SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2209.02816 · 2022

A probabilistic framework for task-aligned intra- and inter-area neural manifold estimation

arXiv · PDF · Open in the Atlas

Code that ran

11 of 12 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 11 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/2209.02816.json · badge: /api/ran/2209.02816.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
balzaniedoardo/tame-gp canonical (classifier label)
classifier basis not recorded on the link
11 ran · 1 no recorded run
FunctionStatusWhere it lives
GPLogLike Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/inference.py
code served (permissive licence) · get_code("37c9910bd5d7a1ee")
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

# GPLogLike expects:
# z: TxK latent factor (2D array)
# Kinv: KT x KT precision matrix (2D array)

# Let's use T=2 (time steps), K=4 (latent dimensions)
T = 2
K = 4

# z is TxK
z = np.random.randn(T, K)

# Kinv is KT x KT (so 8x8 in this case)
# Make it a valid precision matrix (symmetric positive definite)
KT = K * T
# Create a random matrix and make it symmetric positive definite
A = np.random.randn(KT, KT)
Kinv = A @ A.T + np.eye(KT)  # Ensure positive definiteness

ARGS = {
    "z": z,
    "Kinv": Kinv,
}
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
MStepGauss Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/learnGaussianParam.py
code served (permissive licence) · get_code("3018164749c26e46")
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

# Set random seed for reproducibility
np.random.seed(0)

# Define dimensions
T = 5  # number of time steps (sum of all trial lengths)
N = 3  # observation dimension
K = 2  # latent dimension

# Create x1: Gaussian observations for all trials
# Shape: (T, N) = (5, 3)
x1 = np.random.randn(T, N)

# Create mean_post: posterior mean for all trials
# Shape: (T, K) = (5, 2)
mean_post = np.random.randn(T, K)

# Create cov_post: posterior covariance
# Shape: (T, K, K) = (5, 2, 2)
cov_post = np.random.randn(T, K, K)
# Make each covariance matrix positive semi-definite by using outer product
for i in range(T):
    cov_post[i] = np.dot(cov_post[i], cov_post[i].T)

ARGS = {
    "x1": x1,
    "mean_post": mean_post,
    "cov_post": cov_post,
}
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
block_inv Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/data_processing_tools.py
code served (permissive licence) · get_code("e30898c47a64f3a0")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family float, shape (2, 4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
from numba import jit

# Set random seed for reproducibility
np.random.seed(0)

# Create a small batch of invertible square matrices
# Shape: (batch_size, matrix_dim, matrix_dim)
batch_size = 2
matrix_dim = 4

# Generate random matrices and ensure they're invertible by using QR decomposition
A_list = []
for _ in range(batch_size):
    # Generate random matrix and decompose to get orthogonal matrix (always invertible)
    random_matrix = np.random.randn(matrix_dim, matrix_dim)
    Q, R = np.linalg.qr(random_matrix)
    A_list.append(Q.astype(np.float64))

A = np.stack(A_list, axis=0)

# Verify shapes and invertibility
assert A.shape == (batch_size, matrix_dim, matrix_dim), f"Expected shape (2, 4, 4), got {A.shape}"
assert A.dtype == np.float64, f"Expected dtype float64, got {A.dtype}"

# Verify each matrix is invertible (non-zero determinant)
for t in range(A.shape[0]):
    det = np.linalg.det(A[t])
    assert abs(det) > 1e-10, f"Matrix {t} is singular with determinant {det}"

ARGS = {
    "A": A
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numba, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (2, 4, 4) — shape (2, 4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compileTrialStackedObsAndLatent Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/data_processing_tools.py
code served (permissive licence) · get_code("9ca8df1a25d01713")
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

class PosteriorInf:
    def __init__(self, T_tr, K0, K1, idx_latent):
        self.mean = {
            0: np.random.randn(K0, T_tr),
            idx_latent: np.random.randn(K1, T_tr)
        }
        self.cov_t = {
            0: np.random.randn(T_tr, K0, K0),
            idx_latent: np.random.randn(T_tr, K1, K1)
        }
        self.cross_cov_t = {
            idx_latent: np.random.randn(T_tr, K0, K1)
        }

