# store the whole datasets warehouse in one hdf5 file
hdf = h5py.File(pathjoin(tempdir, 'myhdf5.hdf5'), 'w')
for d in datasets:
obj2hdf(hdf, datasets[d], d)
hdf.close()
hdf = h5py.File(pathjoin(tempdir, 'myhdf5.hdf5'), 'r')
rc_ds = {}
for d in hdf:
rc_ds[d] = hdf2obj(hdf[d])
hdf.close()
#cleanup temp dir
shutil.rmtree(tempdir, ignore_errors=True)
# return the reconstructed datasets (for use in datasets warehouse)
return rc_ds
datasets = generate_testing_datasets(specs)
if cfg.getboolean('tests', 'use hdf datasets', False):
if not externals.exists('h5py'):
raise RuntimeError(
"Cannot perform HDF5 dump of all datasets in the warehouse, "
"because 'h5py' is not available")
datasets = saveload_warehouse()
print "Replaced all dataset warehouse for HDF5 loaded alternative."
# an output dataset (something x nfeatures of input ds)
raise NotImplementedError
def _forward_dataset(self, ds):
reg_names, X = self._build_design(ds)
model, out = self._fit_model(ds, X, reg_names)
out.fa.update(ds.fa)
out.a.update(ds.a) # this last one might be a bit to opportunistic
if self.params.return_design:
if not len(out) == len(X.T):
raise ValueError("cannot include GLM regressors as sample "
"attributes (dataset probably contains "
"something other than parameter estimates")
out.sa['regressors'] = X.T
if self.params.return_model:
out.a['model'] = model
return out
# TODO: this is not unreasonable, forward+reverse cycle throws away residuals...
#def _reverse_dataset(self, ds):
# reconstruct timeseries from model fit
from mvpa2 import externals
if externals.exists('nipy'):
from .nipy_glm import NiPyGLMMapper
__all__.append('NiPyGLMMapper')
if externals.exists('statsmodels'):
from .statsmodels_glm import StatsmodelsGLMMapper
__all__.append('StatsmodelsGLMMapper')
# an output dataset (something x nfeatures of input ds)
raise NotImplementedError
def _forward_dataset(self, ds):
reg_names, X = self._build_design(ds)
model, out = self._fit_model(ds, X, reg_names)
out.fa.update(ds.fa)
out.a.update(ds.a) # this last one might be a bit to opportunistic
if self.params.return_design:
if not len(out) == len(X.T):
raise ValueError("cannot include GLM regressors as sample "
"attributes (dataset probably contains "
"something other than parameter estimates")
out.sa['regressors'] = X.T
if self.params.return_model:
out.a['model'] = model
return out
# TODO: this is not unreasonable, forward+reverse cycle throws away residuals...
#def _reverse_dataset(self, ds):
# reconstruct timeseries from model fit
from mvpa2 import externals
if externals.exists('nipy'):
from .nipy_glm import NiPyGLMMapper
__all__.append('NiPyGLMMapper')
if externals.exists('statsmodels'):
from .statsmodels_glm import StatsmodelsGLMMapper
__all__.append('StatsmodelsGLMMapper')
Dimensionality of target space
data : array, optional
Some data (should have rank high enough) to derive
rotation
"""
if nt is None:
nt = ns
# figure out some "random" rotation
d = max(ns, nt)
if data is None:
data = np.random.normal(size=(d*10, d))
_u, _s, _vh = np.linalg.svd(data[:, :d])
R = _vh[:ns, :nt]
if ns == nt:
# Test if it is indeed a rotation matrix ;)
# Lets flip first axis if necessary
if np.linalg.det(R) < 0:
R[:, 0] *= -1.0
return R
datasets = generate_testing_datasets(specs)
if cfg.getboolean('tests', 'use hdf datasets', False):
if not externals.exists('h5py'):
raise RuntimeError(
"Cannot perform HDF5 dump of all datasets in the warehouse, "
"because 'h5py' is not available")
datasets = saveload_warehouse()
print "Replaced all dataset warehouse for HDF5 loaded alternative."
