Whats New

15.12 (Released March 21, 2016)

Highlights

  • now supports SciDB15.12; use branch scidb15.7 for use with SciDB15.7

14.10 (Released November 4, 2014)

Highlights

API Changes in 14.10

  • Slicing an array by a boolean array now produces a sparse result, preserving the location of the selected cells. The collapse method converts this array to the 1D dense array previously returned by boolean masking.
  • The merge() method was changed from a direct AFL call to a high-level join operator. Use robust.merge for the old behavior

14.8 (Released August 22, 2014)

Highlights

  • Support for authenticated and encrypted connections to ScIDB
  • Fixed a bug where uploading large arrays using from_data resulted in scrambled cell locations in the database
  • Proper treatment of elementwise arithmetic on sparse arrays

14.7 (Released August 1, 2014)

Highlights

  • Wider support for wrapping all of SciDB’s built-in datatypes, including strings, datetimes, and nullable values:

    >>> x
    SciDBArray('py1101328071989_00045<f0:string> [i0=0:2,1000,0]')
    >>> x.toarray()
    array([u'abc', u'def', u'ghi'], dtype=object)
    
  • A groupby() method for performing aggregation over groups:

    x.groupby('gender').aggregate('mean(age)')
    
  • Boolean comparison and filtering of arrays:

    >>> x = sdb.random((5,5))
    >>> (x > 0.7).toarray()
    array([[ True,  True, False, False, False],
           [ True,  True,  True, False, False],
           [False, False, False,  True,  True],
           [False, False,  True, False, False],
           [False,  True,  True, False, False]], dtype=bool)
    >>> x[x>0.7].toarray()
    array([ 0.83500619,  0.95791602,  0.94745933,  0.89868099,  0.97664716,
            0.7045693 ,  0.88949448,  0.88112397,  0.73766701,  0.94612052])
    
  • Pandas-like syntax for accessing and defining new array attributes:

    array['b'] = 'sin(f0+3)'
    array['b'].toarray()
    
  • Lazy evaluation of arrays. Computation for most array operations are deferred until needed.

  • AFL queries now return lazy SciDBArrays instead of special class instances, which makes it easy to mix SciDBArray methods with raw AFL calls:

    >>> sdb.afl.build('<x:float>[i=0:5,10,0]', 'i').max()[0]
    
  • A cleaner syntax for chaining several AFL calls at once. The following two lines are equivalent:

    f = sdb.afl
    f.subarray(f.project(f.apply(x, 'f2', 'f*2'), 'f2'), 0, 5)
    
    x.apply('f2', 'f*2').project('f2').subarray(0, 5)
    
  • New element-wise operators: sqrt, floor, ceil, isnan

  • A cumulate() method for performing cumulative aggregation over arrays

  • Numerous bugfixes.