SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2007.09390 · ICML · 2020

Autoregressive flow-based causal discovery and inference

Ilyes Khemakhem, Aapo Hyvärinen, Ricardo Pio Monti

arXiv · PDF · Open in the Atlas

Code that ran

11 of 15 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (10 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 10 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/2007.09390.json · badge: /api/ran/2007.09390.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
piomonti/carefl community
Papers with Code archive
11 ran · 4 no recorded run
FunctionStatusWhere it lives
ARMLP Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("b793bc8bea09122a")
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, ARMLP: 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 numpy as np

torch.manual_seed(0)

ARGS = {
    'nin': 4,
    'nout': 8,
    'nh': 3,
}
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 ARMLP — 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 ARMLP 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
AffineCL Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("ee35a766a467c7ab")
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, AffineCL: 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
from torch import nn

torch.manual_seed(0)

class MLP4(nn.Module):
    """ a simple 4-layer MLP """

    def __init__(self, nin, nout, nh):
        super().__init__()
        self.net = nn.Sequential(nn.Linear(nin, nh), nn.LeakyReLU(0.2), nn.Linear(nh, nh), nn.LeakyReLU(0.2), nn.Linear(nh, nh), nn.LeakyReLU(0.2), nn.Linear(nh, nout))

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

ARGS = {
    "dim": 4,
    "parity": False,
    "net_class": MLP4,
    "nh": 24,
    "scale": True,
    "shift": True,
    "scale_base": False,
    "shift_base": False,
    "checkerboard": 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 AffineCL — 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 AffineCL 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
CustomSyntheticDatasetDensity Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("da6d3d9a336b2a9c")
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, CustomSyntheticDatasetDensity: 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 numpy as np

torch.manual_seed(0)

# Create a small synthetic dataset
X = np.random.randn(2, 4).astype(np.float32)

ARGS = {
    'X': X,
    'device': 'cpu'
}
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 CustomSyntheticDatasetDensity — 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 CustomSyntheticDatasetDensity output the recorded ndim 2 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
MADE Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("03e61cb03e4a473b")
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, MADE: 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 numpy as np

torch.manual_seed(0)

ARGS = {
    'nin': 4,
    'hidden_sizes': [8, 8],
    'nout': 8,
    'num_masks': 1,
    'natural_ordering': False,
}
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 MADE — 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 MADE 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
MAF Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("49de88957949ca25")
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, MAF: 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 numpy as np

torch.manual_seed(0)

class ARMLP(nn.Module):
    """ a 4-layer auto-regressive MLP, wrapper around MADE net """

    def __init__(self, nin, nout, nh):
        super().__init__()
        self.net = MADE(nin, [nh, nh, nh], nout, num_masks=1, natural_ordering=True)

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

class MADE(nn.Module):

    def __init__(self, nin, hidden_sizes, nout, num_masks=1, natural_ordering=False):
        super().__init__()
        self.nin = nin
        self.nout = nout
        self.hidden_sizes = hidden_sizes
        assert self.nout % self.nin == 0, 'nout must be integer multiple of nin'
        self.net = []
        hs = [nin] + hidden_sizes + [nout]
        for h0, h1 in zip(hs, hs[1:]):
            self.net.extend([MaskedLinear(h0, h1), nn.ReLU()])
        self.net.pop()
        self.net = nn.Sequential(*self.net)
        self.natural_ordering = natural_ordering
        self.num_masks = num_masks
        self.seed = 0
        self.m = {}
        self.update_masks()

    def update_masks(self):
        if self.m and self.num_masks == 1:
            return
        L = len(self.hidden_sizes)
        rng = np.random.RandomState(self.seed)
        self.seed = (self.seed + 1) % self.num_masks
        self.m[-1] = np.arange(self.nin) if self.natural_ordering else rng.permutation(self.nin)
        for l in range(L):
            self.m[l] = rng.randint(self.m[l - 1].min(), self.nin - 1, size=self.hidden_sizes[l])
        masks = [self.m[l - 1][:, None] <= self.m[l][None, :] for l in range(L)]
        masks.append(self.m[L - 1][:, None] < self.m[-1][None, :])
        if self.nout > self.nin:
            k = int(self.nout / self.nin)
            masks[-1] = np.concatenate([masks[-1]] * k, axis=1)
        layers = [l for l in self.net.modules() if isinstance(l, MaskedLinear)]
        for l, m in zip(layers, masks):
            l.set_mask(m)

