Remove deprecated CK Test validator (#258)

Author: clonker

deeptime/decomposition/_vamp.py

@@ -8,14 +8,12 @@
 
 import numpy as np
 
-from ._koopman import CovarianceKoopmanModel, KoopmanObservable
+from ._koopman import CovarianceKoopmanModel
 from ..base import EstimatorTransformer
 from ..basis import Identity
 from ..covariance import Covariance, CovarianceModel
 from ..numeric import spd_inv_split
-from ..util.decorators import deprecated_method
 from ..util.types import to_dataset
-from ..util.validation import DeprecatedCKValidator
 
 
 class VAMP(EstimatorTransformer):
@@ -460,31 +458,3 @@ def fetch_model(self) -> CovarianceKoopmanModel:
             self._model = self._decompose(self._covariance_estimator.fetch_model())
             self._covariance_estimator = None
         return self._model
-
-    @deprecated_method("Deprecated in v0.4.1 and will be removed soon, please use model.ck_test.")
-    def chapman_kolmogorov_validator(self, mlags, test_model: CovarianceKoopmanModel = None,
-                                     n_observables=None, observables='phi', statistics='psi'):
-        r""" Replaced by `deeptime.decomposition.CovarianceKoopmanModel.ck_test`. """
-        test_model = self.fetch_model() if test_model is None else test_model
-        assert test_model is not None, "We need a test model via argument or an estimator which was already " \
-                                       "fit to data."
-
-        def fit_for_lag(data, lagtime):
-            est = VAMP(lagtime=lagtime, dim=self.dim, var_cutoff=self.var_cutoff, scaling=self.scaling,
-                       epsilon=self.epsilon, observable_transform=self.observable_transform)
-            return est.fit(data).fetch_model()
-
-        if isinstance(observables, str) and observables == 'phi':
-            observables = test_model.singular_vectors_right[:, :n_observables]
-            observables_mean_free = True
-        else:
-            observables_mean_free = False
-
-        if isinstance(statistics, str) and statistics == 'psi':
-            statistics = test_model.singular_vectors_left[:, :n_observables]
-            statistics_mean_free = True
-        else:
-            statistics_mean_free = False
-
-        observable = KoopmanObservable(observables, statistics, observables_mean_free, statistics_mean_free)
-        return DeprecatedCKValidator(self, fit_for_lag, mlags, observable, test_model)

deeptime/markov/hmm/_bayesian_hmm.py

@@ -3,29 +3,19 @@
 from typing import Optional, Union, List
 
 import numpy as np
-
-from deeptime.markov.tools.analysis import is_connected
-from deeptime.markov.tools.estimation import sample_tmatrix, transition_matrix
+from deeptime.markov.hmm._hmm_bindings import util as _bd_util
 
 from deeptime.base import Estimator
+from deeptime.markov import TransitionCountModel, compute_dtrajs_effective, number_of_states
 from deeptime.markov._base import BayesianMSMPosterior
 from deeptime.markov._transition_matrix import stationary_distribution
 from deeptime.markov.hmm import HiddenMarkovModel
-from ._output_model import DiscreteOutputModel
-from ._util import observations_in_state, sample_hidden_state_trajectory
 from deeptime.markov.msm import MarkovStateModel
-from deeptime.markov import TransitionCountModel, compute_dtrajs_effective, number_of_states
-from deeptime.markov.hmm._hmm_bindings import util as _bd_util
+from deeptime.markov.tools.analysis import is_connected
+from deeptime.markov.tools.estimation import sample_tmatrix, transition_matrix
 from deeptime.util.types import ensure_dtraj_list
-
-__author__ = 'noe, clonker'
-
-__all__ = [
-    'BayesianHMMPosterior',
-    'BayesianHMM',
-]
-
-from ...util.decorators import deprecated_method
+from ._output_model import DiscreteOutputModel
+from ._util import observations_in_state, sample_hidden_state_trajectory
 
 from ...util.platform import handle_progress_bar
 from ...util.validation import ck_test
@@ -651,28 +641,3 @@ def _append_sample(self, models, prior, sample_model):
                                               reversible=self.reversible, count_model=count_model),
             output_model=model_copy.output_model, initial_distribution=model_copy.initial_distribution,
             hidden_state_trajectories=model_copy.hidden_trajs))
-
-    @deprecated_method("Deprecated in v0.4.1 and will be removed soon, please use model.ck_test.")
-    def chapman_kolmogorov_validator(self, mlags, test_model: BayesianHMMPosterior = None):
-        r""" Replaced by `deeptime.markov.hmm.BayesianHMMPosterior.ck_test`. """
-        test_model = self.fetch_model() if test_model is None else test_model
-        assert test_model is not None, "We need a test model via argument or an estimator which was already" \
-                                       "fit to data."
-
-        from . import DiscreteOutputModel
-        assert isinstance(test_model.prior.output_model, DiscreteOutputModel), \
-            "Can only perform CKTest for discrete output models"
-
-        from deeptime.markov._observables import MembershipsObservable
-        obs = MembershipsObservable(test_model, np.eye(test_model.prior.n_hidden_states))
-        from deeptime.util.validation import DeprecatedCKValidator
-
-        def fit_for_lag(data, lag):
-            estimator = BayesianHMM.default(dtrajs=data, n_hidden_states=self.initial_hmm.n_hidden_states,
-                                            lagtime=lag, n_samples=self.n_samples, stride=self.stride,
-                                            initial_distribution_prior=self.initial_distribution_prior,
-                                            transition_matrix_prior=self.transition_matrix_prior,
-                                            reversible=self.reversible, stationary=self.stationary)
-            return estimator.fit(data).fetch_model()
-
-        return DeprecatedCKValidator(self, fit_for_lag, mlags, obs, test_model)

