Ah no, this creates an additional dependency on pyemma which we want to avoid (that's the purpose of the msmtools package). We'd need to include our k-means code too in _external. But I'd be fine to just use the k-means initialization for now - it's not clear why k-means would help.
Moreover I'd reduce the sklearn GMM module to what is really used, which is only a small subset (we only need to execute fit and retrieve the result)
Ah no, this creates an additional dependency on pyemma which we want to avoid (that's the purpose of the msmtools package). We'd need to include our k-means code too in _external. But I'd be fine to just use the k-means initialization for now - it's not clear why k-means would help.
Moreover I'd reduce the sklearn GMM module to what is really used, which is only a small subset (we only need to execute fit and retrieve the result)