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

class MaskedLinear(nn.Linear):
    """ same as Linear except has a configurable mask on the weights """

    def __init__(self, in_features, out_features, bias=True):
        super().__init__(in_features, out_features, bias)
        self.register_buffer('mask', torch.ones(out_features, in_features))

    def set_mask(self, mask):
        self.mask.data.copy_(torch.from_numpy(mask.astype(np.uint8).T))

    def forward(self, input):
        import torch.nn.functional as F
        return F.linear(input, self.mask * self.weight, self.bias)

class MAF(nn.Module):
    """ Masked Autoregressive Flow that uses a MADE-style network for fast forward """

    def __init__(self, dim, parity, net_class=ARMLP, nh=24):
        super().__init__()
        self.dim = dim
        self.net = net_class(dim, dim * 2, nh)
        self.parity = parity

    def forward(self, x):
        st = self.net(x)
        s, t = st.split(self.dim, dim=1)
        z = x * torch.exp(s) + t
        z = z.flip(dims=(1,)) if self.parity else z
        log_det = torch.sum(s, dim=1)
        return (z, log_det)

    def backward(self, z):
        x = torch.zeros_like(z)
        log_det = torch.zeros(z.size(0))
        z = z.flip(dims=(1,)) if self.parity else z
        for i in range(self.dim):
            st = self.net(x.clone())
            s, t = st.split(self.dim, dim=1)
            x[:, i] = (z[:, i] - t[:, i]) * torch.exp(-s[:, i])
            log_det += -s[:, i]
        return (x, log_det)

ARGS = {
    'dim': 4,
    'parity': True,
    'net_class': ARMLP,
    'nh': 24,
}
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 MAF — 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 MAF 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
MLP1layer Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("756d26a591b1b7cf")
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, MLP1layer: 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 MLP1layer.__init__ takes nin, nout, nh as parameters
# nin: input dimension
# nout: output dimension
# nh: hidden dimension

# These should be distinct sizes for probing
nin = 4
nout = 3
nh = 8

ARGS = {
    'nin': nin,
    'nout': nout,
    'nh': nh
}
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 MLP1layer — 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 MLP1layer 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
MLP4 Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("2eefac96b1b3db97")
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, MLP4: 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)

# MLP4.__init__ takes nin, nout, nh
# nin: input dimension
# nout: output dimension
# nh: hidden dimension

ARGS = {
    "nin": 4,
    "nout": 3,
    "nh": 8,
}
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 MLP4 — 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 MLP4 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
MaskedLinear Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("ed7abf1170b542ee")
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, MaskedLinear: 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 numpy as np

torch.manual_seed(0)

ARGS = {
    'in_features': 4,
    'out_features': 3,
    'bias': True,
}
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 MaskedLinear — 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 MaskedLinear 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
NormalizingFlow Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("003866981408459c")
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, NormalizingFlow: 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
from torch import nn

torch.manual_seed(0)

# Create a minimal flow module that satisfies the interface
class DummyFlow(nn.Module):
    def __init__(self, dim):
        super().__init__()
        self.dim = dim
        self.weight = nn.Parameter(torch.randn(dim, dim))
    
    def forward(self, x):
        # x shape: (batch, dim)
        # Return transformed x and log determinant for each sample
        batch_size = x.shape[0]
        transformed = torch.matmul(x, self.weight)
        # Simple log_det: sum of log absolute diagonal elements
        log_det = torch.ones(batch_size) * torch.log(torch.abs(self.weight.diag()).sum())
        return transformed, log_det
    
    def backward(self, z):
        # z shape: (batch, dim)
        # Inverse transform
        batch_size = z.shape[0]
        try:
            weight_inv = torch.linalg.inv(self.weight)
            transformed = torch.matmul(z, weight_inv)
            log_det = torch.ones(batch_size) * torch.log(torch.abs(weight_inv.diag()).sum())
        except:
            # Fallback if singular
            transformed = torch.matmul(z, self.weight.T)
            log_det = torch.ones(batch_size) * torch.log(torch.abs(self.weight.diag()).sum())
        return transformed, log_det

# Create flows
flow1 = DummyFlow(dim=4)
flow2 = DummyFlow(dim=4)
flows = [flow1, flow2]

ARGS = {
    "flows": flows
}
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 NormalizingFlow — 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 NormalizingFlow 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
NormalizingFlowModel Ran
ran on a synthesized input · constructed an object; no result was computed
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("332907cc3fc9bc91")
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, NormalizingFlowModel: 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 numpy as np

torch.manual_seed(0)

# Create a simple flow module that satisfies the interface
class SimpleFlow(nn.Module):
    def __init__(self, dim):
        super().__init__()
        self.dim = dim
        self.scale = nn.Parameter(torch.ones(dim))
    
    def forward(self, x):
        # x: (batch, dim)
        # Returns transformed x and log determinant
        m = x.shape[0]
        log_det = torch.sum(torch.log(torch.abs(self.scale))).unsqueeze(0).expand(m)
        z = x * self.scale
        return z, log_det
    
    def backward(self, z):
        # z: (batch, dim)
        # Returns inverse transformed z and log determinant
        m = z.shape[0]
        log_det = -torch.sum(torch.log(torch.abs(self.scale))).unsqueeze(0).expand(m)
        x = z / self.scale
        return x, log_det

