'v0' must have shape {shape}

[desanou]

Since the recent update of scipy library version 1.8.0 (February 5th), I'm encountering a bug in my usual pylearn-parsimony based routine. It happens when fitting a Conesta solver with a non-zero total variation parameter. The error occurred in the svds function of scipy.sparse.linalg, called by nipals.py#343. So I needed to downgrade scipy. I am not sure if I should report the issue here.

[DimitriPapadopoulos]

Could be a SciPy issue. This is not the SciPy site.

Also, you should post an example that reproduces the problem, if at all possible, and at least the complete error messages.

[DimitriPapadopoulos]

The 'v0' must have shape {shape} error message originates in SciPy:

scipy/sparse/linalg/_eigen/_svds.py

        shape = (A.shape[0],) if solver == 'propack' else (min(A.shape),)
        if v0.shape != shape:
            message = "`v0` must have shape {shape}."
            raise ValueError(message)

[DimitriPapadopoulos]

Note that the https://github.com/scipy/scipy/pull/15566 ticket I have opened against SciPy does not address this issue.

[desanou]

Okay, thanks for your feedback. I have a reproducible example but it is not minimal. I'll look into the different libraries a bit more and get back to you if the problem is parsimony related.

I'll close the issue.

[desanou]

@DimitriPapadopoulos I reopen the issue with a minimal example that can produce it. I used the same examples provided by the parsimony-pylearn quick start section.

import numpy as np
np.random.seed(42)
shape = (1, 4, 4)  # Three-dimension matrix
num_samples = 10  # The number of samples
num_ft = shape[0] * shape[1] * shape[2] # The number of features per sample

Randomly generate X

X = np.random.rand(num_samples, num_ft)
beta = np.random.rand(num_ft, 1) # Define beta
# Add noise to y
y = np.dot(X, beta) + 0.001 * np.random.rand(num_samples, 1)
X_train = X[0:6, :]
y_train = y[0:6]
X_test = X[6:10, :]
y_test = y[6:10]

import parsimony.estimators as estimators
import parsimony.algorithms as algorithms
import parsimony.functions.nesterov.tv as tv
l = 0.0  # l1 lasso coefficient
k = 0.0  # l2 ridge regression coefficient
g = 1.0  # tv coefficient
A = tv.linear_operator_from_shape(shape)  # Memory allocation for TV
olstv = estimators.LinearRegressionL1L2TV(l, k, g, A, mu=0.0001,
                                         algorithm=algorithms.proximal.FISTA(),
                                         algorithm_params=dict(max_iter=1000))

res = olstv.fit(X_train, y_train)
print("Estimated beta error = ", np.linalg.norm(olstv.beta - beta))
print("Prediction error = ", np.linalg.norm(olstv.predict(X_test) - y_test))

The code has been compiled with pylearn=parsimony 0.3.1 python 3.8 and scipy last version 1.9.0. I got the error: ValueError: v0 must have shape (16,).

A traceback is given below.

---------------------------------------------------------------------------
**ValueError**                                Traceback (most recent call last)
Input In [3], in <cell line: 1>()
----> 1 res = olstv.fit(X_train, y_train)

File ** .\parsimony\estimators.py:888**, in LinearRegressionL1L2TV.fit(self, X, y, beta)
--> 888 self.beta = self.algorithm.run(function, beta)

File **.\parsimony\algorithms\bases.py:101**, in force_reset.<locals>.wrapper(self, function, *args, **kwargs)
--> 101 return f(self, function, *args, **kwargs)

File **.\parsimony\algorithms\bases.py:81**, in check_compatibility.<locals>.wrapper(self, function, *args, **kwargs)
---> 81     return f(self, function, *args, **kwargs)

File **.\parsimony\algorithms\proximal.py:401**, in FISTA.run(self, function, beta)
--> 401 step = function.step(z)

File **.\parsimony\functions\combinedfunctions.py:1044**, in LinearRegressionL1L2TV.step(self, x, **kwargs)
-> 1044     return 1.0 / self.L()

File **.\parsimony\functions\combinedfunctions.py:769**, in LinearRegressionL1L2TV.L(self)
--> 769          + self.tv.L()

File **.\parsimony\functions\nesterov\tv.py:156**, in TotalVariation.L(self)
--> 156 lmaxA = self.lambda_max()

File **.\parsimony\functions\nesterov\tv.py:189**, in TotalVariation.lambda_max(self)
--> 189 v = RankOneSparseSVD().run(A)  # , max_iter=max_iter)

File **.\parsimony\algorithms\nipals.py:343**, in RankOneSparseSVD.run(self, X, start_vector)
--> 343     [_, _, v] = sparse_linalg.svds(X, k=1, v0=v0,

File **.\scipy\sparse\linalg\_eigen\_svds.py:252**, in svds(A, k, ncv, tol, which, v0, maxiter, return_singular_vectors, solver, random_state, options)
--> 252 args = _iv(A, k, ncv, tol, which, v0, maxiter, return_singular_vectors,

File **.\scipy\sparse\linalg\_eigen\_svds.py:84**, in _iv(A, k, ncv, tol, which, v0, maxiter, return_singular, solver, random_state)
---> 84         raise ValueError(message)

**ValueError**: `v0` must have shape (16,).

Using the scipy version 1.7.1 and the relevant numpy version does not raised an error. The error might be due to update of scipy > 1.7.1.

[DimitriPapadopoulos]

The error message had been introduced in SciPy 1.8.0rc1 by https://github.com/scipy/scipy/pull/14433.

v0 is initialised here, it would be great if you could print the value of X.shape: https://github.com/neurospin/pylearn-parsimony/blob/969ea182573dddf8cd70e93846613713598e3afa/parsimony/algorithms/nipals.py#L151

[desanou]

X.shape (10, 16)

[DimitriPapadopoulos]

According to the documentation of scipy.sparse.linalg.svds, the default solver, which we use here as we do not pass a solver argument, is ‘arpack’:

solver{‘arpack’, ‘propack’, ‘lobpcg’}, optional The solver used. ‘arpack’, ‘lobpcg’, and ‘propack’ are supported. Default: ‘arpack’.

According to the documentation of svds(solver=’arpack’):

In the descriptions below, let M, N = A.shape.

v0ndarray, optional The starting vector for iteration: an (approximate) left singular vector if N > M and a right singular vector otherwise. Must be of length min(M, N). Default: random

Because in our case we have M = 10 and N = 16. I would expect the length of v0 to be 10, and hence its shape (10,). ⟹ Could you perhaps print v0.shape?

Then, I don't understand why the input validation function _iv() of svds() raises this error where v0 must have shape (16,).