SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.07556 · ICLR · 2021

INTERPRETABLE NEURAL ARCHITECTURE SEARCH VIA BAYESIAN OPTIMISATION WITH WEISFEILER-LEHMAN KERNELS

Binxin Ru, Xingchen Wan, Xiaowen Dong, Michael Osborne

arXiv · PDF · Open in the Atlas

Code that ran

15 of 19 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; 1 honoured the contract we drafted, 0 violated it and 13 have no contract verdict; 1 failure of our instrument; 4 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/2006.07556.json · badge: /api/ran/2006.07556.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
automl/nas_benchmarks canonical (classifier label)
classifier basis: census
4 ran · 2 no recorded run
xingchenwan/nasbowl community (classifier label)
classifier basis not recorded on the link
10 ran · 2 no recorded run
google-research/nasbench community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
add_color Ran
ran on a synthesized input
xingchenwan/nasbowl/bayesopt/utils.py
code served (permissive licence) · get_code("202d26d511c5371d")
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 DiGraph. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-26T22:17:14Z. 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 networkx as nx

# Create a simple directed graph that mimics an architecture graph
arch = nx.DiGraph()

# Add nodes with 'op_name' data attribute
arch.add_node(0, op_name='conv')
arch.add_node(1, op_name='relu')
arch.add_node(2, op_name='pool')
arch.add_node(3, op_name='fc')

# Add some edges to make it a valid graph
arch.add_edge(0, 1)
arch.add_edge(1, 2)
arch.add_edge(2, 3)

# Create a color map for operation types
color_map = {
    'conv': '#FF0000',
    'relu': '#00FF00',
    'pool': '#0000FF',
    'fc': '#FFFF00'
}

ARGS = {
    'arch': arch,
    'color_map': color_map
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DiGraph — a DiGraph output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DiGraph output the recorded ndim 1 and dtype family int describe that coercion, not the output, and are not stated
Environment
image syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); 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
build_graph Ran
ran on a synthesized input
xingchenwan/nasbowl/loader.py
code served (permissive licence) · get_code("2a97c65c41cd8219")
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 DiGraph. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import networkx as nx

# Create a simple adjacency matrix
# 5 nodes with some connections
A = np.array([
    [0, 1, 0, 0, 0],
    [0, 0, 1, 1, 0],
    [0, 0, 0, 1, 0],
    [0, 0, 0, 0, 1],
    [0, 0, 0, 0, 0]
], dtype=int)

# Create node names corresponding to the 5 nodes
nodes_names = ['input', 'conv1x1-bn-relu', 'conv3x3-bn-relu', 'maxpool3x3', 'output']

ARGS = {
    'A': A,
    'nodes_names': nodes_names,
    'create_directed_graph': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DiGraph — a DiGraph output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DiGraph output the recorded ndim 1 and dtype family int 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
create_nasbench201_graph Ran
ran on a synthesized input
xingchenwan/nasbowl/bayesopt/generate_test_graphs.py
code served (permissive licence) · get_code("3d48c1d48f8af9a9")
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 DiGraph. 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 networkx as nx

# The function expects:
# - op_node_labelling: a list of 6 operation names (strings)
# - edge_attr: a boolean flag

# Create a valid op_node_labelling with 6 elements
# These should be operation names like 'conv_3x3', 'conv_1x1', 'skip_connect', 'none', etc.
op_node_labelling = [
    'conv_3x3',
    'conv_1x1',
    'skip_connect',
    'avg_pool_3x3',
    'none',
    'conv_3x3'
]

ARGS = {
    'op_node_labelling': op_node_labelling,
    'edge_attr': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DiGraph — a DiGraph output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DiGraph output the recorded ndim 1 and dtype family int 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
encoding_to_nx Ran
ran on a synthesized input
xingchenwan/nasbowl/bayesopt/utils.py
code served (permissive licence) · get_code("1744de6273e76dc3")
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 DiGraph. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-26T22:17:14Z. 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 logging
import networkx as nx

# Test case 1: Simple encoding without color_map
encoding_simple = 'input~maxpool3x3~conv3x3'

# Test case 2: Color map with matching operations
color_map_valid = {
    'input': '#FF0000',
    'maxpool3x3': '#00FF00',
    'conv3x3': '#0000FF'
}

