[pymvpa] Fw: Hyperalignment
salim_alwasity at yahoo.com
Thu Dec 1 00:06:53 UTC 2016
1. My datasets have similar number of features and every time I ran (hyper=Hyperalignment ( )), the reference is the first dataset ds_h (0).
2. I also did (hyper= Hyperalignment (ref_ds=0)) for the second scenario and still getting different result tha scenario 1.
Sent from Yahoo Mail on Android
On Wed, 30 Nov, 2016 at 11:14 pm, Yaroslav Halchenko<yoh at onerussian.com> wrote: On November 30, 2016 5:54:36 PM EST, Salim Al-wasity <salim_alwasity at yahoo.com> wrote:
I am running a Hyperalignment analysis to investigate theeffect of the reference subject on the common model and between subject classification. I ran two scenarios to double check my results, and in both cases I got different classification accuracies.If I am not wrong, the below scenarios must giving me identical results.I had 10 subjects whose Hyperalignment data are stored in (ds_h) and task data in (ds_task) which I am classifying
hyper=Hyperalignment(ref_ds=2)hypermaps=hyper(ds_h)ds_hyper = [ hypmaps[i].forward(d_all) for i, d_all in enumerate(ds_task)]..........Then continue with classification
Scenario-2-:new_ds_h=[ds_h, ds_h, ds_h, ds_h, ds_h, ds_h, ds_h, ds_h, ds_h, ds_h]hyper=Hyperalignment()hyper.train(new_ds_h)hypermaps=hyper(ds_h)ds_hyper = [ hypmaps[i].forward(d_all) for i, d_all in enumerate(ds_task)]..........Then continue with classification
Pkg-ExpPsy-PyMVPA mailing list
Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
ref_ds : int or None, optional
Index of a dataset to use as 1st-level common space reference. If None, then the dataset with the maximum number of features is used. Constraints: (value must be in range [0, inf], and value must be convertible to type ‘int’), or value must be None. [Default: None]
So utter is not necessarily the first (0th) dataset which is chosen... So you get the same results if your set ref_ds=0
You could enable_ca=["chosen_ref_ds"] and then check ca.chosen_ref_ds on which one is chosen it you don't explicitly specify any
Sent from a phone which beats iPhone.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Pkg-ExpPsy-PyMVPA