class Data:
    def __init__(self, trial_durs, K0, K1, idx_latent, xDim):
        self.trialDur = trial_durs
        self.posterior_inf = {}
        self.observations = {}
        
        for tr_idx, T_tr in enumerate(trial_durs):
            self.posterior_inf[tr_idx] = PosteriorInf(T_tr, K0, K1, idx_latent)
            obs_list = [None] * (idx_latent + 1)
            obs_list[idx_latent - 1] = np.random.randn(T_tr, xDim)
            self.observations[tr_idx] = (None, obs_list)
    
    def get_observations(self, tr):
        return self.observations[tr]

K0 = 3
K1 = 2
idx_latent = 2
xDim = 4
trial_durs = [5, 3]
T = sum(trial_durs)
trial_list = [0, 1]

data = Data(trial_durs, K0, K1, idx_latent, xDim)

ARGS = {
    "data": data,
    "idx_latent": idx_latent,
    "trial_list": trial_list,
    "T": T,
    "xDim": xDim,
    "K0": K0,
    "K1": K1
}
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
fast_stackCSRHes_memoryPreAllocation Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/data_processing_tools.py
code served (permissive licence) · get_code("77dc3857c3b4d606")
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 torch

class MockCSRMatrix:
    """Mock CSR matrix class with necessary attributes"""
    def __init__(self, values, rowptrs, colinds, nrows, ncols, nnz):
        self.values = values
        self.rowptrs = rowptrs
        self.colinds = colinds
        self.nrows = nrows
        self.ncols = ncols
        self.nnz = nnz

# Create mock CSR matrices
torch.manual_seed(0)

# Create a small hessian matrix for new trial
# 3x3 matrix with 4 non-zero elements
new_hes_values = np.array([1.5, 2.3, 0.8, 1.2], dtype=np.float64)
new_hes_rowptrs = np.array([0, 2, 3, 4], dtype=np.int32)
new_hes_colinds = np.array([0, 2, 1, 2], dtype=np.int32)
newHes = MockCSRMatrix(
    values=new_hes_values,
    rowptrs=new_hes_rowptrs,
    colinds=new_hes_colinds,
    nrows=3,
    ncols=3,
    nnz=4
)

# Pre-allocated arrays for stacking (assume space for 2 matrices)
vals = np.zeros(10, dtype=np.float64)
rowsptr = np.zeros(10, dtype=np.int32)
colindices = np.zeros(10, dtype=np.int32)

# Initial state - first matrix already placed
# Assume first matrix has 3 rows, 4 nnz
vals[0:4] = np.array([0.5, 1.0, 1.5, 0.2], dtype=np.float64)
rowsptr[0:4] = np.array([0, 2, 3, 4], dtype=np.int32)
colindices[0:4] = np.array([0, 1, 1, 2], dtype=np.int32)

# Parameters for the function
nnz = 4
nrows = 3
ncols = 3
i0PTR = 3  # pointer offset for next block (after first matrix's row pointers)
i0Val = 4  # value offset for next block
sumK = 2   # number of trials (or similar)

ARGS = {
    'vals': vals,
    'rowsptr': rowsptr,
    'colindices': colindices,
    'nnz': nnz,
    'nrows': nrows,
    'ncols': ncols,
    'i0PTR': i0PTR,
    'i0Val': i0Val,
    'newHes': newHes,
    'sumK': sumK
}
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
grad_GPLogLike Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/inference.py
code served (permissive licence) · get_code("afc943a0bf54fdc6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family float, shape (12,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct valid arguments for grad_GPLogLike
# The function expects:
# - z: a 2D array that will be transposed and reshaped to 1D
# - Kinv: an inverse covariance matrix to multiply with the flattened z

# Create z as a 2D array (e.g., 3x4 shape)
z = np.array([[1.0, 2.0, 3.0, 4.0],
              [5.0, 6.0, 7.0, 8.0],
              [9.0, 10.0, 11.0, 12.0]])

