torch.randn(B, N, L, L)
        return out, attn

# Create instances
d_model = 8
n_heads = 2
d_keys = 4
d_values = 4

correlation_module = MinimalCorrelation()

ARGS = {
    'correlation': correlation_module,
    'd_model': d_model,
    'n_heads': n_heads,
    'd_keys': d_keys,
    'd_values': d_values,
}
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 CrossCorrelationLayer — 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 CrossCorrelationLayer 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
DataEmbedding Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("c4a745f5f73145a0")
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, DataEmbedding: 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 json
import tempfile
import os
import numpy as np
from sklearn.preprocessing import StandardScaler

torch.manual_seed(0)

# Create a temporary directory with meta_info.json
temp_dir = tempfile.mkdtemp()
meta_info = {
    "station_0": {"latitude": 40.7128, "ELEVATION": 10},
    "station_1": {"latitude": 34.0522, "ELEVATION": 285},
    "station_2": {"latitude": 41.8781, "ELEVATION": 181},
    "station_3": {"latitude": 29.7604, "ELEVATION": 47},
}

meta_info_path = os.path.join(temp_dir, "meta_info.json")
with open(meta_info_path, "w") as f:
    json.dump(meta_info, f)

# Create sample data
c_in = 3  # number of input channels
d_model = 8  # model dimension
node_num = 2  # number of nodes
batch_size = 2  # batch size
seq_len = 4  # sequence length

# Create dummy x tensor: (batch, seq_len, c_in)
x = torch.randn(batch_size, seq_len, c_in)

# Create dummy x_mark tensor with temporal information
# Format expected by TemporalEmbedding: (batch, seq_len, 5) where
# indices are [month, day, weekday, hour, minute]
x_mark = torch.tensor([
    [[1, 15, 3, 10, 0], [1, 15, 3, 11, 0], [1, 15, 3, 12, 0], [1, 15, 3, 13, 0]],
    [[1, 16, 4, 8, 0], [1, 16, 4, 9, 0], [1, 16, 4, 10, 0], [1, 16, 4, 11, 0]]
], dtype=torch.float32)

ARGS = {
    "c_in": c_in,
    "d_model": d_model,
    "root_path": temp_dir,
    "node_num": node_num,
    "embed_type": "fixed",
    "freq": "h",
    "dropout": 0.1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataEmbedding — 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 DataEmbedding 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
Decoder Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("2601733ce09afb32")
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, Decoder: 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

torch.manual_seed(0)

# Create a minimal decoder layer that matches the expected interface
class DecoderLayer(nn.Module):
    def __init__(self, d_model=4):
        super(DecoderLayer, self).__init__()
        self.linear = nn.Linear(d_model, d_model)
    
    def forward(self, x, cross, x_mask=None, cross_mask=None):
        # Returns (output, residual_trend)
        output = self.linear(x)
        residual_trend = torch.zeros(x.shape[0], 1, dtype=x.dtype)  # batch_size x 1
        return output, residual_trend

# Create minimal layers list
layers = [
    DecoderLayer(d_model=4),
    DecoderLayer(d_model=4)
]

# Create a norm layer (called in forward)
norm_layer = nn.LayerNorm(4)

# Create a projection layer (called in forward)
projection = nn.Linear(4, 3)

# Prepare input tensors for forward pass
batch_size = 2
seq_len = 5
d_model = 4

x = torch.randn(batch_size, seq_len, d_model)
cross = torch.randn(batch_size, seq_len, d_model)
x_mask = None
cross_mask = None
trend = torch.zeros(batch_size, 1)

ARGS = {
    "layers": layers,
    "norm_layer": norm_layer,
    "projection": projection
}
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 Decoder — 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 Decoder 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
DecoderLayer Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("ba4b7db0656f9fd6")
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, DecoderLayer: 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

torch.manual_seed(0)

# Build minimal mock attention/correlation modules
class MockMultiCorrelation(nn.Module):
    def forward(self, x, y, mask=None):
        # Should return tensor of same shape as x
        return torch.zeros_like(x)

class MockCrossMultiCorrelation(nn.Module):
    def forward(self, x, cross, mask=None):
        # Should return tensor of same shape as x
        return torch.zeros_like(x)

# Create instances
multi_correlation = MockMultiCorrelation()
cross_multi_correlation = MockCrossMultiCorrelation()

# Parameters for DecoderLayer
d_model = 4  # model dimension
c_out = 3    # output channels (distinct from d_model)
d_ff = 16    # feed-forward dimension
moving_avg_kernel = 25  # kernel size for moving average
dropout = 0.1
activation = 'relu'

