Dealing with issues related to Active set

[GSINGH006]

Hi @clonker recently while building a msm i faced a issue where my state fraction was less than 1 I know that msm is built on the active sets but i wanted indices for the states after performing pcca+ "deeptime.markov.sample.indices_by_distribution " but due to shape mismatch that was not possible how should I proceed with this case ?

[clonker]

I assume you want to sample in pcca state space then? In that case you have to map the actual state of the trajectory to the pcca state (via the pcca model sets property). Although this is argmaxed. For a real quantitative analysis, you should be using the assignment probabilities and take it from there. Generally, if states in a trajectory are not within the active set you should filter them out. Then, depending on your use case, you should either divide up the trajectories into individual chunks of states that do not leave the active set or you just remove the non-active data. The latter can only be done if subsequent analysis does not rely on time correlations.

[GSINGH006]

@clonker So,I just want to sample structures from that particular states I first use this indices = dt.markov.sample.compute_index_states(dtrajs) to get indices then pcca_dist=cg.metastable_distributions to get metastable distributions followed by dt.markov.sample.indices_by_distribution(indices,distributions=pcca_dist1,nsample=100) , so length of indices list obatined from compute_index_stateswill be equal to my K value in K means can I remove the states from indices here which are not part of the cg.sets as I just want the structures belonging to that state