""" This module provides access to higher level functions and constants for ieee special values such as Not a Number (nan) and infinity (inf). >>> from numarray import * The special values are designated using lower case as follows: >> inf inf >> plus_inf inf >> minus_inf -inf >> nan nan >> plus_zero 0.0 >> minus_zero -0.0 Note that the representation of IEEE special values is platform dependent so your Python might, for instance, say 'Infinity' rather than 'inf'. Below, inf is seen to arise as the result of floating point division by 0 and nan is seen to arise from 0 divided by 0: >>> a = arange(2.0) >>> b = a/0 Warning: Encountered invalid numeric result(s) in divide Warning: Encountered divide by zero(s) in divide Here are the results for linux, but the repr problem causes this example to fail for windows: >> b array([ nan, inf]) A curious property of nan is that it does not compare to *itself* as equal (results also from linux): >> b == nan array([0, 0], type=Bool) The isnan(), isinf(), and isfinite() functions return boolean arrays which have the value True where the corresponding predicate holds. These functions detect bit ranges and are therefore more robust than simple equality checks. >>> isnan(b) array([1, 0], type=Bool) >>> isinf(b) array([0, 1], type=Bool) >>> isfinite(b) array([0, 0], type=Bool) Array based indexing provides a convenient way to replace special values: >>> b[isnan(b)] = 999 >>> b[isinf(b)] = 5 >>> b array([ 999., 5.]) Here's an easy approach for compressing your data arrays to remove NaNs: >>> x, y = arange(10.), arange(10.); x[5] = nan; y[6] = nan; >>> keep = ~isnan(x) & ~isnan(y) >>> x[keep] array([ 0., 1., 2., 3., 4., 7., 8., 9.]) >>> y[keep] array([ 0., 1., 2., 3., 4., 7., 8., 9.]) ======================================================================= # >>> inf # the repr() of inf may vary from platform to platform # inf # >>> nan # the repr() of nan may vary from platform to platform # nan # Create a couple inf values in 4,4 array >>> a=arange(16.0, shape=(4,4)) >>> a[2,3] = 0.0 >>> b = 1/a Warning: Encountered divide by zero(s) in divide # Locate the positions of the inf values >>> getinf(b) (array([0, 2]), array([0, 3])) # Change the inf values to something else >>> isinf(b) array([[1, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 1], [0, 0, 0, 0]], type=Bool) >>> isinf(inf) 1 >>> isinf(1) 0 >>> isinf(nan) 0 >>> isfinite(inf) 0 >>> isfinite(1) 1 >>> isfinite(nan) 0 >>> isnan(inf) 0 >>> isnan(1) 0 >>> isnan(nan) 1 >>> isfinite(b) array([[0, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 0], [1, 1, 1, 1]], type=Bool) >>> a[getinf(b)] = 999 >>> a array([[ 999., 1., 2., 3.], [ 4., 5., 6., 7.], [ 8., 9., 10., 999.], [ 12., 13., 14., 15.]]) # Set a bunch of locations to a special value >>> a[0,1] = nan; a[1,2] = nan; a[2,3] = nan >>> getnan(a) (array([0, 1, 2]), array([1, 2, 3])) IEEE Special Value support 32-bit >>> import ieeespecial >>> a = arange(5.0, type=Float32) >>> b = (a*a)/a Warning: Encountered invalid numeric result(s) in divide >>> ieeemask(b, NAN) array([1, 0, 0, 0, 0], type=Bool) >>> ieeemask(b, NUMBER) array([0, 1, 