A simple start

Here we show how to perform a simple cross-validated classification analysis with PyMVPA. This script is the exact equivalent of the A simple start (on the command line) example, but using the Python API instead of the command line interface.

First, we import the PyMVPA suite to enable all PyMVPA building blocks

from mvpa2.suite import *

Now we load an example fMRI dataset that comes with PyMVPA. It has some attributes associated with each volume, and is masked to exclude voxels outside of the brain.

# load PyMVPA example dataset with literal labels
dataset = load_example_fmri_dataset(literal=True)

Next we remove linear trends by polynomial regression for each voxel and each chunk (recording run) of the dataset individually.

poly_detrend(dataset, polyord=1, chunks_attr='chunks')

For this example we are only interested in data samples that correspond to the face or to the house condition.

dataset = dataset[np.array([l in ['face', 'house'] for l in dataset.sa.targets],
                          dtype='bool')]

The setup for our cross-validation analysis include the selection of a classifier, and a partitioning scheme, and an error function to convert literal predictions into a quantitative performance metric.

cv = CrossValidation(SMLR(), OddEvenPartitioner(), errorfx=mean_mismatch_error)
error = cv(dataset)

The resulting dataset contains the computed accuracy.

# UC: unique chunks, UT: unique targets
print "Error for %i-fold cross-validation on %i-class problem: %f" \
      % (len(dataset.UC), len(dataset.UT), np.mean(error))

See also

The full source code of this example is included in the PyMVPA source distribution (doc/examples/start_easy.py).