deeptime/markov/hmm/_maximum_likelihood_hmm.py

@@ -10,7 +10,6 @@
 from ..msm import MarkovStateModel
 from .. import TransitionCountModel, compute_dtrajs_effective
 from ._hmm_bindings import util as _util
-from ...util.decorators import deprecated_method
 from ...util.types import ensure_timeseries_data
 
 
@@ -423,30 +422,3 @@ def _update_model(self, model: _HMMModelStorage, observations: List[np.ndarray],
         model.initial_distribution[:] = pi
         model.transition_matrix[:] = T
         model.output_model.fit(observations, gammas)
-
-    @deprecated_method("Deprecated in v0.4.1 and will be removed soon, please use model.ck_test.")
-    def chapman_kolmogorov_validator(self,  mlags, test_model: HiddenMarkovModel = None):
-        r""" Replaced by `deeptime.markov.hmm.MaximumLikelihoodHMM.ck_test`. """
-        test_model = self.fetch_model() if test_model is None else test_model
-        assert test_model is not None, "We need a test model via argument or an estimator which was already" \
-                                       "fit to data."
-        from . import DiscreteOutputModel
-        assert isinstance(test_model.output_model, DiscreteOutputModel), \
-            "Can only perform CKTest for discrete output models"
-
-        from deeptime.markov._observables import MembershipsObservable
-        obs = MembershipsObservable(test_model, np.eye(test_model.n_hidden_states))
-        from deeptime.util.validation import DeprecatedCKValidator
-
-        def fit_for_lag(data, lag):
-            from .init.discrete import metastable_from_data
-            initial_model = metastable_from_data(data, n_hidden_states=test_model.n_hidden_states, lagtime=lag,
-                                                 stride=self.stride, reversible=self.reversible,
-                                                 stationary=self.stationary)
-            estimator = MaximumLikelihoodHMM(initial_model, lagtime=lag, reversible=self.reversible,
-                                             stationary=self.stationary, accuracy=self.accuracy,
-                                             maxit=self.maxit, maxit_reversible=self.maxit_reversible)
-            hmm = estimator.fit(data).fetch_model()
-            return hmm.submodel_largest(dtrajs=data)
-
-        return DeprecatedCKValidator(self, fit_for_lag, mlags, obs, test_model)

deeptime/markov/msm/_bayesian_msm.py

@@ -3,12 +3,11 @@
 
 import numpy as np
 
+from . import MarkovStateModel, MaximumLikelihoodMSM
+from .._base import _MSMBaseEstimator, BayesianMSMPosterior
 from .._transition_counting import TransitionCountEstimator
 from ...base import Estimator
 from ...numeric import is_square_matrix
-from .._base import _MSMBaseEstimator, BayesianMSMPosterior
-from . import MarkovStateModel, MaximumLikelihoodMSM
-from ...util.decorators import deprecated_method
 
 __author__ = 'noe, marscher, clonker'
 
@@ -353,25 +352,3 @@ def fit_from_counts(self, counts, callback=None, **kw):
             sparse=self.sparse, maxiter=self.maxiter, maxerr=self.maxerr
         ).fit(counts).fetch_model()
         return self.fit_from_msm(msm, callback=callback, **kw)
-
-    @deprecated_method("Deprecated in v0.4.1 and will be removed soon, please use model.ck_test.")
-    def chapman_kolmogorov_validator(self, n_metastable_sets: int, mlags, test_model=None):
-        r""" Replaced by `deeptime.markov.msm.BayesianMSMPosterior.ck_test`. """
-        test_model = self.fetch_model() if test_model is None else test_model
-        assert test_model is not None, "We need a test model via argument or an estimator which was already" \
-                                       "fit to data."
-        prior = test_model.prior
-        assert prior.has_count_model, "The test model needs to have a count model, i.e., be estimated from data."
-        pcca = prior.pcca(n_metastable_sets)
-        from deeptime.markov._observables import MembershipsObservable
-        obs = MembershipsObservable(test_model, pcca)
-        from deeptime.util.validation import DeprecatedCKValidator
-
-        def fit_for_lag(data, lag):
-            from deeptime.markov import TransitionCountEstimator
-            counting_mode = test_model.prior.count_model.counting_mode
-            counts = TransitionCountEstimator(lag, counting_mode).fit(data, n_jobs=1).fetch_model().submodel_largest()
-            return self.fit(counts).fetch_model()
-
-        return DeprecatedCKValidator(self, fit_for_lag, mlags, obs, test_model)
-

deeptime/markov/msm/_maximum_likelihood_msm.py

@@ -9,7 +9,6 @@
 from .._base import _MSMBaseEstimator
 from .._transition_counting import TransitionCountModel, TransitionCountEstimator
 from ...numeric import is_square_matrix
-from ...util.decorators import deprecated_method
 
 log = logging.getLogger(__file__)
 