1, 1, 1], type=Bool) >>> index(b, NAN) (array([0]),) >>> getnan(b) (array([0]),) >>> setnan(b, 42.0) >>> b[0] 42.0 >>> a = arange(1.0, 6.0, type=Float32) >>> b = a/zeros((5,), type=Float32) Warning: Encountered divide by zero(s) in divide >>> ieeemask(b, POS_INFINITY) array([1, 1, 1, 1, 1], type=Bool) >>> ieeemask(b, NEG_INFINITY) array([0, 0, 0, 0, 0], type=Bool) >>> ieeemask(b, INFINITY) array([1, 1, 1, 1, 1], type=Bool) >>> b = (-a)/zeros((5,), type=Float32) Warning: Encountered divide by zero(s) in divide >>> ieeemask(b, POS_INFINITY) array([0, 0, 0, 0, 0], type=Bool) >>> ieeemask(b, NEG_INFINITY) array([1, 1, 1, 1, 1], type=Bool) >>> ieeemask(b, INFINITY) array([1, 1, 1, 1, 1], type=Bool) >>> ieeemask(b, NUMBER) array([0, 0, 0, 0, 0], type=Bool) >>> ieeemask(array([0], type=Float32), POS_ZERO) array([1], type=Bool) >>> ieeemask(array([0], type=Float32), NEG_ZERO) array([0], type=Bool) >>> ieeemask(array([0], type=Float32), ZERO) array([1], type=Bool) >>> neginf = (array([-1],type=Float32)/array([0], type=Float32)) Warning: Encountered divide by zero(s) in divide >>> negzero = array([1], type=Float32)/neginf >>> ieeemask(negzero, POS_ZERO) array([0], type=Bool) >>> ieeemask(negzero, NEG_ZERO) array([1], type=Bool) >>> ieeemask(array([-0], type=Float32), ZERO) array([1], type=Bool) IEEE Special Value support 64-bit >>> import ieeespecial >>> a = arange(5.0, type=Float64) >>> b = (a*a)/a Warning: Encountered invalid numeric result(s) in divide >>> ieeemask(b, NAN) array([1, 0, 0, 0, 0], type=Bool) >>> ieeemask(b, NUMBER) array([0, 1, 1, 1, 1], type=Bool) >>> index(b, NAN) (array([0]),) >>> getnan(b) (array([0]),) >>> setnan(b, 42.0) >>> b[0] 42.0 >>> a = arange(1.0, 6.0, type=Float64) >>> b = a/zeros((5,), type=Float64) Warning: Encountered divide by zero(s) in divide >>> ieeemask(b, POS_INFINITY) array([1, 1, 1, 1, 1], type=Bool) >>> ieeemask(b, NEG_INFINITY) array([0, 0, 0, 0, 0], type=Bool) >>> ieeemask(b, INFINITY) array([1, 1, 1, 1, 1], type=Bool) >>> b = (-a)/zeros((5,), type=Float64) Warning: Encountered divide by zero(s) in divide >>> ieeemask(b, POS_INFINITY) array([0, 0, 0, 0, 0], type=Bool) >>> ieeemask(b, NEG_INFINITY) array([1, 1, 1, 1, 1], type=Bool) >>> ieeemask(b, INFINITY) array([1, 1, 1, 1, 1], type=Bool) >>> ieeemask(b, NUMBER) array([0, 0, 0, 0, 0], type=Bool) >>> ieeemask(array([0], type=Float64), POS_ZERO) array([1], type=Bool) >>> ieeemask(array([0], type=Float64), NEG_ZERO) array([0], type=Bool) >>> ieeemask(array([0], type=Float64), ZERO) array([1], type=Bool) >>> neginf = (array([-1],type=Float64)/array([0], type=Float64)) Warning: Encountered divide by zero(s) in divide >>> negzero = array([1], type=Float64)/neginf >>> ieeemask(negzero, POS_ZERO) array([0], type=Bool) >>> ieeemask(negzero, NEG_ZERO) array([1], type=Bool) >>> ieeemask(array([-0], type=Float64), ZERO) array([1], type=Bool) """ import numarrayall as _na from numarray.ufunc import isnan # Define *ieee special values* _na.Error.pushMode(all="ignore") plus_inf = inf = (_na.array(1.0)/_na.array(0.0))[()] minus_inf = (_na.array(-1.0)/_na.array(0.0))[()] nan = (_na.array(0.0)/_na.array(0.0))[()] plus_zero = zero = 0.0 minus_zero = (_na.array(-1.0)*0.0)[()] _na.Error.popMode() # Define *mask condition bits* class _IeeeMaskBit(_na.NumArray): pass def _BIT(x): a = _na.array((1 << x), type=_na.Int32) a.