# Test case 3: Color map with missing operation (to test warning)
color_map_partial = {
    'input': '#FF0000',
    'maxpool3x3': '#00FF00'
}

ARGS = {
    'encoding': encoding_simple,
    'color_map': color_map_valid
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DiGraph — a DiGraph output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DiGraph output the recorded ndim 1 and dtype family int describe that coercion, not the output, and are not stated
Environment
image syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); 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
fix Ran
ran on a synthesized input
automl/nas_benchmarks/data_collection/train_fcnet.py
code served (permissive licence) · get_code("82c1defce60e94b6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "epoch": 0,
    "initial_lr": 0.001
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_data Ran
ran on a synthesized input
automl/nas_benchmarks/data_collection/train_fcnet.py
code served (permissive licence) · get_code("25eb8182e13b6c57")
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. 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 os
import numpy as np
import tempfile
import shutil

# Create a temporary directory to hold the mock data files
temp_dir = tempfile.mkdtemp()

# Create mock dataset files for 'protein_structure'
dataset_name = 'protein_structure'

# Create small dummy data with distinct dimensions
train_data = np.random.randn(10, 4).astype(np.float64)
train_targets = np.random.randn(10, 3).astype(np.float64)
valid_data = np.random.randn(5, 4).astype(np.float64)
valid_targets = np.random.randn(5, 3).astype(np.float64)
test_data = np.random.randn(5, 4).astype(np.float64)
test_targets = np.random.randn(5, 3).astype(np.float64)

# Save the mock data to the temporary directory
np.save(os.path.join(temp_dir, f'{dataset_name}_train_data.npy'), train_data)
np.save(os.path.join(temp_dir, f'{dataset_name}_train_targets.npy'), train_targets)
np.save(os.path.join(temp_dir, f'{dataset_name}_valid_data.npy'), valid_data)
np.save(os.path.join(temp_dir, f'{dataset_name}_valid_targets.npy'), valid_targets)
np.save(os.path.join(temp_dir, f'{dataset_name}_test_data.npy'), test_data)
np.save(os.path.join(temp_dir, f'{dataset_name}_test_targets.npy'), test_targets)

ARGS = {
    'path': temp_dir,
    'dataset': 'protein_structure'
}
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 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: none by the harness; any seed is in the setup block
hash_module Ran
ran; our contract draft was wrong, not the code
google-research/nasbench/nasbench/lib/graph_util.py
code served (permissive licence) · get_code("a6de71afa55f4f9b")
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 str. 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 hashlib
import numpy as np

# Create a small valid upper-triangular adjacency matrix
# 4x4 matrix for a directed graph
matrix = np.array([
    [0, 1, 1, 0],
    [0, 0, 1, 1],
    [0, 0, 0, 1],
    [0, 0, 0, 0]
], dtype=np.int32)

# Labels for each vertex (4 vertices, so 4 labels)
labeling = [0, 1, 0, 2]

ARGS = {
    "matrix": matrix,
    "labeling": labeling
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
load_from_pickle Ran
ran on a synthesized input
xingchenwan/nasbowl/loader.py
code served (permissive licence) · get_code("b7eaf69387d2ac0c")
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. 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 pickle
import numpy as np
import tempfile
import os

# Create a minimal pickle file with the expected structure
def create_test_pickle():
    # Create a temporary file to hold the pickle data
    temp_fd, temp_path = tempfile.mkstemp(suffix='.pkl')
    os.close(temp_fd)
    
    # Build test data matching the expected structure
    dump = {
        'model_graph_specs': [
            {
                'adjacency': np.array([
                    [0, 1, 0, 0, 0],
                    [0, 0, 1, 1, 0],
                    [0, 0, 0, 0, 1],
                    [0, 0, 0, 0, 1],
                    [0, 0, 0, 0, 0]
                ], dtype=np.int32),
                'node_labels': ['input', 'conv1x1-bn-relu', 'conv3x3-bn-relu', 'maxpool3x3', 'output']
            },
            {
                'adjacency': np.array([
                    [0, 1, 0, 0],
                    [0, 0, 1, 0],
                    [0, 0, 0, 1],
                    [0, 0, 0, 0]
                ], dtype=np.int32),
                'node_labels': ['input', 'conv3x3-bn-relu', 'conv1x1-bn-relu', 'output']
            }
        ],
        'validation_err': [0.05, 0.08]
    }
    
    with open(temp_path, 'wb') as f:
        pickle.dump(dump, f)
    
    return temp_path

pickle_path = create_test_pickle()

