An additional sqrt() was introduced in 703dc23 --- minpack takes the sqrt internally, so it should not be done again. The routines can still fail if the problem is badly scaled. The test is here changed to be less badly scaled. The correct choice of epsfcn is problem-dependent and in general needs to be left to the user. Fixes gh-2639
…e same information is in the docstring
The documentation describing the `sigma` parameter of the `optimize.curve_fit` function was updated to remove the reference to representing the standard deviation of the data. This is misleading because the `sigma` value is used purely as a relative weighting factor. See also: http://mail.scipy.org/pipermail/scipy-user/2013-February/034191.html
…fixes Trac #920) In particular, if the objective function returns float32, use a large enough epsilon for gradient estimation.
…inverting curvature matrix
…nused options Fix also unused option warnings in tests.
Fixes a number of tickets - #1528, #1385, #1576, #1632 - and all test warnings. 61c61e9 TST: fix ResourceWarnings under python 3.2. Closes #1385. 8231c9f TST: clean up cluster.vq test file. d6c56db TST: io.wavfile: filter out warning. e089c85 MAINT: fix some build warnings. Closes ticket 1528. d56afef TST: use warning filters correctly in all tests. dcfc275 TST: filter test warnings from bivariate spline tests in interpolate. 8a8a0a2 BUG: py3k: remove unnecessary use of types.NoneType; removed in py3k. e60098c BUG: py3k: func_name attribute of functions was removed, use __name__. dc3a6f6 STY: fix odd use of array.nonzero(). Closes #1632. 242106b STY: simplify opening data files in cluster tests.
…ssing 'abs' around relerr.
…e. don't special-case the n==1 call, but always return an array).