# Create Kinv as a square matrix that matches the flattened size of z
# z.T.reshape(-1) will have size 12 (3*4), so Kinv should be 12x12
Kinv = np.random.randn(12, 12)

ARGS = {
    "z": z,
    "Kinv": Kinv
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (12,) — shape (12,) 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
grad_Poisson_ithunit Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/learnPoissonParam.py
code served (permissive licence) · get_code("300256c468a22a4b")
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

# Set random seed for reproducibility
np.random.seed(0)

# Define dimensions
xdim = 4
xdim1 = 3
batch_size = 2

# C: shape (xdim,) - a vector of size 4
C = np.random.randn(xdim)

# C1: shape (xdim1,) - a vector of size 3
C1 = np.random.randn(xdim1)

# d: scalar
d = np.random.randn()

# mean_post: shape (batch_size, xdim + xdim1) - posterior mean
# The function uses mean_post[:, :xdim] and mean_post[:, xdim:] and also einsum 'j,tj->t' with Cout
mean_post = np.random.randn(batch_size, xdim + xdim1)

# cov_post: shape (batch_size, xdim + xdim1, xdim + xdim1) - posterior covariance
# The function uses cov_post[:, :xdim, :xdim], cov_post[:, :xdim, xdim:], cov_post[:, xdim:, xdim:], cov_post[:, xdim:, :xdim]
# and also reshapes it to (batch_size, (xdim + xdim1)^2)
cov_post = np.random.randn(batch_size, xdim + xdim1, xdim + xdim1)

ARGS = {
    'C': C,
    'C1': C1,
    'd': d,
    'mean_post': mean_post,
    'cov_post': cov_post
}
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
grad_expectedLLPoisson Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/learnPoissonParam.py
code served (permissive licence) · get_code("86af6b6c728cd3c8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family float, shape (12,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Define dimensions
ydim = 2      # output dimension
xdim = 3      # latent dimension
xdim1 = 2     # additional latent dimension (for C1)
T = 4         # time steps

# Create x: observed data, shape (T, ydim)
x = np.random.poisson(2, size=(T, ydim)).astype(float)

# Create C: weight matrix, shape (ydim, xdim)
C = np.random.randn(ydim, xdim)

# Create C1: additional weight matrix, shape (ydim, xdim1)
C1 = np.random.randn(ydim, xdim1)

# Create d: bias term, shape (ydim,)
d = np.random.randn(ydim)

# Create mean_post: posterior mean, shape (T, xdim + xdim1)
mean_post = np.random.randn(T, xdim + xdim1)

# Create cov_post: posterior covariance, shape (T, xdim + xdim1, xdim + xdim1)
cov_post = np.random.randn(T, xdim + xdim1, xdim + xdim1)
# Make covariance symmetric positive semi-definite
cov_post = np.einsum('tij,tkj->tik', cov_post, cov_post)

ARGS = {
    'x': x,
    'C': C,
    'd': d,
    'mean_post': mean_post,
    'cov_post': cov_post,
    'C1': C1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (12,) — shape (12,) 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
grag_GaussLL_wrt_Wd Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/learnGaussianParam.py
code served (permissive licence) · get_code("080e1e0050cc6ef8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Define dimensions
T = 2  # number of time steps
d_x = 4  # dimension of x
d_z = 3  # dimension of latent z

# x1: shape (T, d_x) - observations
x1 = np.random.randn(T, d_x)

# W1: shape (d_x, d_z) - weight matrix
W1 = np.random.randn(d_x, d_z)

# d1: shape (d_x,) - bias vector
d1 = np.random.randn(d_x)

# Rinv: shape (d_x, d_x) - inverse covariance (not used in function but required)
Rinv = np.eye(d_x)

# mean_post: shape (T, d_z) - posterior mean
mean_post = np.random.randn(T, d_z)

# cov_post: shape (T, d_z, d_z) - posterior covariance
cov_post = np.array([np.eye(d_z) for _ in range(T)])

ARGS = {
    'x1': x1,
    'W1': W1,
    'd1': d1,
    'Rinv': Rinv,
    'mean_post': mean_post,
    'cov_post': cov_post
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (4, 4) — shape (4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
hess_GPLogLike Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/inference.py
code served (permissive licence) · get_code("752cd888d54fd526")
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 Tensor, ndim 2, dtype family float, shape (4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# The function hess_GPLogLike takes two parameters:
# - z: appears unused in the function body
# - Kinv: a matrix that gets negated and returned
# 
# Based on the docstring mentioning "GP prior" and "Hessian",
# Kinv is likely an inverse covariance matrix (symmetric, positive definite)
# z is likely latent variables or inducing points

