Pyemma

For now all future versions will only support Python 3. 1395

**New features**:

- msm: for Bayesian MSMs we show an optional progress bar for the counting computation. 1344
- msm: ImpliedTimescales allow to only store timescales and its samples for the purpose of saving memory. 1377
- double_well_discrete: allows the setting of a random seed 1388


**Fixes**:

- msm: fix connected sets for Bayesian MSMs in case the mincount connectivity (ergodic cutoff) parameter truncated the count matrix. 1343
- plots: fix vmin, vmax keyword arguments for plot_contour(). 1376
- coordinates: forcefully enable checking of coordinates data streaming for invalid (not initialized) data. 1384
- coordinates: for sake of faster strided MD data reading, we now require a version of MDTraj >= 1.9.2 1391
- coordinates: VAMP: changed default projection vector from right to left, since only the left singular functions induce a kinetic map wrt. the conventional forward propagator, while the right singular functions induce
a kinetic map wrt. the backward propagator. 1394
- coordinates: VAMP: fixed variance cutoff to really include as many dimensions to meet subspace variance criterion. 1397


**Contributors**:

- marscher
- fabian-paul
- brookehus
- thempel

**New features**:

- coordinates: added Nystroem-TICA, which uses sparse sampling to approximate the input space. 1261, 1273
- plots: added multi-dimensional stacked histogram plot function. 1264

**Fixes**:

- msm: Chapman Kolmogorov validator ensures there are no side effects on the tested model. 1255
- datasets: Fix default values for kT to ensure integrator produces sane values. 1272, 1275
- coordinates: fixed handling of default chunksize. 1284

Quick fix release to repair chunking in the coordinates package.

**Fixes**:

- msm: fix bug in ImpliedTimescales, which happened when an estimation failed for a given lag time. 1248
- coordinates: fixed handling of default chunksize. 1247, 1251, 1252
- base: updated pybind to 2.2.2. 1249

As of this version the usage of Python 2.7 is officially deprecated. Please upgrade
your Python installation to at least version 3.5.

**New features**:
- base: most Estimators and Models in msm, thermo and coordinates packages can be saved to disk now.
Multiple models/estimators can be stored in the same file, which uses HDF5 as backend. 849, 867, 1155, 1200, 1205
- coordinates: Added Variational Approach for Markov Processes (VAMP) estimator and scoring method, see [2]. 1237
- coodinates: selection based features allow alignment to a reference structure. 1184
- coordinates: two new center of mass features: ResidueCOMFeature() and GroupCOMFeature()
- coordinates: new configuration variable 'default_chunksize' can be set to limit the size of a fragmented
extracted per iteration from a data source. This is invariant to the dimension of data sets. 1190
- msm: Added Augmented Markov Models. A way to include averaged experimental
data into estimation of Markov models from molecular simulations. The method is described in [1]. 1111
- msm: Added mincount_connectivity argument to MSM estimators. This option enables to omit counts below
a given threshold. 1106
- datasets: added Prinz potential (quadwell). 1226


References:
[1] Olsson S, Wu H, Paul F, Clementi C, Noe F: Combining experimental and simulation data of molecular
processes via augmented Markov models. PNAS 114, 8265-8270 (2017).
[2] Wu H, Noé F: Variational approach for learning Markov processes from time series data.
arXiv:1707.04659 (2017).

**Fixes**:

- datasets: fixed get_multi_temperature_data and get_umbrella_sampling_data for Python 3. 1102
- coordinates: fixed StreamingTransformers (TICA, Kmeans, etc.) not respecting the in_memory flag. 1112
- coordinates: made TrajectoryInfoCache more fail-safe in case of concurrent processes. 1122
- msm: fix setting of dt_model for BayesianMSM. This bug led to wrongly scaled time units for mean first passage times,
correlation and relaxation times as well for timescales for this estimator. 1116
- coordinates: Added the covariance property of time-lagged to CovarianceLagged. 1125
- coordinates: clustering code modernized in C++ with pybind11 interface. 1142
- variational: covartools code modernized in C++ with pybind11 interface. 1147
- estimators: n_jobs setting does not look for OMP_NUM_THREADS, but for PYEMMA_NJOBS and SLURM_CPUS_ON_NODE to avoid
multiplying OpenMP threads with PyEMMA processes. On SLURM the number of allocated cores is used.
If nothing is set, the physical cpu count is considered.
- msm: calling score_cv does not modify the object anymore. 1178
- base:estimator: fixed signature of fit function for compatability with scikit-learn. 1193
- coordinates: assign_to_centers now handles stride argument again. 1190

**New features**:

- msm: variational scores for model selection of MSMs. The scores are based on the variational
approach for Markov processes [1, 2] and can be employed for both reversible and non-reversible
MSMs. Both the Rayleigh quotient as well as the kinetic variance [3] and their non-reversible
generalizations are available. The scores are implemented in the `score` method of the MSM
estimators `MaximumLikelihoodMSM` and `OOMReweightedMSM`. Rudimentary support for Cross-validation
similar as suggested in [4] is implemented in the `score_cv` method, however this is currently
inefficient and will be improved in future versions. 1093

- config: Added a lot of documentation and added `mute` option to silence PyEMMA (almost completely).

- References:
[1] Noe, F. and F. Nueske: A variational approach to modeling slow processes
in stochastic dynamical systems. SIAM Multiscale Model. Simul. 11, 635-655 (2013).
[2] Wu, H and F. Noe: Variational approach for learning Markov processes
from time series data (in preparation).
[4] Noe, F. and C. Clementi: Kinetic distance and kinetic maps from molecular
dynamics simulation. J. Chem. Theory Comput. 11, 5002-5011 (2015).
[3] McGibbon, R and V. S. Pande: Variational cross-validation of slow
dynamical modes in molecular kinetics, J. Chem. Phys. 142, 124105 (2015).

- coordinates:
- kmeans: allow the random seed used for initializing the centers to be passed. The prior behaviour
was to init the generator by time, if fixed_seed=False. Now bool and int can be passed. 1091

- datasets:
- added a multi-ensemble data generator for the 1D asymmetric double well. 1097

**Fixes**:

- coordinates:
- StreamingEstimators: If an exception occurred during flipping the `in_memory` property,
the state is not updated. 1096
- Removed deprecated method parametrize. Use estimate or fit for now. 1088
- Readers: nice error messages for file handling errors (which file caused the error). 1085
- TICA: raise ZeroRankError, if the input data contained only constant features. 1055
- KMeans: Added progress bar for collecting the data in pre-clustering phase. 1084

- msm:
- ImpliedTimescales estimation can be interrupted (strg+c, stop button in Jupyter notebooks). 1079

- general:
- config: better documentation of the configuration parameters. 1095

-----------------

**New features**:

- thermo:
- Allow for periodicity in estimate_umbrella_sampling()
- Add *_full_state getter variants to access stationary properties on the full set of states
instead of the active set.

**Fixes**:

- coordinates:
- [TICA] fixed regularization of timescales for the non-default feature **commute_map**. 1037, 1038