def plotTimescales(self,
lags=None,
errors=None,
nits=None,
results=False,
plot=True):
""" Plot the implied timescales of MSMs of various lag times
Parameters
----------
lags : list
The lag times at which to compute the timescales. By default it spreads out 25 lag times linearly from lag
10 until the mode length of the trajectories.
errors : errors
Calculate errors using Bayes (Refer to pyEMMA documentation)
nits : int
Number of implied timescales to calculate. Default: all
results : bool
If the method should return the calculated implied timescales
plot : bool
If the method should display the plot of implied timescales
Returns
-------
If given `results`=True this method will return the following data
its : np.ndarray
The calculated implied timescales. 2D array with dimensions (len(`lags`), `nits`)
lags : np.ndarray
A list of the lag times that were used to calculate the implied timescales
Examples
--------
>>> model = Model(data)
>>> model.plotTimescales()
>>> model.plotTimescales(lags=list(range(1,100,5)))
"""
import pyemma.plots as mplt
import pyemma.msm as msm
self._integrityCheck()
if lags is None:
lags = self._defaultLags()
if nits is None:
nits = np.min((self.data.K, 20))
from htmd.config import _config
its = msm.its(self.data.St.tolist(),
lags=lags,
errors=errors,
nits=nits,
n_jobs=_config['ncpus'])
if plot:
plt.ion()
plt.figure()
mplt.plot_implied_timescales(its, dt=self.data.fstep, units='ns')
plt.show()
if results:
return its.get_timescales(), its.lags
def plotTimescales(self, lags=None, units='frames', errors=None, nits=None, results=False, plot=True):
""" Plot the implied timescales of MSMs of various lag times
Parameters
----------
lags : list
The lag times at which to compute the timescales. By default it spreads out 25 lag times linearly from lag
10 until the mode length of the trajectories.
units : str
The units of lag. Can be 'frames' or any time unit given as a string.
errors : errors
Calculate errors using Bayes (Refer to pyEMMA documentation)
nits : int
Number of implied timescales to calculate. Default: all
results : bool
If the method should return the calculated implied timescales
plot : bool
If the method should display the plot of implied timescales
Returns
-------
If given `results`=True this method will return the following data
its : np.ndarray
The calculated implied timescales. 2D array with dimensions (len(`lags`), `nits`)
lags : np.ndarray
A list of the lag times that were used to calculate the implied timescales
Examples
--------
>>> model = Model(data)
>>> model.plotTimescales()
>>> model.plotTimescales(lags=list(range(1,100,5)))
"""
import pyemma.plots as mplt
import pyemma.msm as msm
self._integrityCheck()
if lags is None:
lags = self._defaultLags()
else:
lags = unitconvert(units, 'frames', lags, fstep=self.data.fstep).tolist()
if nits is None:
nits = np.min((self.data.K, 20))
from htmd.config import _config
its = msm.its(self.data.St.tolist(), lags=lags, errors=errors, nits=nits, n_jobs=_config['ncpus'])
if plot:
from matplotlib import pylab as plt
plt.ion()
plt.figure()
mplt.plot_implied_timescales(its, dt=self.data.fstep, units='ns')
plt.show()
if results:
return its.get_timescales(), its.lags
def setUpClass(cls):
P = np.array([
[0.5, .25, .25, 0.],
[0., .25, .5, .25],
[.25, .25, .5, 0],
[.25, .25, .25, .25],
])
# bogus its object
lags = [1, 2, 3, 5, 10]
cls.its = its(generate_traj(P, 100), lags=lags, errors='bayes')
cls.refs = cls.its.timescales[-1]
return cls
def test_its_bmsm(self):
estimator = msm.its([self.double_well_data.dtraj_T100K_dt10_n6good], lags = [10, 50, 200],
errors='bayes', nsamples=1000)
ref = np.array([[ 284.87479737, 6.68390402, 3.0375248, 2.65314172, 1.93066562],
[ 320.08583492, 11.14612743, 10.3450663, 9.42799075, 8.2109752 ],
[ 351.41541961, 42.87427869, 41.17841657, 37.35485197, 23.24254608]])
# rough agreement with MLE
assert np.allclose(estimator.timescales, ref, rtol=0.1, atol=10.0)