# Create a small inverse covariance matrix (Kinv)
# Make it 4x4 symmetric positive definite
Kinv = torch.tensor([
    [2.0, 0.5, 0.1, 0.2],
    [0.5, 3.0, 0.3, 0.1],
    [0.1, 0.3, 2.5, 0.4],
    [0.2, 0.1, 0.4, 1.8]
], dtype=torch.float32)

# Create z - just a vector or matrix of latent variables
z = torch.randn(2, 4, dtype=torch.float32)

ARGS = {
    "z": z,
    "Kinv": Kinv,
}
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 Tensor, ndim 2, dtype family float, shape (4, 4) — shape (4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
logLike_Gauss Ran
ran on a synthesized input
balzaniedoardo/tame-gp/core/learnGaussianParam.py
code served (permissive licence) · get_code("1587807b549bfccc")
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

# Set seed for reproducibility
np.random.seed(0)

# Define dimensions with distinct sizes for clarity
T = 2        # time/batch dimension
D = 4        # data/feature dimension
K = 3        # latent dimension
M = 5        # output dimension

# x1: shape (T, D) - observations
x1 = np.random.randn(T, D)

# W1: shape (D, K) - weight matrix
W1 = np.random.randn(D, K)

# d1: shape (D,) - bias/offset vector
d1 = np.random.randn(D)

# Rinv: can be 1D or 2D
# Using 1D version (diagonal precision matrix)
Rinv = np.abs(np.random.randn(D)) + 0.1  # ensure positive for log

# mean_post: shape (T, K) - posterior mean
mean_post = np.random.randn(T, K)

# cov_post: shape (T, K) - posterior covariance (diagonal elements)
cov_post = np.abs(np.random.randn(T, K)) + 0.1

ARGS = {
    "x1": x1,
    "W1": W1,
    "d1": d1,
    "Rinv": Rinv,
    "mean_post": mean_post,
    "cov_post": cov_post
}
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
expectedLLPoisson Not yet run
unverified (harvested, no recorded run)
balzaniedoardo/tame-gp/core/learnPoissonParam.py
code served (permissive licence) · get_code("e1f68ee82159b776")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 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.

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

Latent manifolds provide a compact characterization of neural population activity and of shared co-variability across brain areas. Nonetheless, existing statistical tools for extracting neural manifolds face limitations in terms of interpretability of latents with respect to task variables, and can be hard to apply to datasets with no trial repeats. Here we propose a novel probabilistic framework that allows for interpretable partitioning of population variability within and across areas in the context of naturalistic behavior. Our approach for task aligned manifold estimation (TAME-GP) extends a probabilistic variant of demixed PCA by (1) explicitly partitioning variability into private and shared sources, (2) using a Poisson noise model, and (3) introducing temporal smoothing of latent trajectories in the form of a Gaussian Process prior. This TAME-GP graphical model allows for robust estimation of task-relevant variability in local population responses, and of shared co-variability between brain areas. We demonstrate the efficiency of our estimator on within model and biologically motivated simulated data. We also apply it to neural recordings in a closed-loop virtual navigation task in monkeys, demonstrating the capacity of TAME-GP to capture meaningful intra- and inter-area neural variability with single trial resolution.

For agents

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

get_harvested_code_for_paper("2209.02816")
get_code_for_paper("2209.02816")
have("2209.02816")

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

curl https://syntology.ai/api/ran/2209.02816.json

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

[![Syntology run record](https://syntology.ai/api/ran/2209.02816.svg)](https://syntology.ai/paper/2209.02816)

Connect an agent — have() is free.