[pymvpa] Accuracy for Each Timepoint in Event-Related Analysis

[Nick Oosterhof]

On Jan 10, 2014, at 8:15 PM, Daniel P. Bliss wrote:

> I think this question was based on a misinterpretation of what the event-related analysis actually does.  I was assuming that it conducted a separate classification analysis for each timepoint, but now I'm inferring that it conducts a single analysis, treating the timepoints as independent features.  Is that true?  

Not necessarily independent, but yes, data from different timepoints is used in the analysis.

> Is there a way to automate separate analyses for different timepoints in PyMVPA?

Yes, but it depends a bit on what type analysis you want to do.
I assume you have a dataset ds with trials that have N timepoints each. Add a sample attribute, say 'time_in_trial', and assign to each sample a value from 1 to N. Then do

ds_splits=split_by_sample_attribute(ds, 'time_in_trial')

and run your analysis separately for each dataset in ds_splits, e.g.

ds_results=[]
for ds_split in ds_splits:
   # your code here to get ds_result
  ds_results.append(ds_result)

result=hstack(ds_results)