ARGS = {
    'multi_correlation': multi_correlation,
    'cross_multi_correlation': cross_multi_correlation,
    'd_model': d_model,
    'c_out': c_out,
    'd_ff': d_ff,
    'moving_avg': moving_avg_kernel,
    'dropout': dropout,
    'activation': activation
}
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 DecoderLayer — 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 DecoderLayer 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
Encoder Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("23a94c41f4335756")
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, Encoder: 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

torch.manual_seed(0)

# Create a simple attention layer that accepts x and attn_mask
class SimpleAttentionLayer(nn.Module):
    def __init__(self, dim):
        super(SimpleAttentionLayer, self).__init__()
        self.linear = nn.Linear(dim, dim)
    
    def forward(self, x, attn_mask=None):
        # Simple pass-through with linear transformation
        return self.linear(x)

# Create attention layers
attn_layers = [
    SimpleAttentionLayer(4),
    SimpleAttentionLayer(4)
]

# Create a norm layer
norm_layer = nn.LayerNorm(4)

# Define ARGS dict
ARGS = {
    'attn_layers': attn_layers,
    'norm_layer': norm_layer
}
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 Encoder — 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 Encoder 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
EncoderLayer Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("ed00dfdc4c788366")
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, EncoderLayer: 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

torch.manual_seed(0)

# Create a minimal multi_correlation module that satisfies the usage pattern
class DummyMultiCorrelation(nn.Module):
    def __init__(self, d_model):
        super().__init__()
        self.d_model = d_model
        self.linear = nn.Linear(d_model, d_model)
    
    def forward(self, q, k, attn_mask=None):
        # The function calls: self.multi_correlation(x, x, attn_mask)
        # x shape is (batch, seq_len, d_model)
        # Return same shape as input
        return self.linear(q)

# Create instances
d_model = 4
d_ff = 16
moving_avg_kernel = 25
dropout = 0.1
activation = 'relu'

multi_correlation = DummyMultiCorrelation(d_model)

ARGS = {
    'multi_correlation': multi_correlation,
    'd_model': d_model,
    'd_ff': d_ff,
    'moving_avg': moving_avg_kernel,
    'dropout': dropout,
    'activation': activation
}
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 EncoderLayer — 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 EncoderLayer 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
GeoPositionalEmbedding Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("3d2068507fa9d360")
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, GeoPositionalEmbedding: 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

torch.manual_seed(0)

# GeoPositionalEmbedding takes c_in and d_model as constructor parameters
# c_in: input feature dimension (number of input channels/coordinates)
# d_model: output embedding dimension

c_in = 4  # input coordinate dimension
d_model = 3  # output embedding dimension

ARGS = {
    'c_in': c_in,
    'd_model': d_model,
}
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 GeoPositionalEmbedding — 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 GeoPositionalEmbedding 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
MultiCorrelation Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("55fc5b1b9eb20a44")
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, MultiCorrelation: 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

torch.manual_seed(0)

# Create a minimal auto_correlation module
class AutoCorrelation(nn.Module):
    def forward(self, q, k, v, attn_mask=None):
        # Return tuple with tensor of shape (B, L, C)
        B, L, C = q.shape
        output = torch.randn(B, L, C)
        return (output, None)

# Create a minimal cross_correlation module
class CrossCorrelation(nn.Module):
    def forward(self, q, k, v, attn_mask=None, n_nodes=None):
        # Return tuple with tensor of shape (B, S, C)
        B, S, C = q.shape
        output = torch.randn(B, S, C)
        return (output, None)

# Instantiate the correlation modules
auto_corr = AutoCorrelation()
cross_corr = CrossCorrelation()

# Parameters for MultiCorrelation
node_num = 2  # B // node_num should divide batch size evenly
node_list = [2, 2]  # Product should divide batch size
dropout_rate = 0.1

ARGS = {
    'auto_correlation': auto_corr,
    'cross_correlation': cross_corr,
    'node_num': node_num,
    'nodel_list': node_list,  # Note: parameter has typo "nodel_list"
    'dropout': dropout_rate
}
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 MultiCorrelation — 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 MultiCorrelation 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
moving_avg Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("627811d6e342bdbf")
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, moving_avg: 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

torch.manual_seed(0)

# The moving_avg module expects:
# - kernel_size: int, used in AvgPool1d and for repeat operations
# - stride: int, used in AvgPool1d
# - forward() expects x: tensor of shape (batch, time_steps, features)

# Create small test tensor: batch=2, time_steps=4, features=3
x = torch.randn(2, 4, 3)

ARGS = {
    'kernel_size': 3,
    'stride': 1
}
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 moving_avg — 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 moving_avg 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
my_Layernorm Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("bf5431d9a57a906b")
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, my_Layernorm: 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

torch.manual_seed(0)

