This module provides:
- A set of new numpy
ndarrayducktypes, includingMaskedArray,ArrayCollection, andUnitArrayusing the new ducktype numpy api. See below for descriptions. - Helper tools to help you define your own numpy ducktypes. This including helper methods, mixins, and most importantly printing functionality in the
duckprintsubmodule to help implement your ducktype'sstrandrepr.
This package supports standard python package installation using setuptools. The simplest way is to use pip to install from the cloned source directory:
$ pip install /path/to/ndarray_ducktypes
and uninstall by "pip uninstall ndarray_ducktypes".
To use these ducktypes, numpy's __array_function__ must be enabled, and it is currently disabled by default. To enable, in your shell do:
export NUMPY_EXPERIMENTAL_ARRAY_FUNCTION=1
- MakedArray documentation.
- ArrayCollection documentation
- UnitArray documentation (TODO)
- LazyArray documentation (TODO)
- Making Ducktypes documentaiton
MaskedArray provides missing-value support for numpy arrays, and is meant as a new iteration of numpy's np.ma.MaskedArray. MaskedArrays from this module can be used as drop-in replacements in arbitrary numpy expressions. (95% complete.)
See the MakedArray documentation.
ArrayCollection mimics numpy's structured arrays but avoids binary representation issues of structured arrays and can have more optimized memory layout. ArrayCollection is meant for use as a "pandas-lite" for working with tabular data, which avoids the difficulties of using structured arrays for this purpose. (5% complete)
See the ArrayCollection documentation
This module also defines a composite MaskedArrayCollection type which allows you to mask individual fields in elements of an ArrayCollection.
Will implement a ducktype which stores unit information in addition to the dtype. 0% complete.
Will implement a ducktype which delays evaluation of numpy operations until .eval() is called. Doing np.sin(a + b) will return an unevaluated LazyArray on which you can call .eval(). The default .eval will avoid creation of intermediate arrays using numpy's out= method arguments for a small speedup, but it should be possible to override .eval for bigger gains, eg numba compilation. 0% complete.
This module implements helper classes and methods and mixins to help simplify defining new ducktypes, with some recommendations for implementation which are reuired by some of the helper methods, see docs linked below.
To make ducktypes quicker to define, the duckprint provides ready-made printing routines to help implement the str/repr of ducktypes. duckprint is used internally by the MaskedArray and ArrayCollection classes defined in this repository.