# within left / right intervals. This test should fail only 1 out of 1000 times.
L, R = estimator.get_sample_conf(conf=0.999)
np.testing.assert_array_less(L, estimator.timescales)
np.testing.assert_array_less(estimator.timescales, R)
def test_its_bmsm(self):
estimator = msm.its([self.double_well_data.dtraj_T100K_dt10_n6good], lags = [10, 50, 200],
errors='bayes', nsamples=1000, n_jobs=2)
ref = np.array([[ 284.87479737, 6.68390402, 3.0375248, 2.65314172, 1.93066562],
[ 320.08583492, 11.14612743, 10.3450663, 9.42799075, 8.2109752 ],
[ 351.41541961, 42.87427869, 41.17841657, 37.35485197, 23.24254608]])
# rough agreement with MLE
assert np.allclose(estimator.timescales, ref, rtol=0.1, atol=10.0)
# within left / right intervals. This test should fail only 1 out of 1000 times.
L, R = estimator.get_sample_conf(conf=0.999)
# we only test the first timescale, because the second is already ambiguous (deviations after the first place),
# which makes this tests fail stochastically.
np.testing.assert_array_less(L[0], estimator.timescales[0])
np.testing.assert_array_less(estimator.timescales[0], R[0])
def _calculateITS(self):
is_converged = False
# its
print(("Calculating implied time-scales, when it's done will prompt "
"for confirmation on the validity of the lagtimes..."))
while not is_converged:
if not self.error:
itsErrors = None
elif self.error:
itsErrors = "bayes"
if self.lagtimes and self.lagtimes is not None:
# workaround to get new its plot at each iteration, the
# plot_implied_timescales function is calling plt.gca() and
# recovers the previous plot's axes, by creating a new figure
# gca gets a set of empty axes and plots are fine
plt.figure()
its_object = msm.its(self.dtrajs, lags=self.lagtimes, errors=itsErrors)
mplt.plot_implied_timescales(its_object, outfile=self.itsOutput, nits=self.numberOfITS)
plt.savefig("its.png")
if self.lagtime is not None:
return self.lagtime
while True:
plt.show()
convergence_answer = raw_input("Has the ITS plot converged?[y/n] ")
convergence_answer.rstrip()
convergence_answer = convergence_answer or "y" # Making yes the default answer
if convergence_answer.lower() == "y" or convergence_answer.lower() == "yes":
is_converged = True
lagtime_str = raw_input("Please input the lagtime to construct the MSM: ")
lagtime = int(lagtime_str.rstrip())
break
elif convergence_answer.lower() == "n" or convergence_answer.lower() == "no":
break
else:
print("Answer not valid. Please answer yes or no")
if not is_converged:
new_lagtimes = raw_input("Do you want to define new lagtimes or add to the previous?[add(a)/new(n)] ")
new_lagtimes.rstrip()
if new_lagtimes.lower() == "add" or new_lagtimes.lower() == "a":
lag_list = raw_input("Please input the lagtimes you want to add separated by a space: ")
lag_list.rstrip()
self.lagtimes.extend(map(int, lag_list.split(" ")))
elif new_lagtimes.lower() == "new" or new_lagtimes.lower() == "n":
lag_list = raw_input("Please input the new lagtimes separated by a space: ")
lag_list.rstrip()
self.lagtimes = map(int, lag_list.split(" "))
self.lagtimes.sort()
return lagtime
def _calculateITS(self):
print("Calculating implied time-scales")
if not self.error:
itsErrors = None
elif self.error:
itsErrors = "bayes"
if self.lagtimes and self.lagtimes is not None:
# workaround to get new its plot at each iteration, the
# plot_implied_timescales function is calling plt.gca() and
# recovers the previous plot's axes, by creating a new figure
# gca gets a set of empty axes and plots are fine
plt.figure()
its_object = msm.its(self.dtrajs, lags=self.lagtimes, errors=itsErrors)
mplt.plot_implied_timescales(its_object, outfile=self.itsOutput, nits=self.numberOfITS)
plt.savefig("its.png")
if self.lagtime is not None:
return self.lagtime
def plot_implied_timescales(dtrajs, nits=15, model_name=""):
""" Compute and plot implied timescales.
Parameters
----------
dtrajs
nits
model_name
Returns
-------
its : :class:`ImpliedTimescales <pyemma.msm.estimators.implied_timescales.ImpliedTimescales>` object
"""
its = msm.its(dtrajs, lags=lags, nits=nits, errors="bayes")
mplt.plot_implied_timescales(its, dt=0.25, units="ns")
plt.title(model_name)
return its
def plot_implied_timescales(dtrajs,nits=15,model_name=''):
''' Compute and plot implied timescales.
