Release Notes – PyMVPA 0.5¶
For The Impatient¶
- Datasets are no longer relatively static objects, but become flexible
multi-purpose containers that can handle attributes for samples, feature,
or whole datasets. There is some inital support for other datatypes than
NumPy’s
ndarrays
, e.g. sparse matrices.
Critical API Changes¶
states
->ca
(for conditional attributes). All attributes stored in collections (parameters for Classifiers in.params
, states in.ca
) should be accessed not at top level of the object but through a collection.- Dataset: behaves more like a NumPy array. No specialized Dataset classes,
but constructors
- MaskedDataset ->
dataset_wizard
- NiftiDataset ->
fmri_dataset
- ERNiftiDataset ->
fmri_dataset
+eventrelated_dataset
(see event-related analysis example)
- MaskedDataset ->
- MRI volumes: 3,4D volumes (and coordinates) are exposed with following order of axes: t,x,y,z. Previously we followed a convention of t,z,y,x order of axis in volume data (to be consistent with PyNIfTI).
- Masks (
mask_mapper
)
- now
[1,1,0]
is not the same as[True, True, False]
- We have weird (but consistent) conventions now - classes are CamelCased - factory functions (even for whatever might have been before a class) are in pythonic_style
detrend
->poly_detrend
perchunk=bool
(in zscore/detrend) got refactored intochunks_attr=None or string
to specify on which sample attribute to operate.- internally and as provided by mvpa2.suite,
numpy
is imported asnp
, andpylab
is imported aspl