Permalink
Commits on Jun 4, 2009
  1. scipy 0.7.1 RC1.

    cournape committed Jun 4, 2009
Commits on Jun 3, 2009
  1. Fix bdist_wininst_arch task on win32.

    (cherry picked from commit 5a0c61ea9da2077c5dc08daaa92ab4ca9d14d430)
    cournape committed Jun 3, 2009
  2. Fix paver file to handle bdist targets on windows.

    (cherry picked from commit c20e347ccf18da54a98842ddffba042fec107724)
    cournape committed Jun 3, 2009
Commits on Jun 2, 2009
  1. Workaround for PyInt_Check failures for python 2.6

    If an int is an instance of np.int32, PyInt_Check on it will fail for
    python 2.6 and later. Instead of converting the related argument to a
    'pure' python int in the Data class, I added a workaround at the C level
    to avoid changing anything in the ODR API by accident.
    
    (cherry picked from commit 2256fa8ccefd95362f4e4ea0a23de05fcfc69272)
    cournape committed Jun 2, 2009
Commits on Jun 1, 2009
  1. Change mode from 'wa' to 'w' for array_import.

    Mode 'wa' is not recognized on at least windows (python 2.6 and later),
    and the mode 'wa' does not do what one might expect, that is it looks
    like successful open(filename, 'wa') is equivalent to open(filename,
    'w') on linux at least.
    
    (cherry picked from commit 7d1349335b43a1b2eb7537e14408f83d0a15eb49)
    cournape committed Jun 1, 2009
Commits on May 31, 2009
  1. Fix bdist_superpack paver task (forgot to change name from numpy to s…

    …cipy).
    
    (cherry picked from commit 121e78c2c8e254a6b78cb6ee78c8dd719b0cbc6e)
    cournape committed May 31, 2009
Commits on May 30, 2009
  1. Mark two tests as known failures: corresponding code is numericaly un…

    …stable, and compiler dependent.
    cournape committed May 30, 2009
Commits on May 26, 2009
  1. 0.7.x: enable fixed test

    cournape committed May 26, 2009
  2. 0.7.x: special: ensure NAN/INFINITY are available in amos_wrappers (e…

    …ven on non-C99 compilers)
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5699 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 332a8ddf77e5117f288436471505d2e2fa5b2e27)
    cournape committed May 26, 2009
  3. 0.7.x: special: adjust cephes/yv test tolerances, and mark some as kn…

    …own failures
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5695 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit df44fb8c5b508a7c862cba807ca76d7d58b1ec45)
    cournape committed May 26, 2009
  4. 0.7.x: Remove inline keyword to fix ticket #924.

    Backported: http://svn.scipy.org/svn/scipy/trunk@5677 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit bd5286702db43711d47ae01901d71f9b39ad57b2)
    cournape committed May 26, 2009
  5. 0.7.x: Add some tests for pbdv (cf. #803)

    Backported: http://svn.scipy.org/svn/scipy/trunk@5660 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 746e23519a6881a2e213c788f1c7d2d1df1f4200)
    cournape committed May 26, 2009
  6. 0.7.x: Fix off-by-one bug in pbdv wrappers

    Backported: http://svn.scipy.org/svn/scipy/trunk@5658 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit c6bd1c821a0a09e6204180137d24d888b0f8fc9b)
    cournape committed May 26, 2009
  7. 0.7.x: special: Fix pbdv test

    Backported: http://svn.scipy.org/svn/scipy/trunk@5657 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 6ea644e51bdb7a0061be68b2febfb1f6ea2c4ade)
    cournape committed May 26, 2009
  8. 0.7.x: special/iv: add a more meticulous test against amos

    Backported: http://svn.scipy.org/svn/scipy/trunk@5641 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit b2f897cc749c2db932b50e5f050d10f8909e1669)
    cournape committed May 26, 2009
  9. 0.7.x: Fixing #503: reimplement cephes/iv

    Cephes's implementation of Iv via a relation to hyp1f1 is problematic,
    especially in the asymptotic regime where intermediate results overflow.
    
    This commit introduces a new implementation for the Bessel I function,
    partly converted from the one in the Boost C++ library:
    
    - For large orders, the uniform asymptotic expansion is used.
    
    - Otherwise, use Temme's method.
    
    Performance-wise, this implementation seems to be typically about 2-4
    times faster than the Amos routines.
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5640 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 68ffca175091fb49dcd8040c52c6c8b9c7cce3ac)
    cournape committed May 26, 2009
  10. 0.7.x: cephes/hyperg: restore and explain hyp2f0 asymptotic series te…

    …rmination condition
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5639 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit efab8380ae64ddbf0f479c5b7225bf5723d05cfa)
    cournape committed May 26, 2009
  11. 0.7.x: Fix working for multigammaln doc.

    Backported: http://svn.scipy.org/svn/scipy/trunk@5629 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 23a9c7146883668955354aebc4adcbc093ef42d1)
    cournape committed May 26, 2009
  12. 0.7.x: Update multigammaln docstring.

    Backported: http://svn.scipy.org/svn/scipy/trunk@5627 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 5e38974176cd3470d6e292cc977b65247ebde0fe)
    cournape committed May 26, 2009
  13. 0.7.x: Do not use C99-style initialization.