Parameters
----------
dtrajs
nits
model_name
Returns
-------
its : :class:`ImpliedTimescales <pyemma.msm.estimators.implied_timescales.ImpliedTimescales>` object
'''
its = msm.its(dtrajs, lags=lags, nits=nits)#,errors='bayes')
mplt.plot_implied_timescales(its,dt=0.25,units='ns')
plt.title(model_name)
plt.savefig('{0}_implied_timescales.pdf'.format(model_name))
plt.close()
return its
def maxConnectedLag(self, lags):
""" Heuristic for getting the lagtime before a timescale drops.
It calculates the last lagtime before a drop occurs in the first implied timescale due to disconnected states.
If the top timescale is closer to the second top timescale at the previous lagtime than to itself at the previous
lagtime it means that a drop occured. The lagtime before the drop is returned.
Parameters
----------
lags : np.ndarray or list
A list of lag times for which to calculate the implied timescales
Returns
-------
ml : int
The maximum lagtime before a drop occurs in the top timescale
Examples
--------
>>> model = Model(data)
>>> model.maxConnectedLag(list(range(1, 100, 5)))
"""
if len(lags) == 1:
return lags
if isinstance(lags, np.ndarray):
lags = lags.astype(int)
import pyemma.msm as msm
itime = msm.its(self.data.St.tolist(), lags=lags,
nits=2).get_timescales()
for i in range(1, np.size(itime, 0)):
if abs(itime[i, 0] - itime[i - 1, 1]) < abs(itime[i, 0] -
itime[i - 1, 0]):
lagidx = i - 1
break
else:
lagidx = i
return lags[lagidx], itime
def maxConnectedLag(self, lags):
""" Heuristic for getting the lagtime before a timescale drops.
It calculates the last lagtime before a drop occurs in the first implied timescale due to disconnected states.
If the top timescale is closer to the second top timescale at the previous lagtime than to itself at the previous
lagtime it means that a drop occured. The lagtime before the drop is returned.
Parameters
----------
lags : np.ndarray or list
A list of lag times for which to calculate the implied timescales
Returns
-------
ml : int
The maximum lagtime before a drop occurs in the top timescale
Examples
--------
>>> model = Model(data)
>>> model.maxConnectedLag(list(range(1, 100, 5)))
"""
if len(lags) == 1:
return lags
if isinstance(lags, np.ndarray):
lags = lags.astype(int)
import pyemma.msm as msm
itime = msm.its(self.data.St.tolist(), lags=lags, nits=2).get_timescales()
for i in range(1, np.size(itime, 0)):
if abs(itime[i, 0] - itime[i-1, 1]) < abs(itime[i, 0] - itime[i-1, 0]):
lagidx = i-1
break
else:
lagidx = i
return lags[lagidx], itime
if i == 3:
for j in range(4):
axes[i][j].set_xlabel("TIC " + str(j + 2), fontsize=20)
axes[0][0].annotate("TICA " + f_str,
fontsize=24,
xy=(0, 0),
xytext=(1.8, 1.1),
xycoords="axes fraction",
textcoords="axes fraction")
fig.savefig(msm_savedir + "/tic_hist_grid.pdf")
n_clusters = 300
msm_lags = [1, 10, 20, 50, 100, 200]
cluster = coor.cluster_kmeans(k=n_clusters)
coor.pipeline([reader, tica, cluster])
its = msm.its(cluster.dtrajs, lags=msm_lags)
plt.figure()
mplt.plot_implied_timescales(its)
plt.title(msm_savedir)
plt.savefig(msm_savedir + "/its_vs_lag_ylog.pdf")
#plt.figure()
#plt.plot(np.arange(1,21), M.timescales()[:20], 'o')
#ymin, ymax = plt.ylim()
#plt.ylim(0, ymax)
#plt.savefig("msm_ti.pdf")
import pyemma.coordinates as coor
import numpy as np
import pyemma.msm as msm
import pyemma.plots as pyemma_plots
import matplotlib.pyplot as plt
sys = 'fdis'
n_clusters = 100
dtrajs = coor.load(f'cluster_data/{sys}_{n_clusters}_cluster_dtrajs.h5')
max_lag = 80
dt2 = [i.astype(np.int_) for i in dtrajs]
dt3 = [i.reshape((i.shape[0])) for i in dt2]
its = msm.its(dt3, lags=max_lag, nits=8, errors='bayes', nsamples=200)
fig, ax = plt.subplots()
pyemma_plots.plot_implied_timescales(its, units='ns', ax=ax)
fig.savefig(f'{sys}_implied_timescale_{max_lag}.pdf')
elapsed_time = final_time - initial_time
print('Elapsed time %.3f s' % elapsed_time)
# Save cluster centers
np.save('clustercenters', clustering.clustercenters)