# The function __init__ takes 'channels' parameter
# Looking at the forward method, x is expected to be a tensor
# The forward uses:
# - self.layernorm(x) which is nn.LayerNorm(channels)
# - torch.mean(x_hat, dim=1) which reduces along dimension 1
# - unsqueeze(1) and repeat(1, x.shape[1], 1)
# This suggests x should have shape (batch, seq_len, channels)
# where channels matches the parameter

channels = 4
batch_size = 2
seq_len = 3

ARGS = {
    'channels': channels
}
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 my_Layernorm — 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 my_Layernorm 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
series_decomp Ran
ran on a synthesized input · constructed an object; no result was computed
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("0476441daf861ec1")
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, series_decomp: 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

torch.manual_seed(0)

# series_decomp takes one parameter: kernel_size
# It creates a moving_avg module internally with kernel_size and stride=1
# The forward method expects input x and calls self.moving_avg(x)
# moving_avg expects x to have shape (batch, seq_len, features)

ARGS = {
    "kernel_size": 3
}
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 series_decomp — 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 series_decomp 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
Model Not yet run
unverified (harvested, no recorded run)
taohan10200/weather-5k/models/Corrformer.py
code served (permissive licence) · get_code("b8cc3b42dec2929e")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 5 are also present in other papers whose code we hold (5 with at least one paper dated earlier, 2 the same month, 3 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
series_decomp
brick of this paper's assembly: helper
15 other papers
8 dated earlier, 1 the same month, 6 later · 10 only as an assembly brick
2106.13008 (2021-06)
Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting
2201.12740 (2022-01)
FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting
2205.13504 (2022-05)
Are Transformers Effective for Time Series Forecasting?
and 12 more not named here
moving_avg
brick of this paper's assembly: helper
13 other papers
6 dated earlier, 1 the same month, 6 later · 11 only as an assembly brick
2106.13008 (2021-06)
Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting
2205.13504 (2022-05)
Are Transformers Effective for Time Series Forecasting?
2310.01728 (2023-10)
Time-LLM: Time Series Forecasting by Reprogramming Large Language Models
and 10 more not named here
Decoder
brick of this paper's assembly: model
3 other papers
3 dated earlier · 2 only as an assembly brick
2106.13008 (2021-06)
Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting
2201.12740 (2022-01)
FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting
2206.04038 (2022-06)
Scaleformer: Iterative Multi-scale Refining Transformers for Time Series Forecasting
my_Layernorm
brick of this paper's assembly: helper
3 other papers
3 dated earlier · 2 only as an assembly brick
2106.13008 (2021-06)
Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting
2201.12740 (2022-01)
FEDformer: Frequency Enhanced Decomposed Transformer for Long-term Series Forecasting
2206.04038 (2022-06)
Scaleformer: Iterative Multi-scale Refining Transformers for Time Series Forecasting
Chomp1d
brick of this paper's assembly: helper
2 other papers
1 dated earlier, 1 later · 1 only as an assembly brick
2112.01641 (2021-12)
Hamiltonian latent operators for content and motion disentanglement in image sequences
2511.03168 (2025-11)
UnCLe: Towards Scalable Dynamic Causal Discovery in Non-linear Temporal Systems

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

The development of Time-Series Forecasting (TSF) models is often constrained by the lack of comprehensive datasets, especially in Global Station Weather Forecasting (GSWF), where existing datasets are small, temporally short, and spatially sparse. To address this, we introduce WEATHER-5K, a large-scale observational weather dataset that better reflects real-world conditions and supports improved model training and evaluation. While recent TSF methods perform well on benchmarks, they still lag behind operational Numerical Weather Prediction (NWP) systems in capturing complex weather dynamics and extreme events. We propose PhysicsFormer, a physicsinformed forecasting model that combines a dynamic core with a Transformer residual to predict future weather states. Physical consistency is enforced via pressure-wind alignment and energy-aware smoothness losses, ensuring plausible dynamics while capturing complex temporal patterns. We benchmark PhysicsFormer and other TSF models against operational systems across several weather variables, extreme event prediction, and model complexity, providing a comprehensive assessment of the gap between academic TSF models and operational forecasting. The dataset and benchmark implementation are available at: https://github.com/ taohan10200/WEATHER-5K.

For agents

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

get_harvested_code_for_paper("2406.14399")
get_code_for_paper("2406.14399")
have("2406.14399")

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

curl https://syntology.ai/api/ran/2406.14399.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.14399.svg)](https://syntology.ai/paper/2406.14399)

Connect an agent — have() is free.