    Backported: http://svn.scipy.org/svn/scipy/trunk@5607 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 64f6f20958342ba6eff7c63def5c877d70feecf2)
    cournape committed May 26, 2009
  14. 0.7.x: Use PyErr_Warn instead of PyErr_WarnEx (which is not available…

    … in Python 2.4)
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5528 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 6ebfc167ee91377e7b250990d0074ad94b0dc1cd)
    cournape committed May 26, 2009
  15. 0.7.x: special: return infs for Jv/Iv/Yv/Kv overflows in AMOS

    Backported: http://svn.scipy.org/svn/scipy/trunk@5524 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit e422b08fc369f48ee59e0cd779fb93e356d41de4)
    cournape committed May 26, 2009
  16. 0.7.x: Fixing #854: make iv, jv, etc. return NaN when out-of-domain o…

    …r failure
    
    - Fix Cephes iv, jv return values; on DOMAIN return NaN, not zero
    
    - Fix amos_wrappers return value handling: return NaNs when Amos signals
      that it didn't do any computation
    
    - Fix _cephesmodule signatures: all real-valued functions should return
      NaN when the result would be complex, instead of just returning the
      real part.
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5523 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 0295e463f8819e364c0fd4029cd8ca48e6e1a66e)
    cournape committed May 26, 2009
  17. 0.7.x: Fixing #853: also I_v: use correct symmetry for negative order…

    …s (complex x)
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5521 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit f8bcf052d991adeb082a0c9503f89d08b6ede7ea)
    cournape committed May 26, 2009
  18. 0.7.x: Fix #853: use correct symmetries for J_v / Y_v for negative no…

    …n-int. orders
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5516 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit a391787474deb4102b8cdd3e4798de93bf4d5a0e)
    cournape committed May 26, 2009
  19. 0.7.x: Fix #623: better Cephes jv.c:recur continued fraction terminat…

    …ion conditions
    
    There were two problems in the old termination conditions for the
    continued fraction for Jn/Jn-1:
    
    1) It could terminate spuriously on the first iteration, because the
       initial value for `ans` corresponded to a possible value of the fraction
       but not one obtained from the iteration.
    
    2) `jv` could call `recur` with values of `n` and `x` such that the
       continued fraction iteration would not converge and the maximum
       iteration limit was reached.
    
    The code is now more careful with 1) and enforces a minimum number of
    iterations, and 2) is worked around by bumping the maximum iteration
    limit so that the CF should converge for all calls that come from `jv`.
    The iteration limits can be estimated as the number of iterations
    needed is known for this continued fraction.
    
    Additional tests included.
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5515 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 98fdf6af4f0053244667c0b542ec455ae21cb7fb)
    cournape committed May 26, 2009
  20. 0.7.x: Show Cephes errors as Python warnings instead of fprintfing.

    This changes cephes/mtherr.c to signal warnings via PyErr_WarnEx rather
    than fprintf. (Though error printing is disabled by default, as previously.)
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5514 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 8ea62852f764d1719aa65ebe69684fa597c5c357)
    cournape committed May 26, 2009
  21. 0.7.x: Fixing #852: make specfun.f:JYZO work also for N >= 101.

    In addition to removing the checks for N, this requires rewriting
    the Newton iterations in JYZO to be more robust, because for large
    N the initial guess used between successive zeros becomes inaccurate.
    
    Additional tests for jn/yn zeros are included.
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5508 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 527094f56e9c7095dec008994d46ed9ee484a11f)
    cournape committed May 26, 2009
  22. 0.7.x: Fix #852: improve accuracy of jn_zeros by refactoring Specfun …

    …JYNDD, JYNB
    
    Code to calculate the value of Bessel functions of integer order was
    duplicated in specfun.f routines JYNDD and JYNB. The former routine
    didn't do asymptotic expansion, and its results were bad for large
    values of x, resulting to errors in `scipy.special.jn_zeros`.
    
    This commit refactors the common code of these two functions to
    JYNBH, and makes JYZO Newton tolerances stricter. The changes to JYNDD
    also remove the limitation n <= 101.
    
    Improved tests for jn_zeros are included.
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5507 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 8ed30385f492878a24514a8d70cca41fd433ca2c)
    cournape committed May 26, 2009
  23. 0.7.x: Fixing #640: Make Cephes iv and hyperg behave better for large…

    … |z|
    
    Bump iteration limits and handle infs and nans more carefully.
    
    However, since iv is implemented in Cephes using 1F1, it overflows to inf
    at smaller |z| than necessary, as intermediate results are larger than
    floating point max. Fixing this would likely require a complete rewrite.
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5478 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit c9c1abed9d0c02ab9163540102e16cb5be2e3951)
    cournape committed May 26, 2009
  24. 0.7.x: Fix #640: bad output from scipy.special.iv(k,z) for large real z.

    This occurred because Cephes hyperg.c:hyp1f1p estimated the error
    for power series expansion of Kummer confluent 1F1 as zero when
    the number of terms was too large. This commit makes Cephes to
    assume 100% error when the series does not converge, which makes
    it to switch to an asymptotic expansion.
    
    The Cephes hyp1f1 is not exposed by scipy.special (it uses Specfun's
    implementation), but it's used internally by Cephes.
    
    Improved tests for Iv are included.
    
    Backported: http://svn.scipy.org/svn/scipy/trunk@5477 d6536bca-fef9-0310-8506-e4c0a848fbcf
    (cherry picked from commit 617c0a78999bb4dd5e8d664b397ba8eaa73b6136)
    cournape committed May 26, 2009
Commits on May 22, 2009
Commits on May 19, 2009
  1. backport r:5604 use float in mstats.kendalltau

    josef committed May 19, 2009
Commits on May 17, 2009