__class__ = _IeeeMaskBit return a POS_QUIET_NAN = _BIT(0) NEG_QUIET_NAN = _BIT(1) POS_SIGNAL_NAN = _BIT(2) NEG_SIGNAL_NAN = _BIT(3) POS_INFINITY = _BIT(4) NEG_INFINITY = _BIT(5) POS_DENORMALIZED = _BIT(6) NEG_DENORMALIZED = _BIT(7) POS_NORMALIZED = _BIT(8) NEG_NORMALIZED = _BIT(9) POS_ZERO = _BIT(10) NEG_ZERO = _BIT(11) INDETERM = _BIT(12) BUG = _BIT(15) NAN = POS_QUIET_NAN | NEG_QUIET_NAN | POS_SIGNAL_NAN | NEG_SIGNAL_NAN | INDETERM INFINITY = POS_INFINITY | NEG_INFINITY SPECIAL = NAN | INFINITY NORMALIZED = POS_NORMALIZED | NEG_NORMALIZED DENORMALIZED = POS_DENORMALIZED | NEG_DENORMALIZED ZERO = POS_ZERO | NEG_ZERO NUMBER = NORMALIZED | DENORMALIZED | ZERO FINITE = NUMBER def mask(a, m): """mask(a, m) returns the values of 'a' satisfying category 'm'. mask does a parallel check for values which are not classifyable by the categorization code, raising a RuntimeError exception if any are found. """ a = _na.asarray(a) if isinstance(a.type(), _na.IntegralType): a = a.astype('Float64') if isinstance(a.type(), _na.ComplexType): f = _na.ieeemask(a.real, m) | _na.ieeemask(a.imag, m) g = _na.ieeemask(a.real, BUG) | _na.ieeemask(a.imag, BUG) else: f = _na.ieeemask(a, m) g = _na.ieeemask(a, BUG) if _na.bitwise_or.reduce(_na.ravel(g)) != 0: raise RuntimeError("Unclassifyable floating point values.") if f.rank == 0: f = f[()] return f def index(a, msk): """index returns the tuple of indices where the values satisfy 'mask'""" return _na.nonzero(mask(a, msk)) def getinf(a): """getinf returns a tuple of indices of 'a' where the values are infinite.""" return index(a, INFINITY) def setinf(a, value): """setinf sets elements of 'a' which are infinite to 'value' instead. DEPRECATED: use 'a[getinf(a)] = value' instead. """ _na.put(a, getinf(a), value) def isinf(a): """Idenitfies elements of 'a' which are infinity. """ return mask(a, INFINITY) def getposinf(a): """getposinf returns a tuple of indices of 'a' where the values are +inf.""" return index(a, POS_INFINITY) def getneginf(a): """getneginf returns a tuple of indices of 'a' where the values are -inf.""" return index(a, NEG_INFINITY) def getnan(a): """getnan returns a tuple of indices of 'a' where the values are not-a-numbers""" return _na.nonzero(isnan(a)) def setnan(a, value): """setnan sets elements of 'a' which are NANs to 'value' instead. DEPRECATED: use 'a[getnan(a)] = value' instead. """ a[isnan(a)]= value #def isnan(a): # """Idenitfies elements of 'a' which are NANs, not a number. # """ # return _na.isnan(a) # # This function has been replaced by isnan macro added to the numarray.ufunc module. def isfinite(a): """Identifies elements of an array which are neither nan nor infinity.""" return _na.logical_not(isinf(a)| isnan(a)) def getbug(a): """getbug returns a tuple of indices of 'a' where the values are not classifyable.""" return index(a, BUG) def test(): import doctest, ieeespecial return doctest.testmod(ieeespecial)