def _unpack_simple_data(self, start, end=None):
self.all_data = np.array([], dtype=np.float32)
self.all_data_aux = np.array([], dtype=DTYPE)
self.n_samples = np.array([]).astype(int)
self.bias_mat = None
# If autocorr not set, set it ouselves
if self.calc_autocorr:
self.autocorr = np.zeros(self.n_windows)
for i, (ds_name, ds) in enumerate(self.dr.datasets.items()):
log.info('extracting ds {}'.format(i))
if self.ts == None:
self.ts = []
# Sanity check - every input should have same timestep
#else:
# np.testing.assert_almost_equal(self.ts, ds.ts)
self.ts.append(ds.ts)
data = ds.data[start:end]
dataframe = np.array(data)
if self.calc_autocorr:
autocorr_len = np.ceil(
pymbar.timeseries.integratedAutocorrelationTime(
dataframe[:, -1]))
self.autocorr[i] = ds.ts * autocorr_len
self.n_samples = np.append(self.n_samples, dataframe.shape[0])
if self.bias_mat is None:
self.bias_mat = self.beta * dataframe.copy()
else:
self.bias_mat = np.vstack(
(self.bias_mat, self.beta * dataframe))
self.all_data = np.append(self.all_data, dataframe[:, -1])
#self.all_data_aux = np.append(self.all_data_aux, ds.aux_data[start:end])
del ds
#if do_autocorr:
log.info("saving integrated autocorr times (in ps) to 'autocorr.dat'")
np.savetxt(
'autocorr.dat',
self.autocorr,
header='integrated autocorrelation times for each window (in ps)')
self.ts = np.array(self.ts)
dr.clearData()
def _unpack_xvg_data(self, start, end=None):
self.all_data = None
self.n_samples = np.array([], dtype=np.int32)
self.bias_mat = None
if self.autocorr is None:
do_autocorr = True
self.autocorr = np.zeros(self.n_windows)
else:
do_autocorr = False
for i, (ds_name, ds) in enumerate(self.dr.datasets.iteritems()):
log.info("Unpacking {}th dataset ({:s})".format(i, ds_name))
if self.ts == None:
self.ts = ds.ts
else:
np.testing.assert_almost_equal(self.ts, ds.ts)
dataframe = np.array(ds.data[start:end], dtype=np.float32)
if do_autocorr:
log.info(" Calculating autocorrelation time...")
autocorr_nsteps = 1
for k in xrange(self.n_windows):
if k == i:
continue
try:
autocorr_res = 2 * np.ceil(pymbar.timeseries.integratedAutocorrelationTime(dataframe[:,k]))
if autocorr_res > autocorr_nsteps:
autocorr_nsteps = autocorr_res
except:
continue
self.autocorr[i] = self.ts * autocorr_nsteps * 0.5
log.info(" Tau={} ps".format(self.autocorr[i]))
bias = self.beta*dataframe # biased values for all windows
self.n_samples = np.append(self.n_samples, dataframe.shape[0])
if self.all_data is None:
self.all_data = dataframe
else:
self.all_data = np.vstack((self.all_data, dataframe))
if self.bias_mat is None:
self.bias_mat = bias
else:
self.bias_mat = np.vstack((self.bias_mat, bias))
dr.clearData()
def _unpack_pmf_data(self, start, end=None):
self.all_data = np.array([], dtype=DTYPE)
self.all_data_aux = np.array([], dtype=DTYPE)
self.n_samples = np.array([]).astype(int)
# If autocorr not set, set it ouselves
if self.calc_autocorr:
self.autocorr = np.zeros(self.n_windows)
for i, (ds_name, ds) in enumerate(self.dr.datasets.items()):
if self.ts == None:
self.ts = ds.ts
# Sanity check - every input should have same timestep
else:
np.testing.assert_almost_equal(self.ts, ds.ts)
data = ds.data[start:end]
dataframe = np.array(data['dr_sq'])
if self.calc_autocorr:
autocorr_len = np.ceil(
pymbar.timeseries.integratedAutocorrelationTime(
dataframe[:]))
self.autocorr[i] = ds.ts * autocorr_len
self.n_samples = np.append(self.n_samples, dataframe.shape[0])
self.all_data = np.append(self.all_data, dataframe)
self.all_data_aux = np.append(self.all_data_aux, dataframe)
self.bias_mat = np.zeros((self.n_tot, self.n_windows),
dtype=np.float32)
# Ugh !
for i, (ds_name, ds) in enumerate(self.dr.datasets.items()):
self.bias_mat[:,
i] = self.beta * (0.5 * ds.kappa * (self.all_data))
log.info("saving integrated autocorr times (in ps) to 'autocorr.dat'")
np.savetxt(
'autocorr.dat',
self.autocorr,
header='integrated autocorrelation times for each window (in ps)')
dr.clearData()
def _unpack_phi_data(self, start, end=None):
self.all_data = np.array([], dtype=np.float32)
self.all_data_N = np.array([]).astype(int)
self.n_samples = np.array([]).astype(int)
if self.autocorr is None:
do_autocorr = True
self.autocorr = np.zeros(self.n_windows)
else:
do_autocorr = False
for i, (ds_name, ds) in enumerate(self.dr.datasets.iteritems()):
if self.ts == None:
self.ts = ds.ts
# Sanity check - every input should have same timestep
else:
np.testing.assert_almost_equal(self.ts, ds.ts)
data = ds.data[start:end]
dataframe = np.array(data['$\~N$'])
dataframe_N = np.array(data['N']).astype(np.int32)
if do_autocorr:
autocorr_len = np.ceil(pymbar.timeseries.integratedAutocorrelationTime(dataframe[:]))
self.autocorr[i] = self.ts * autocorr_len
self.n_samples = np.append(self.n_samples, dataframe.shape[0])
self.all_data = np.append(self.all_data, dataframe)
self.all_data_N = np.append(self.all_data_N, dataframe_N)
self.bias_mat = np.zeros((self.n_tot, self.n_windows), dtype=np.float32)