# Create a simple prior that satisfies the interface
class SimplePrior(nn.Module):
    def __init__(self, dim):
        super().__init__()
        self.dim = dim
    
    def log_prob(self, z):
        # z: (batch, dim)
        # Returns log probability for each sample
        # Standard normal log prob: -0.5 * sum(z^2) - 0.5 * dim * log(2*pi)
        return -0.5 * torch.sum(z ** 2, dim=1, keepdim=True)
    
    def sample(self, shape):
        # shape: tuple like (num_samples,)
        # Returns samples of shape (num_samples, dim)
        return torch.randn(*shape, self.dim)

# Create instances
input_dim = 4
prior = SimplePrior(dim=input_dim)
flows = [SimpleFlow(dim=input_dim), SimpleFlow(dim=input_dim)]

ARGS = {
    "prior": prior,
    "flows": flows,
}
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 NormalizingFlowModel — 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 NormalizingFlowModel 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
unconstrained_RQS Ran
ran; our contract draft was wrong, not the code
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("af7c19b8eaf6e4a4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 2, 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.functional as F
import numpy as np

torch.manual_seed(0)

# Create small tensors with distinct dimensions
batch_size = 2
num_bins = 4

# inputs: shape (batch_size,) - values between -tail_bound and tail_bound
inputs = torch.tensor([-0.5, 0.3])

# unnormalized_widths: shape (batch_size, num_bins)
unnormalized_widths = torch.randn(batch_size, num_bins)

# unnormalized_heights: shape (batch_size, num_bins)
unnormalized_heights = torch.randn(batch_size, num_bins)

# unnormalized_derivatives: shape (batch_size, num_bins)
unnormalized_derivatives = torch.randn(batch_size, num_bins)

ARGS = {
    'inputs': inputs,
    'unnormalized_widths': unnormalized_widths,
    'unnormalized_heights': unnormalized_heights,
    'unnormalized_derivatives': unnormalized_derivatives,
    'inverse': False,
    'tail_bound': 1.0,
    'min_bin_width': 0.001,
    'min_bin_height': 0.001,
    'min_derivative': 0.001,
}
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, ndim 2, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
CAREFL Not yet run
unverified (harvested, no recorded run)
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("c201e2ca52b456ff")
How it ran

no invocation recorded.

NSF_AR Not yet run
unverified (harvested, no recorded run)
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("87c69bdae0f2ca05")
How it ran

no invocation recorded.

RQS Not yet run
unverified (harvested, no recorded run)
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("db027a8f670e12d3")
How it ran

no invocation recorded.

searchsorted Not yet run
unverified (harvested, no recorded run)
piomonti/carefl/models/carefl.py
code served (permissive licence) · get_code("e85d747983c4f5d8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 14 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
searchsorted
brick of this paper's assembly: helper
9 other papers
9 later · 1 only as an assembly brick
2106.06103 (2021-06)
Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech
2206.04658 (2022-06)
BigVGAN: A Universal Neural Vocoder with Large-Scale Training
2210.06370 (2022-10)
Can we use Common Voice to train a Multi-Speaker TTS system?
and 6 more not named here

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

We posit that autoregressive flow models are wellsuited to performing a range of causal inference tasks -ranging from causal discovery to making interventional and counterfactual predictions. In particular, we exploit the fact that autoregressive architectures define an ordering over variables, analogous to a causal ordering, in order to propose a single flow architecture to perform all three aforementioned tasks. We first leverage the fact that flow models estimate normalized log-densities of data to derive a bivariate measure of causal direction based on likelihood ratios. Whilst traditional measures of causal direction often require restrictive assumptions on the nature of causal relationships (e.g., linearity), the flexibility of flow models allows for arbitrary causal dependencies. Our approach compares favorably against alternative methods on synthetic data as well as on the Cause-Effect Pairs benchmark dataset. Subsequently, we demonstrate that the invertible nature of flows naturally allows for direct evaluation of both interventional and counterfactual predictions, which require marginalization and conditioning over latent variables respectively. We present examples over synthetic data where autoregressive flows, when trained under the correct causal ordering, are able to make accurate interventional and counterfactual predictions.

For agents

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

get_harvested_code_for_paper("2007.09390")
get_code_for_paper("2007.09390")
have("2007.09390")

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

curl https://syntology.ai/api/ran/2007.09390.json

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

[![Syntology run record](https://syntology.ai/api/ran/2007.09390.svg)](https://syntology.ai/paper/2007.09390)

Connect an agent — have() is free.