ARGS = {
    'path': pickle_path,
    'n_samples': 2,
    'log_validation': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
networkx, numpy — 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: none by the harness; any seed is in the setup block
nll Ran
ran on a synthesized input
xingchenwan/nasbowl/perf_metrics.py
code served (permissive licence) · get_code("764c6062e2d15820")
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 float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# pred: predictions (array-like, will be reshaped to 1D)
# pred_std: standard deviation of predictions (scalar)
# target: target values (array-like, will be reshaped to 1D)

pred = np.array([1.5, 2.3, 0.8, 3.1])
pred_std = 0.5
target = np.array([1.2, 2.5, 0.9, 3.0])

ARGS = {
    "pred": pred,
    "pred_std": pred_std,
    "target": target
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
normalize_y Ran
ran on a synthesized input
xingchenwan/nasbowl/bayesopt/utils.py
code served (permissive licence) · get_code("cca2009ba2f2a630")
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. 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

torch.manual_seed(0)

# The function takes a single parameter 'y' which can be either a torch.Tensor or numpy array
# It computes mean, std, normalizes, and returns a tuple of (normalized_y, y_mean, y_std)

# Create a small torch tensor for testing
y_tensor = torch.randn(2, 4)

ARGS = {
    "y": y_tensor
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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: none by the harness; any seed is in the setup block
prune Ran
ran on a synthesized input
xingchenwan/nasbowl/bayesopt/generate_test_graphs.py
code served (permissive licence) · get_code("43444e233e672713")
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. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import copy

# Construct a valid adjacency matrix representing a DAG
# The matrix represents connections between vertices in a neural architecture
# Row i, column j = 1 means there's an edge from vertex i to vertex j
# We need at least input (vertex 0) and output (vertex num_vertices-1)

num_vertices = 5
original_matrix = np.array([
    [0, 1, 1, 0, 0],  # vertex 0 (input) connects to 1 and 2
    [0, 0, 0, 1, 0],  # vertex 1 connects to 3
    [0, 0, 0, 1, 0],  # vertex 2 connects to 3
    [0, 0, 0, 0, 1],  # vertex 3 connects to 4
    [0, 0, 0, 0, 0],  # vertex 4 (output) - terminal node
], dtype=np.int8)

# ops is a list of operations, one for each vertex
# The function only accesses ops by index for deletion, so simple strings work
ops = ['input', 'conv1', 'conv2', 'relu', 'output']

ARGS = {
    'original_matrix': original_matrix,
    'ops': ops
}
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 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: none by the harness; any seed is in the setup block
rmse Ran
ran on a synthesized input
xingchenwan/nasbowl/perf_metrics.py
code served (permissive licence) · get_code("fd1cb5746c3cdd8b")
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 float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create simple prediction and target arrays
pred = np.array([1.0, 2.0, 3.0, 4.0])
target = np.array([1.1, 2.1, 2.9, 3.9])

ARGS = {
    "pred": pred,
    "target": target
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
scan_top_valid Ran
ran on a synthesized input
automl/nas_benchmarks/experiment_scripts/run_rl.py
code served (permissive licence) · get_code("992e40612fd29072")
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 list, 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

torch.manual_seed(0)

# Analyze the function: scan_top_valid expects 'data' which is a list of runs.
# Each run is a list where:
# - run[0] is ignored (or used as initialization [0.0, 0.0, 0.0])
# - run[1:] contains elements r where r has:
#   - r[0]: some value (kept as-is)
#   - r[1]: a validation metric (compared to find max)
#   - r[2]: a test metric (carried forward)
# Each r appears to be a list/tuple with at least 3 elements [train_loss, val_acc, test_acc]