# Ugh !
for i, (ds_name, ds) in enumerate(self.dr.datasets.iteritems()):
self.bias_mat[:, i] = self.beta*(0.5*ds.kappa*(self.all_data-ds.Nstar)**2 + ds.phi*self.all_data)
dr.clearData()
this_uncorr_ntwid_dat = this_ntwid_dat[rem:].reshape(n_uncorr_sample, g_inef).mean(axis=1)
uncorr_phi_vals.append(this_uncorr_phi)
uncorr_psi_vals.append(this_uncorr_psi)
uncorr_nreg_dat.append(this_uncorr_nreg_dat)
uncorr_ntwid_dat.append(this_uncorr_ntwid_dat)
print(" n_samples: {}, uncorr_n_samples: {}".format(n_sample, n_uncorr_sample))
nreg_dat.append(n_dat['N'])
ntwid_dat.append(n_dat['$\~N$'])
phi_vals.append(this_phi)
psi_vals.append(this_psi)
n_samples.append(n_sample)
uncorr_n_samples.append(n_uncorr_sample)
dr.clearData()
phi_vals = np.concatenate(phi_vals)
psi_vals = np.concatenate(psi_vals)
uncorr_phi_vals = np.concatenate(uncorr_phi_vals)
uncorr_psi_vals = np.concatenate(uncorr_psi_vals)
nreg_dat = np.concatenate(nreg_dat)
ntwid_dat = np.concatenate(ntwid_dat)
uncorr_nreg_dat = np.concatenate(uncorr_nreg_dat)
uncorr_ntwid_dat = np.concatenate(uncorr_ntwid_dat)
assert phi_vals.size == psi_vals.size == ntwid_dat.size == nreg_dat.size
n_samples = np.array(n_samples)
def _unpack_phi_data(self, start, end=None):
self.all_data = np.array([], dtype=DTYPE)
self.all_data_aux = np.array([], dtype=DTYPE)
self.n_samples = np.array([]).astype(int)
# For subvol Ntwid - may not be used.
self.sub_all_data = np.array([], dtype=DTYPE)
self.kappas = np.zeros(self.n_windows, dtype=DTYPE)
self.phis = np.zeros(self.n_windows, dtype=DTYPE)
self.Nstars = np.zeros(self.n_windows, dtype=DTYPE)
self.avg_N = np.zeros(self.n_windows, dtype=DTYPE)
## May not be used
self.sub_kappas = np.zeros_like(self.kappas)
self.sub_phis = np.zeros_like(self.phis)
self.sub_Nstars = np.zeros_like(self.Nstars)
# If autocorr not set, set it ouselves
if self.calc_autocorr:
self.autocorr = np.zeros(self.n_windows)
for i, (ds_name, ds) in enumerate(self.dr.datasets.items()):
log.info("Extracting {}th dataset {}".format(i, ds_name))
if self.ts == None:
self.ts = ds.ts
# Sanity check - every input should have same timestep
else:
np.testing.assert_almost_equal(self.ts, ds.ts)
data = ds.data[start:end]
dataframe = np.array(data['$\~N$'])
dataframe_N = np.array(data['N']).astype(np.int32)
if self.calc_autocorr:
autocorr_len = np.ceil(
pymbar.timeseries.integratedAutocorrelationTime(
dataframe[:]))
self.autocorr[i] = ds.ts * autocorr_len
self.n_samples = np.append(self.n_samples, dataframe.shape[0])
self.all_data = np.append(self.all_data, dataframe)
self.all_data_aux = np.append(self.all_data_aux, dataframe_N)
self.kappas[i] = ds.kappa
self.phis[i] = ds.phi
self.Nstars[i] = ds.Nstar
self.avg_N[i] = dataframe.mean()
log.info(" kappa: {:.5f} phi: {:.5f} Nstar: {:.2f}".format(
ds.kappa, ds.phi, ds.Nstar))
# Load in sub vol data, if provided.
if self.phisub:
sub_data = ds.ds_sub.data[start:end]
sub_dataframe = np.array(sub_data['$\~N$'])
self.sub_all_data = np.append(self.sub_all_data, sub_dataframe)
self.sub_kappas[i] = ds.ds_sub.kappa
self.sub_phis[i] = ds.ds_sub.kappa
self.sub_Nstars[i] = ds.ds_sub.Nstar
self.bias_mat = np.zeros((self.n_tot, self.n_windows),
dtype=np.float32)
# Ugh !
for i, (ds_name, ds) in enumerate(self.dr.datasets.items()):
self.bias_mat[:, i] = self.beta * (0.5 * ds.kappa *
(self.all_data - ds.Nstar)**2 +
ds.phi * self.all_data)
if self.phisub:
self.bias_mat[:, i] += self.beta * (
0.5 * ds.ds_sub.kappa *
(self.sub_all_data - ds.ds_sub.Nstar)**2 +
ds.ds_sub.phi * self.sub_all_data)
log.info("saving integrated autocorr times (in ps) to 'autocorr.dat'")
np.savetxt(
'autocorr.dat',
self.autocorr,
header='integrated autocorrelation times for each window (in ps)')
dr.clearData()
