part of numpy.oldnumeric from numpy v 1.8 was integrated into Biskit (Biskit.oldnumeric)
all references to numpy.oldnumeric replaced by calls to Biskit.oldnumeric or numpy
stats and pstat modules (Biskit.Statistics) removed -- these had been copied from elsewhere and only served as (error-prone) fall-back option in case scipy was not installed (affects ROCalyzer.py)
scipy is now considered a hard dependency
addressed FutureWarnings from numpy concerning array comparison with None replacing == None by is None actually leads to a significant performance gain (about 20% shaved off core testing suite).
Cut all ties with Scientific Python
Scientific Python has the same numpy.oldnumeric problem just bigger and Konrad doesn't seem to want / have the time / to address it. Biskit relies on Scientific.IO.PDBFile for pdb file reading and writing. This part has been extracted and integrated into Biskit:
small part of Scientific.IO sub-package integrated into Biskit as Biskit.ScientificIO
ScientificIO uncoupled from numpy.oldnumeric (only "Vector" and "Tensor" class affected
and these could be replaced by numpy.array or even list)
high-level Structure class removed
a few modules from very early Biskit days were actually using Scientific.IO.Structure -- deprecated these modules:
ProsaII (removed, superseeded by Prosa2003 since 10 years)
ChainCleaner/ChainSeperator/ChainWriter + pdb2xplor.py (deprecated but kept)
Table (removed, needed by ProsaII only)
Testing status
python test.py -e pvm exe
yields the following:
SUMMARY:
=======
A total of 112 tests from 117 modules were run.
- 110 passed
- 2 failed
- error : Biskit.ProfileCollection.Test.test_concat
- error : Biskit.PDBModel.Test.test_compareChains
DONE
The two test failures were already happening in the master branch (in fact, one third failure spontaneously disappeared during the numpy conversion). One additional failure seems to be specific to MacOSX. These 2 / 3 will be adressed on master.
I have not so far tested the full suite with third-party software installed. Code coverage of the "core" testing suite is reported with 55%.
Make Biskit compatible with latest numpy version.
This includes:
FutureWarnings
from numpy concerning array comparison withNone
replacing== None
byis None
actually leads to a significant performance gain (about 20% shaved off core testing suite).Cut all ties with Scientific Python
Scientific Python has the same numpy.oldnumeric problem just bigger and Konrad doesn't seem to want / have the time / to address it. Biskit relies on Scientific.IO.PDBFile for pdb file reading and writing. This part has been extracted and integrated into Biskit:
Testing status
yields the following:
The two test failures were already happening in the master branch (in fact, one third failure spontaneously disappeared during the numpy conversion). One additional failure seems to be specific to MacOSX. These 2 / 3 will be adressed on
master
.I have not so far tested the full suite with third-party software installed. Code coverage of the "core" testing suite is reported with 55%.