~mnemonic-dev/mnemonic/trunk : changes

~mnemonic-dev/mnemonic/trunk » Changes from revision 31

From Revision 31 to 12

Rev	Summary	Authors	Date
31	Initial port of semantic_crp from matlab Initial port of semantic_crp from matlab	Richard Lawrence	15 years ago
30	Began changing transitions functions to accomodate multiple ... Began changing transitions functions to accomodate multiple masks. conditional_transitions now accepts two additional keyword arguments, from_mask and to_mask, in addition to the mask keyword. It will generate from_mask and/or to_mask from mask as necessary, but from_mask and to_mask will take precedence over mask if all three are given. Made some performance improvements as well (eliminated use of priors in conditional_transitions, initialized arrays to filler value instead of building them up at each iteration, etc.). Still to do: change documentation and interface to transitions(), make unit tests for new behavior.	Richard Lawrence	15 years ago
29	Merged in conditional-revamp branch Merged in conditional-revamp branch	Richard Lawrence	15 years ago
28	Added io roadmap Added io roadmap	Richard Lawrence	15 years ago
27	Changed mnemonic.helpers.transitions functions to preserve o... Changed mnemonic.helpers.transitions functions to preserve output information. These functions now insert a "filler" value (default: numpy.NaN) into transitions arrays at positions where a transition is not calculated (e.g., where the from_pt is masked out). This guarantees that the output of transitions() and the transitions array output by conditional_transitions() have a length equal to the number of items in the input from which a transition is computed (which is len(input) - step), and that the value at position i in the output is either the computed transition from input[i], or the filler. For example, the previous behavior was: >>> data = array([4, 3, 2, 1, 5]) >>> mask = array([True, True, False, True, True]) >>> transitions(data, mask=mask) array([-1, 4]) # represents the transitions 4->3 and 1->5 Now, the result is: >>> transitions(data, mask=mask, filler=numpy.NaN) array([-1., NaN, NaN, 4.]) # shows that 3->2 and 2->1 were skipped There is a lingering issue with types here, because inserting the filler directly is pretty much asking for a type explosion, but I haven't figured out how to sanity check it yet; the docstring just has a warning. Changed docstrings and tests for transitions() and conditional_transitions().	Richard Lawrence	15 years ago
26	Added some quick output tests for mnemonic.helpers.transitio... Added some quick output tests for mnemonic.helpers.transitions.* Added an iterative implementation of the cond_helper function inside mnemonic.helpers.transitions.conditional_transitions. This is to stay in sync with the MATLAB version, and will hopefully speed things up slightly. I'll probably remove the old tail-recursive version in the next revision. Sigh.	Richard Lawrence	15 years ago
25	Added some tests for mnemonic.helpers.masks; a few slight co... Added some tests for mnemonic.helpers.masks; a few slight code changes cropped up during testing.	Richard Lawrence	15 years ago
24	Added tests for apply_by_index in helpers.arrayops Added tests for apply_by_index in helpers.arrayops	Richard Lawrence	15 years ago
23	Added output value test for apply_by_index Added output value test for apply_by_index	Richard Lawrence	15 years ago
22	Tiny changes in crp.py to make possible_transitions less mag... Tiny changes in crp.py to make possible_transitions less magical	Richard Lawrence	15 years ago
21	Small changes in crp docstrings Small changes in crp docstrings	Richard Lawrence	15 years ago
20	Changed helpers.arrayops.collect() and count() to operate fr... Changed helpers.arrayops.collect() and count() to operate freely on arrays with 1 or more dimensions, because converting 1-d to 2-d arrays just to use collect() was going to drive me crazy. Changed collect_by_row to operate on any array with 2 or more dimensions. Changed tests accordingly. Changed paradigms.fr.crp.crp_for_subj to use collect() rather than gather_and_order_lags(), which should now be defunct.	Richard Lawrence	15 years ago
19	Changed implementation of paradigms.fr.spc.spc_for_subj to u... Changed implementation of paradigms.fr.spc.spc_for_subj to use helpers.arrayops.collect instead of numpy.histogram. Changed collect() and collect_by_row() to ensure that array parameter is 2-D.	Richard Lawrence	15 years ago
18	Small changes: line widths, etc. Small changes: line widths, etc.	Richard Lawrence	15 years ago
17	Added tests for helpers.arrayops.{count, collect, collect_by... Added tests for helpers.arrayops.{count, collect, collect_by_row}	Richard Lawrence	15 years ago
16	Added mnemonic.helpers.tests for real, this time... Added mnemonic.helpers.tests for real, this time...	Richard Lawrence	15 years ago
15	Re-wrote helpers.arrayops.{count, collect, collect_by_row} t... Re-wrote helpers.arrayops.{count, collect, collect_by_row} to have non-naive implementations, and added docstrings and doctests.	Richard Lawrence	15 years ago
14	Added test_crp, which I forgot to do before...oops. Added test_crp, which I forgot to do before...oops.	Richard Lawrence	15 years ago
13	Re-wrote mnemonic.helpers.arrayops.eq_with_nans to include p... Re-wrote mnemonic.helpers.arrayops.eq_with_nans to include proper sanity checking and not use numpy.ma. Added mnemonic.helpers.tests and wrote unit tests for arrayops.eq_with_nans. Note on last commit: I apparently forgot to add mnemonic.paradigms.fr.tests.test_crp before committing...so don't rollback before this revision if you need those tests.	Richard Lawrence	15 years ago
12	Finished writing tests for crp module. Finished writing tests for crp module. Made sanity checking more consistent between spc and crp.	Richard Lawrence	15 years ago

Older »