# Construct a valid data structure:
# Multiple runs, each run has an initial dummy element and then metric records
data = [
    # Run 1: validation accuracies go 0.5 -> 0.7 -> 0.6 (so we keep 0.7 and its test acc)
    [
        [0.0, 0.0, 0.0],  # Initial placeholder
        [0.1, 0.5, 0.45],  # epoch 1: train_loss=0.1, val_acc=0.5, test_acc=0.45
        [0.08, 0.7, 0.68],  # epoch 2: train_loss=0.08, val_acc=0.7, test_acc=0.68
        [0.06, 0.6, 0.55],  # epoch 3: train_loss=0.06, val_acc=0.6, test_acc=0.55
    ],
    # Run 2: validation accuracies go 0.4 -> 0.65 -> 0.72
    [
        [0.0, 0.0, 0.0],  # Initial placeholder
        [0.15, 0.4, 0.38],  # epoch 1
        [0.09, 0.65, 0.63],  # epoch 2
        [0.05, 0.72, 0.70],  # epoch 3
    ]
]

ARGS = {
    "data": data
}
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 list, ndim 3, dtype family float — a list 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
spearman Ran
ran on a synthesized input
xingchenwan/nasbowl/perf_metrics.py
code served (permissive licence) · get_code("e7c0e64cbd04d215")
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 float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import torch

torch.manual_seed(0)

# Create sample prediction and target arrays
pred = np.array([0.1, 0.2, 0.3, 0.4, 0.5])
target = np.array([0.15, 0.25, 0.35, 0.45, 0.55])

ARGS = {
    "pred": pred,
    "target": target,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
train_and_eval Ran
ran, honoured the contract we drafted
automl/nas_benchmarks/experiment_scripts/run_regularized_evolution.py
code served (permissive licence) · get_code("17e87ce4fcceba76")
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) and returned Tensor, ndim 0, 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 torch.nn as nn

torch.manual_seed(0)

class MockObjective:
    def objective_function(self, config):
        # Returns (y, cost) tuple
        y = torch.tensor(2.5)
        cost = torch.tensor(1.2)
        return y, cost

# Create a mock module 'b' that has the objective_function method
b = MockObjective()

# Inject it into the module namespace so train_and_eval can access it
import sys
current_module = sys.modules[__name__]
setattr(current_module, 'b', b)

config = {
    'learning_rate': 0.001,
    'batch_size': 2,
    'epochs': 10
}

ARGS = {
    'config': config
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 0, dtype family float — no output shape is stated: observe_signature.py could not observe it (fixture:KeyError)
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
get_grad Not yet run
unverified (harvested, no recorded run)
xingchenwan/nasbowl/bayesopt/gp.py
code served (permissive licence) · get_code("3f165eea0619d62c")
How it ran

no invocation recorded.

mean_squared_err Not yet run
unverified (harvested, no recorded run)
automl/nas_benchmarks/data_collection/train_fcnet.py
code served (permissive licence) · get_code("3e9dfc4b41d4c9db")
How it ran

no invocation recorded.

mutate_arch Not yet run
unverified (harvested, no recorded run)
xingchenwan/nasbowl/bayesopt/generate_test_graphs.py
code served (permissive licence) · get_code("9c64a2257fd875c5")
How it ran

no invocation recorded.

objective_function Not yet run
unverified (harvested, no recorded run)
automl/nas_benchmarks/experiment_scripts/run_smac.py
code served (permissive licence) · get_code("47abffc5ae4aba3f")
How it ran

no invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 19 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 1 the same month, 0 later) and 18 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
train_and_eval 1 other paper
1 the same month
2006.04492 (2020-06)
Speedy Performance Estimation for Neural Architecture Search

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

Current neural architecture search (NAS) strategies focus only on finding a single, good, architecture. They offer little insight into why a specific network is performing well, or how we should modify the architecture if we want further improvements. We propose a Bayesian optimisation (BO) approach for NAS that combines the Weisfeiler-Lehman graph kernel with a Gaussian process surrogate. Our method optimises the architecture in a highly data-efficient manner: it is capable of capturing the topological structures of the architectures and is scalable to large graphs, thus making the high-dimensional and graph-like search spaces amenable to BO. More importantly, our method affords interpretability by discovering useful network features and their corresponding impact on the network performance. Indeed, we demonstrate empirically that our surrogate model is capable of identifying useful motifs which can guide the generation of new architectures. We finally show that our method outperforms existing NAS approaches to achieve the state of the art on both closed-and open-domain search spaces.

For agents

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

get_harvested_code_for_paper("2006.07556")
get_code_for_paper("2006.07556")
have("2006.07556")

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

curl https://syntology.ai/api/ran/2006.07556.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.07556.svg)](https://syntology.ai/paper/2006.07556)

Connect an agent — have() is free.