Recover Dynamics ValueError

Hi Volker, I tried to use the very new dynamical model on one of my datasets. I get an ValueError after 73% of the recovering dynamics step.

Traceback:

ValueError                                Traceback (most recent call last)
<ipython-input-8-5a274db81b65> in <module>
     89     #test
     90 
---> 91     scv.tl.recover_dynamics(adata)
     92     scv.tl.velocity(adata, mode='dynamical')
     93 

/miniconda/lib/python3.7/site-packages/scvelo/tools/dynamical_model.py in recover_dynamics(data, var_names, n_top_genes, max_iter, assignment_mode, t_max, fit_time, fit_scaling, fit_steady_states, fit_connected_states, fit_basal_transcription, use_raw, load_pars, return_model, plot_results, steady_state_prior, add_key, copy, **kwargs)
    331         dm = DynamicsRecovery(adata, gene, use_raw=use_raw, load_pars=load_pars, max_iter=max_iter, fit_time=fit_time,
    332                               fit_steady_states=fit_steady_states, fit_connected_states=conn, fit_scaling=fit_scaling,
--> 333                               fit_basal_transcription=fit_basal_transcription, steady_state_prior=steady_state_prior, **kwargs)
    334         if dm.recoverable:
    335             dm.fit(assignment_mode=assignment_mode)

/miniconda/lib/python3.7/site-packages/scvelo/tools/dynamical_model.py in __init__(self, adata, gene, load_pars, **kwargs)
     21             self.load_pars(adata, gene)
     22         elif self.recoverable:
---> 23             self.initialize()
     24 
     25     def initialize(self):

/miniconda/lib/python3.7/site-packages/scvelo/tools/dynamical_model.py in initialize(self)
     46 
     47         # initialize switching from u quantiles and alpha from s quantiles
---> 48         tstat_u, pval_u, means_u = test_bimodality(u_w, kde=True)
     49         tstat_s, pval_s, means_s = test_bimodality(s_w, kde=True)
     50         self.pval_steady = max(pval_u, pval_s)

/miniconda/lib/python3.7/site-packages/scvelo/tools/utils.py in test_bimodality(x, bins, kde, plot)
    266 
    267     grid = np.linspace(np.min(x), np.percentile(x, 99), bins)
--> 268     kde_grid = gaussian_kde(x)(grid) if kde else np.histogram(x, bins=grid, density=True)[0]
    269 
    270     idx = int(bins / 2) - 2

/miniconda/lib/python3.7/site-packages/scipy/stats/kde.py in __init__(self, dataset, bw_method, weights)
    206             self._neff = 1/sum(self._weights**2)
    207 
--> 208         self.set_bandwidth(bw_method=bw_method)
    209 
    210     def evaluate(self, points):

/miniconda/lib/python3.7/site-packages/scipy/stats/kde.py in set_bandwidth(self, bw_method)
    538             raise ValueError(msg)
    539 
--> 540         self._compute_covariance()
    541 
    542     def _compute_covariance(self):

/miniconda/lib/python3.7/site-packages/scipy/stats/kde.py in _compute_covariance(self)
    550                                                bias=False,
    551                                                aweights=self.weights))
--> 552             self._data_inv_cov = linalg.inv(self._data_covariance)
    553 
    554         self.covariance = self._data_covariance * self.factor**2

/miniconda/lib/python3.7/site-packages/scipy/linalg/basic.py in inv(a, overwrite_a, check_finite)
    943 
    944     """
--> 945     a1 = _asarray_validated(a, check_finite=check_finite)
    946     if len(a1.shape) != 2 or a1.shape[0] != a1.shape[1]:
    947         raise ValueError('expected square matrix')

/miniconda/lib/python3.7/site-packages/scipy/_lib/_util.py in _asarray_validated(a, check_finite, sparse_ok, objects_ok, mask_ok, as_inexact)
    237             raise ValueError('masked arrays are not supported')
    238     toarray = np.asarray_chkfinite if check_finite else np.asarray
--> 239     a = toarray(a)
    240     if not objects_ok:
    241         if a.dtype is np.dtype('O'):

/miniconda/lib/python3.7/site-packages/numpy/lib/function_base.py in asarray_chkfinite(a, dtype, order)
    494     if a.dtype.char in typecodes['AllFloat'] and not np.isfinite(a).all():
    495         raise ValueError(
--> 496             "array must not contain infs or NaNs")
    497     return a
    498 

ValueError: array must not contain infs or NaNs

Versions: scvelo==0.1.24.dev6+geb35fe8 scanpy==1.4.4 anndata==0.6.22.post1 loompy==2.0.17 numpy==1.17.0 scipy==1.2.1 matplotlib==3.0.2 sklearn==0.21.3 pandas==0.25.1

Best, Tobi

[TobiTekath]

Thanks for your quick support, I really appreciate this. Unfortunately your added line did not fix my problem (same error and stacktrace).

But I found the culprit. For one gene, the following code

https://github.com/theislab/scvelo/blob/6dbaafa725c1d1af981ac836e7c9d458310eb70f/scvelo/tools/dynamical_model.py#L25-L33

is only selecting zeroes for u_w. The initial self.u may contain a lot of zeros (~99.5% of the values), but its not completely zero (otherwise it would have been filtered out). The NaNs are produced when dividing by scaling, which subsequently is also 0.

The error is resolved when I remove the gene in question from the var_names - which is totally fine for me at the moment. Nevertheless I would suggest to implement a check to prevent the division by zero.

Best, Tobi

[notimenocall]

It worked, at least in the data set I have. Great job!