Carlos Hernández

Would be nice to have more consistency between the MSM and HMM classes.

try to implement some sort of convergence tests

1

a la pymc: https://pymc-devs.github.io/pymc/modelchecking.html It's not perfect but at least it's a start. We should also consult the stats dept. at some point.

A lot of defaults are not stated....

When I use `DihedralFeaturizer` on the `FsPeptide` dataset, some features seem to have incorrect descriptions: ```python >> DihedralFeaturizer().describe_features(trajectories[0])[77] {'atominds': array([174, 176, 182, 184]), 'featuregroup': 'psi', 'featurizer': 'Dihedral', 'otherinfo': 'cos', 'resids':...

could be interesting to add this to msmb: https://github.com/idiap/eakmeans

see https://en.wikipedia.org/wiki/Robust_measures_of_scale#IQR_and_MAD We can make this an online algorithm fairly easily as shown in this [post](http://stackoverflow.com/questions/1058813/on-line-iterator-algorithms-for-estimating-statistical-median-mode-skewnes)

Transition towards using Pandas DataFrames?

2

I wanted to gauge what the pros and cons are for using `pandas.DataFrame` objects as attributes rather than `numpy.ndarrays`. Personally, I think the switch would be useful to keep track...

Might speed up our workflow for larger datasets: http://arxiv.org/pdf/1502.03167v3.pdf

Add Feature Selection Algorithms

2

We can start by: - [x] porting from [sklearn](http://scikit-learn.org/stable/modules/feature_selection.html) - [ ] adding a few from [this resource](http://featureselection.asu.edu/algorithms.php) (esp the decision tree-based approaches) - [ ] implementing [Gradient Boosted Feature...

This could help out with larger datasets?