# Save discrete trajectories.
dtrajs = clustering.dtrajs
dtrajs_dir = 'dtrajs'
clustering.save_dtrajs(output_dir=dtrajs_dir,
output_format='npy',
extension='.npy')
################################################################################
# Make timescale plots
################################################################################
import matplotlib as mpl
mpl.use('Agg') # Don't use display
import matplotlib.pyplot as plt
from pyemma import msm
from pyemma import plots
lags = [1, 2, 5, 10, 20, 50]
#its = msm.its(dtrajs, lags=lags, errors='bayes')
its = msm.its(dtrajs, lags=lags)
plots.plot_implied_timescales(its)
plt.savefig('plot.pdf')
elapsed_time = final_time - initial_time
print('Elapsed time %.3f s' % elapsed_time)
# Save cluster centers
np.save('clustercenters', clustering.clustercenters)
# Save discrete trajectories.
dtrajs = clustering.dtrajs
dtrajs_dir = 'dtrajs'
clustering.save_dtrajs(output_dir=dtrajs_dir, output_format='npy', extension='.npy')
################################################################################
# Make timescale plots
################################################################################
import matplotlib as mpl
mpl.use('Agg') # Don't use display
import matplotlib.pyplot as plt
from pyemma import msm
from pyemma import plots
lags = [1,2,5,10,20,50]
#its = msm.its(dtrajs, lags=lags, errors='bayes')
its = msm.its(dtrajs, lags=lags)
plots.plot_implied_timescales(its)
plt.savefig('plot.pdf')
# This script is used to test the MSM
# construction with coring after clustering
# using PyEmma
#
# Please refer to PyEmma documentation for more information
import pyemma
import pyemma.msm as msm
import numpy as np
from udpclust import UDPClust as dp
tica_traj=[]
for i in range(4):
fname='DATA/test-its-traj'+str(i)+'.dat'
# tr=tica_traj[i][::25]
tica_traj.append(np.loadtxt(fname))
cl_dpa=dp.cluster_UDP(dim=6,trj_tot=tica_traj,stride=10)
ctrajs=cl_dpa.get_core_traj()
its=msm.its(ctrajs,lags=range(1,10,1))
np.savetxt('out-msm_rho.dat',cl_dpa.rho,fmt="%.6e")
np.savetxt('out-msm_its.dat',its.timescales,fmt="%.6e")
def calculateITS(trajectories, lagtimes, errors=None):
""" Calulate the implied time-scales at the given lagtimes"""
its_object = MSM.its(trajectories, lags=lagtimes, errors=errors)
return its_object
def plotTimescales(self,
lags=None,
units='frames',
errors=None,
nits=None,
results=False,
plot=True,
save=None):
""" Plot the implied timescales of MSMs of various lag times
Parameters
----------
lags : list
The lag times at which to compute the timescales. By default it spreads out 25 lag times linearly from lag
10 until the mode length of the trajectories.
units : str
The units of lag. Can be 'frames' or any time unit given as a string.
errors : errors
Calculate errors using Bayes (Refer to pyEMMA documentation)
nits : int
Number of implied timescales to calculate. Default: all
results : bool
If the method should return the calculated implied timescales
plot : bool
If the method should display the plot of implied timescales
save : str
Path of the file in which to save the figure
Returns
-------
If given results=True this method will return the following data
its : np.ndarray
The calculated implied timescales. 2D array with dimensions (len(`lags`), `nits`)
lags : np.ndarray
A list of the lag times that were used to calculate the implied timescales
Examples
--------
>>> model = Model(data)
>>> model.plotTimescales()
>>> model.plotTimescales(lags=list(range(1,100,5)))
"""
import pyemma.plots as mplt
import pyemma.msm as msm
self._integrityCheck()
if lags is None:
lags = self._defaultLags()
else:
lags = unitconvert(units, 'frames', lags,
fstep=self.data.fstep).tolist()
if nits is None:
nits = np.min((self.data.K, 20))
from htmd.config import _config
its = msm.its(self.data.St.tolist(),
lags=lags,
errors=errors,
nits=nits,
n_jobs=_config['ncpus'])
if plot or (save is not None):
from matplotlib import pylab as plt
plt.ion()
plt.figure()
try:
mplt.plot_implied_timescales(its,
dt=self.data.fstep,
units='ns')
except ValueError as ve:
plt.close()
raise ValueError(
'{} This is probably caused by badly set fstep in the data ({}). '
.format(ve, self.data.fstep) +
'Please correct the model.data.fstep to correspond to the simulation frame step in nanoseconds.'
