cktest bug?

[euhruska]

I updated my conda environment and now getting an error with cktest something about a flag of the array

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-103-4e4c5426a49a> in <module>
----> 1 ck=msm_object.cktest(3, mlags=4)

/scratch1/eh22/conda/envs/py34/lib/python3.7/site-packages/pyemma/msm/estimators/_msm_estimator_base.py in cktest(self, nsets, memberships, mlags, conf, err_est, n_jobs, show_progress)
    846         ck = ChapmanKolmogorovValidator(self, self, memberships, mlags=mlags, conf=conf,
    847                                         n_jobs=n_jobs, err_est=err_est, show_progress=show_progress)
--> 848         ck.estimate(self._dtrajs_full)
    849         return ck

/scratch1/eh22/conda/envs/py34/lib/python3.7/site-packages/pyemma/_base/estimator.py in estimate(self, X, **params)
    405         if params:
    406             self.set_params(**params)
--> 407         self._model = self._estimate(X)
    408         # ensure _estimate returned something
    409         assert self._model is not None

/scratch1/eh22/conda/envs/py34/lib/python3.7/site-packages/pyemma/msm/estimators/lagged_model_validators.py in _estimate(self, data)
    167             model = estimated_models[i]
    168             estimator = estimators[i]
--> 169             self._est.append(self._compute_observables(model, estimator))
    170             if self.has_errors and self.err_est:
    171                 l, r = self._compute_observables_conf(model, estimator)

/scratch1/eh22/conda/envs/py34/lib/python3.7/site-packages/pyemma/msm/estimators/lagged_model_validators.py in _compute_observables(self, model, estimator, mlag)
    465             p0sub = p0[subset]  # map distribution to new active set
    466             p0sub /= p0sub.sum()  # renormalize
--> 467             pksub = model.propagate(p0sub, mlag)
    468             for j in range(self.nsets):
    469                 pk_on_set[i, j] = np.dot(pksub, self.memberships[subset, j])  # map onto set

/scratch1/eh22/conda/envs/py34/lib/python3.7/site-packages/pyemma/msm/models/msm.py in propagate(self, p0, k)
    485                 pk = _np.dot(pk.T, self.transition_matrix)
    486         else:  # dense: employ eigenvalue decomposition
--> 487             self._ensure_eigendecomposition(self.nstates)
    488             from pyemma.util.linalg import mdot
    489             pk = mdot(p0.T,

/scratch1/eh22/conda/envs/py34/lib/python3.7/site-packages/pyemma/msm/models/msm.py in _ensure_eigendecomposition(self, neig)
    357                 # not enough eigenpairs present
    358                 # or eigendecomposition computed for an outdated transition matrix - recompute.
--> 359                 self._compute_eigendecomposition(neig)
    360         except AttributeError:
    361             # no eigendecomposition yet - compute:

/scratch1/eh22/conda/envs/py34/lib/python3.7/site-packages/pyemma/msm/models/msm.py in _compute_eigendecomposition(self, neig)
    320         """ Conducts the eigenvalue decomposition and stores k eigenvalues, left and right eigenvectors """
    321         from msmtools.analysis import rdl_decomposition
--> 322         self._p_id = _hash_numpy_array(self.transition_matrix)
    323 
    324         if self.reversible:

/scratch1/eh22/conda/envs/py34/lib/python3.7/site-packages/pyemma/util/numeric.py in _hash_numpy_array(x)
     45         v = hashlib.sha1(x)
     46     finally:
---> 47         x.flags.writeable = old
     48     hash_value = hash((x.shape, x.strides, v.digest()))
     49 

ValueError: cannot set WRITEABLE flag to True of this array

pyemma 2.5.5 numpy 1.16.2, python 3.7.2
msm_object

MaximumLikelihoodMSM(connectivity='largest', count_mode='sliding',
           dt_traj='1 step', lag=110, maxerr=1e-08, maxiter=1000000,
           mincount_connectivity='1/n', reversible=True, score_k=10,
           score_method='VAMP2', sparse=False, statdist_constraint=None)

[marscher]

can you confirm downgrading numpy to 1.15 solves the issue?

[euhruska]

confirmed

[marscher]

Thanks for testing an reporting this. I have to dig deeper to solve this. Am 27. März 2019 17:46:25 MEZ schrieb Eugen Hruska <notifications@github.com>:

…

confirmed