@@ -358,22 +357,3 @@ def fit(self, data, *args, **kw):
                 raise ValueError("To fit directly from a discrete timeseries, a lagtime must be provided!")
             return self.fit_from_discrete_timeseries(data, kw.pop('lagtime', self.lagtime),
                                                      kw.pop("count_mode", "sliding"))
-
-    @deprecated_method("Deprecated in v0.4.1 and will be removed soon, please use model.ck_test.")
-    def chapman_kolmogorov_validator(self, n_metastable_sets: int, mlags, test_model=None):
-        r""" Removed and replaced by `deeptime.markov.msm.MarkovStateModel.ck_test`. """
-        from deeptime.markov._observables import MembershipsObservable
-        test_model = self.fetch_model() if test_model is None else test_model
-        assert test_model is not None, "We need a test model via argument or an estimator which was already" \
-                                       "fit to data."
-        assert test_model.has_count_model, "The test model needs to have a count model, i.e., be estimated from data."
-        pcca = test_model.pcca(n_metastable_sets)
-        obs = MembershipsObservable(test_model, pcca)
-        from deeptime.util.validation import DeprecatedCKValidator
-
-        def fit_for_lag(data, lag):
-            counting_mode = test_model.count_model.counting_mode
-            counts = TransitionCountEstimator(lag, counting_mode).fit(data).fetch_model().submodel_largest()
-            return self.fit(counts).fetch_model()
-
-        return DeprecatedCKValidator(self, fit_for_lag, mlags, obs, test_model)

deeptime/util/validation.py

@@ -2,7 +2,7 @@
 
 from .decorators import plotting_function
 from .platform import handle_progress_bar
-from ..base import Observable, BayesianModel, Estimator
+from ..base import Observable, BayesianModel
 
 
 def implied_timescales(models, n_its=None):
@@ -364,26 +364,3 @@ def plot(self, height=2.5, aspect=1., conf: float = 0.95, color=None, grid = Non
         from deeptime.plots import plot_ck_test
         return plot_ck_test(self, height=height, aspect=aspect, conf=conf, color=color, grid=grid, legend=legend,
                             xlabel=xlabel, ylabel=ylabel, y01=y01, sharey=sharey, **plot_kwargs)
-
-
-class DeprecatedCKValidator(Estimator):
-
-    def __init__(self, estimator, fit_for_lag, mlags, observable, test_model):
-        super().__init__()
-        self.estimator = estimator
-        self.mlags = mlags
-        self.fit_for_lag = fit_for_lag
-        self.observable = observable
-        self.test_model = test_model
-
-    def fit(self, data, **kwargs):
-        if hasattr(self.test_model, 'prior'):
-            test_lag = self.test_model.prior.lagtime
-        else:
-            test_lag = self.test_model.lagtime
-        models = []
-        for factor in range(1, self.mlags):
-            lagtime = factor * test_lag
-            models.append(self.fit_for_lag(data, lagtime))
-        self._model = ck_test(models, observable=self.observable, test_model=self.test_model)
-        return self

tests/decomposition/test_vamp.py

@@ -255,8 +255,6 @@ def test_cktest():
         models.append(VAMP(lag, dim=1).fit_fetch(traj))
     test_model = models[0]
     ck_test = test_model.ck_test(models)
-    # validator = estimator.chapman_kolmogorov_validator(4)
-    # cktest = validator.fit(traj).fetch_model()
     np.testing.assert_almost_equal(ck_test.predictions, ck_test.estimates, decimal=1)
 
 

tests/plots/test_ck_test.py

@@ -47,25 +47,3 @@ def test_sanity_vamp(fractional):
             plot_ck_test(models[0].ck_test(models))
     else:
         plot_ck_test(models[0].ck_test(models))
-
-
-@pytest.mark.parametrize("hidden", [False, True], ids=lambda x: f"hidden={x}")
-@pytest.mark.parametrize("bayesian", [False, True], ids=lambda x: f"bayesian={x}")
-def test_sanity_deprecated_msm(hidden, bayesian):
-    traj = ellipsoids().observations(20000)
-    dtraj = KMeans(n_clusters=15).fit_transform(traj)
-    est, _ = estimate_markov_model(1, dtraj, hidden=hidden, bayesian=bayesian, n_hidden=2, return_estimator=True)
-    with pytest.deprecated_call():
-        if not hidden:
-            validator = est.chapman_kolmogorov_validator(2, 5)
-        else:
-            validator = est.chapman_kolmogorov_validator(5)
-
-        validator.fit_fetch(dtraj)
-
-
-def test_deprecated_vamp():
-    with pytest.deprecated_call():
-        traj = ellipsoids().observations(20000)
-        validator = VAMP(1, dim=2).fit(traj).chapman_kolmogorov_validator(mlags=10)
-        validator.fit_fetch(traj)