)
if save is not None:
plt.savefig(save, dpi=300, bbox_inches='tight', pad_inches=0.2)
if plot:
plt.show()
if results:
return its.get_timescales(), its.lags
def plotTimescales(self, lags=None, units='frames', errors=None, nits=None, results=False, plot=True, save=None):
""" Plot the implied timescales of MSMs of various lag times
Parameters
----------
lags : list
The lag times at which to compute the timescales. By default it spreads out 25 lag times linearly from lag
10 until the mode length of the trajectories.
units : str
The units of lag. Can be 'frames' or any time unit given as a string.
errors : errors
Calculate errors using Bayes (Refer to pyEMMA documentation)
nits : int
Number of implied timescales to calculate. Default: all
results : bool
If the method should return the calculated implied timescales
plot : bool
If the method should display the plot of implied timescales
save : str
Path of the file in which to save the figure
Returns
-------
If given results=True this method will return the following data
its : np.ndarray
The calculated implied timescales. 2D array with dimensions (len(`lags`), `nits`)
lags : np.ndarray
A list of the lag times that were used to calculate the implied timescales
Examples
--------
>>> model = Model(data)
>>> model.plotTimescales()
>>> model.plotTimescales(lags=list(range(1,100,5)))
"""
import pyemma.plots as mplt
import pyemma.msm as msm
self._integrityCheck()
if lags is None:
lags = self._defaultLags()
else:
lags = unitconvert(units, 'frames', lags, fstep=self.data.fstep).tolist()
if nits is None:
nits = np.min((self.data.K, 20))
from htmd.config import _config
its = msm.its(self.data.St.tolist(), lags=lags, errors=errors, nits=nits, n_jobs=_config['ncpus'])
if plot or (save is not None):
from matplotlib import pylab as plt
plt.ion()
plt.figure()
try:
mplt.plot_implied_timescales(its, dt=self.data.fstep, units='ns')
except ValueError as ve:
plt.close()
raise ValueError('{} This is probably caused by badly set fstep in the data ({}). '.format(ve, self.data.fstep) +
'Please correct the model.data.fstep to correspond to the simulation frame step in nanoseconds.')
if save is not None:
plt.savefig(save, dpi=300, bbox_inches='tight', pad_inches=0.2)
if plot:
plt.show()
if results:
return its.get_timescales(), its.lags
plt.figure(figsize=(5, 3))
plt.plot(time_scale_sep, linewidth=0, marker='o')
#plt.axvline(x=last_slow_index+0.5,color='r')
plt.axvline(x=0.5, color='g')
plt.xlabel('index')
plt.ylabel('timescale separation')
plt.xlim(0, 30)
plt.savefig('timescale_separation.png')
#print('%s macrostates chosen from timescale separation' %n_macrostates_timescales)
print('%s macrostates chosen because thats what we want' % n_macrostates)
plt.clf()
lags = [1, 2, 5, 10, 20, 50, 100, 200, 400]
its = msm.its(clkmeans.dtrajs, lags=lags)
mplt.plot_implied_timescales(its)
plt.savefig('implied_timescale_plot.png')
plt.clf()
print('fraction of states used = ', MSM.active_state_fraction)
print('fraction of counts used = ', MSM.active_count_fraction)
mplt.plot_cktest(MSM.cktest(3))
plt.savefig('cktest_msm.png')
plt.clf()
plt.figure(figsize=(8, 5))
mplt.plot_free_energy(np.hstack(Y1),
print('Elapsed time %.3f s' % elapsed_time)
# Save cluster centers
#import cPickle as pickle
#pickle.dump(clustering.clustercenters, open('clustercenters.p', 'wb'))
np.save('clustercenters.npy', clustering.clustercenters)
# Save discrete trajectories.
dtrajs = clustering.dtrajs
dtrajs_dir = 'dtrajs'
clustering.save_dtrajs(output_dir=dtrajs_dir, output_format='npy', extension='.npy')
################################################################################
# Make timescale plots
################################################################################
import matplotlib as mpl
mpl.use('Agg') # Don't use display
import matplotlib.pyplot as plt
from pyemma import msm
from pyemma import plots
lags = [1,2,5,10,20,50]
its = msm.its(dtrajs, lags=lags, errors='bayes')
plots.plot_implied_timescales(its)
plt.savefig('plot.pdf')
