Index

Module Index

Search Page

NpPlus Numpy enhancements

Homepage: http://github.com/dhmunro/npplus

PyPI: https://pypi.python.org/pypi/npplus

Summary

Numpy and pyplot enhancements and alternatives.

  • A piecewise polynomial class npplus.pwpoly.PwPoly, a more practical alternative to the scipy.interpolate.PPoly. A PwPoly instance p is naturally callable with p(x) returning the value of the piecewise polynomial function. You can combine PwPoly instances p and q with arithmetic operations, so that 3*p*q - p/2 + 0.5 is a new PwPoly. Integration and differentiation p.integ() and p.deriv() also return new PwPoly instances. A PwPoly may also represent a curve in multi dimensional space, so that p(x) has leading dimensions before the dimensions of x; in that case, p[i] is also a PwPoly. For scalar PwPoly instances, p.roots(value) returns the list of all x such that p(x) == value.

  • Provides spline and splfit functions returning PwPoly instances:

    spline(x, y), see npplus.pwpoly.spline()

    The natural cubic spline through points (x, y). You can specify other boundary conditions and any degree polynomials.

    splfit(xk, x, y), see npplus.pwpoly.splfit()

    The cubic spline with knot points at xk, which is the least squares best fit to points (x, y). You can specify boundary conditions and any degree polynomials. Optionally returns the Lagrange multipliers for all continuity abd boundary constraints.

    Unlike the scipy.interpolate spline functions, these do not use the compiled fitpack functions, only numpy and the scipy solve_banded functions. You can mine this code if you have a variant problem. Both functions accept a per=1 keyword to return periodic piecewise linear functions. Both functions accept multi-dimensional y values to return curves in multi-dimensional space. The splfit function allows you to specify standard deviations for the y values.

  • Variants npplus.pwpoly.pline() and npplus.pwpoly.plfit() with simplified arguments for the important special case of piecewise linear polylines.

  • Simple interfaces for linear and non-linear least squares fitting:

    regress(data, m1, m2, ...), see npplus.lsqfit.regress()

    Return the coefficients p of a linear model p[1]*m1+p[2]*m2+... that best fit the given data in a least squares sense. Each of the m1, m2, … must be conformable with data. Optionally returns convariances and other fit statistics.

    levmar(data, f, p0, args), see npplus.lsqfit.levmar()

    Returns a callable m such that m(args) is the best fit of a parametrized non-linear function f(p,args) to the given data. The args are the independent variables of the family of models, and p0 is the intial guess for p such that f(p,args) ~ data. The best fit parameters themselves are m.p, and other methods of m provide a complete statistical description of the fit.

  • A decorator npplus.pcwise.pcwise() to aid writing functions of one variable x which have different algorithms in different domains of x:

    @pcwise
def fun(x):
    def funlo(x):
        return 1. - x
    def funmid(x):
        return numpy.sin(x)
    def funhi(x):
        return x**2 - 1.
    return funlo, xa, funmid, xb, funhi

  • Versions of stack and concatenate that broadcast their components, a generalized linspace, and a repaired logspace:

    a_(a1, a2, ...), see npplus.basic.a_()

    Stack arrays along a new axis, broadcasting first if needed.

    cat_(a1, a2, ...), see npplus.basic.cat_()

    Join arrays along a given axis, brodcasting first if needed.

    span(a, b, n), see npplus.basic.span()

    Like linspace(a,b,n), except a and b may have any conformable shapes to generate a line in multi-dimensional space.

    spanl(a, b, n), see npplus.basic.spanl()

    Like logspace(log10(a),log10(b),n), except a and b may be have any conformable shapes.

  • Elementwise min, max, and abs with any number of arguments:

    max_(a1, a2, ...), see npplus.basic.max_()

    Elementwise max with any number of conformable array-like arguments.

    min_(a1, a2, ...), see npplus.basic.min_()

    Elementwise min with any number of conformable array-like arguments.

    abs_(a1, a2, ...), see npplus.basic.abs_()

    Elementwise linalg.norm with any number of conformable array-like arguments.

  • Rank preserving axis methods for finite difference operations to supplement diff. For example, cum(zcen(y)*diff(x)) is a finite difference indefinite integral:

    cum(x), see npplus.basic.cum()

    cumsum with prepended 0, an inverse of diff

    zcen(x), see npplus.basic.zcen()

    pairwise averages to go with diff (pairwise differences)

    pcen(x), see npplus.basic.pcen()

    zcen, but copy endpoints

  • An attribute-dict class npplus.itemattr.ADict. An ADict instance ad is a mapping whose items can be accessed either as items ad['name'] or as attributes ad.name.

  • A npplus.interactive.reloadx() function to simplify debugging a module in an interactive session. Your workflow becomes a loop of edit source, reloadx, and pdb run or post-mortem without any external IDE required.

  • Wrappers for pyplot plotting functions like plot which return unwanted objects and clutter interactive terminals. It is easier to type plt.plot in the rare case you want the object, rather than _=plot every time you don’t.

  • A simple presentation-ready matplotlib style.

  • A module npplus.pyplotx.interactive you can import in PYTHONSTARTUP that gives you the pylab interactive environment plus all the npplus features.