search

markovmodel / PyEMMA

🚂 Python API for Emma's Markov Model Algorithms 🚂
http://pyemma.org
GNU Lesser General Public License v3.0
311 stars 119 forks source link

cluster_mini_batch_kmeans error isfinite(sum) failed #1348

Closed euhruska closed 6 years ago

euhruska commented 6 years ago

For cluster_mini_batch_kmeans pyemma 2.5.4, python 2.7.14 cl = pyemma.coordinates.cluster_mini_batch_kmeans(data=y, k=msm_states, max_iter=10, n_jobs=None)

I get a fatal error:

kmeans iterations:   0% 0/10 [00:00<?, ?it/s]^[[A^Minitialize kmeans++ centers:   0% 1/300 [00:17<1:26:13, 17.30s/it]python2.7: pyemma/coordinates/clustering/include/bits/metric_base_bits.h:104: dtype euclidean_metric<dtype>::compute(const dtype*, const dtype*) [with dtype = float]: Assertion `std::isfinite(sum)' failed.
marscher commented 6 years ago

This is a very rare condition and says that during euclidean norm computation the sum of squares overflowed, eg did not fit into the float data type any more. This indicates that something is bogus with your input data.

marscher commented 6 years ago

a follow up issue of this one should be "released package on conda-forge uses debug flags for compilation"