Design of a n-dimensional array-like object with dimension variables

test_gridded_array.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
*****************************************************************************
Design document for a ``gridded_array`` class
*****************************************************************************

:Author:    Andreas Hilboll <[email protected]>
:Date:      Wed Sep 26 14:48:57 2012

This document wants to specify my need for an array-like data object which
stores dimension variables, i.e. information on what each dimension of the
array actually means. I call this class ``gridded_array``.
Ideally, the class would support ``datetime`` objects as dimension variable,
and it would be possible to easily down-sample (by averaging) and interpolate
to finer grids.

Most of the requirements would be fulfilled by the excellent ``pandas``
library. However, ``pandas`` only supports up to three dimensions. Another
problem is that the nomenclature (``major_axis``, ``minor_axis``, ``items``)
is kind of unintuitive because the user always has to remember which dimension
is stored in which variable. If a ``pandas.core.index.MultiIndex`` would
better support gridded dimensions, and if this functionality would be
integrated into the core data objects in ``pandas``, this would probably be a
good solution.


About dimension variables
-------------------------
Dimension variables would need to be 1d arrays. I personally would need
floating point and datetime dtypes. Ideally, a dimension variable could be
cyclic. This could be realized with a ``dimvar._cyclic`` attribute defaulting
to ``None``, but possible having a value which 'makes sense', i.e. a tuple
``(begin, end)``.

"""

# Library imports
# ===========================================================================

from collections import OrderedDict
import unittest
from gridded_array import gridded_array

import numpy as np



# Test class definition
# ===========================================================================

class gridded_array_test(unittest.TestCase):
    """Specification of the ``gridded_array`` class."""

    def setup(self):
        self.simple_dims = OrderedDict(longitude=np.linspace(-150., 150., 6),
                                       latitude=np.linspace(-67.5, 67.5, 4))
        self.simple_garr = gridded_array(np.ones((6,4)), self.simple_dims)

    # Basic functionality
    # ------------------------------------------------------------------------
    def test_initialization(self):
        """Initialization of a ``gridded_array`` always requires the data.
        If no dimension variables are given, they are created automatically as
        ``np.arange(len(dimension))``.

        """
        # TODO: gridded_array(data)
        # TODO: gridded_array(data, dims)
        # TODO: gridded_array(existing_arr)
        # TODO: gridded_array(existing_arr, dims)
        # TODO: gridded_array(existing_marr)
        # TODO: gridded_array(existing_marr, dims)
        # TODO: dims as list(variables), list(names)
        # TODO: only one dim, as "arr", "name"
        pass

    def test_add(self):
        """Addition should work as expected by returning the sum of two
        ``gridded_array`` as a new ``gridded_array``. Should raise an
        exception if the dimensions of the two inputs do not agree."""
        # TODO: garr + garr
        # TODO: garr + masked_array
        # TODO: garr + array
        # TODO: garr + scalar
        # TODO: masked_array + garr
        # TODO: array + garr
        # TODO: scalar + garr
        pass

    def test_slicing(self):
        """Slicing should work as expected. Indexing should be possible both
        via the indices and via the according dimensions."""
        # TODO: garr[0]
        # TODO: garr[:,:,0]
        # TODO: garr[:2]
        # TODO: garr[:,5:8]
        pass

    def test_equal(self):
        """Equality should occur only when comparing a ``gridded_array`` with
        another ``gridded_array``. Maybe say in the Exception if the
        underlying data are identical?

        Actually, I don't see a use-case for testing for equality (except for
        writing tests, that is)"""
        # TODO: garr == garr (True, False)
        # TODO: garr == arr
        # TODO: garr == marr
        pass

    def test_broadcast(self):
        """``gridded_array`` should support intelligent broadcasting based on
        the dimension variables."""
        # TODO: (lat, lon) + (lon, lat)
        # TODO: (lat, lon, alt) + (lon, lat)
        # TODO: (alt, lat, lon) + (lat)
        pass

    def test_transpose(self):
        """It should be possible to transpose the array, i.e. swap the axes.
        While this might seem unnecessary, it's important e.g. when writing
        to file, and the writing method doesn't give the option to rearrange
        the dimensions."""
        # TODO: (lat, lon).transpose(('lon', 'lat'))
        # TODO: (alt, lat, lon).transpose(('lon', 'alt', 'lat')
        pass

    def test_to_array(self):
        """``gridded_array.to_array()`` returns a ``ndarray`` object, with
        optionally transposed axes, and optionally filled missing values"""
        # TODO: garr.to_array()
        # TODO: garr.to_array(fill=999.)
        # TODO: (lat, lon).to_array(dims=('lon', 'lat'))
        pass

    def test_to_masked_array(self):
        """Same as to_array, but returns masked_array"""
        pass

    def test_cut(self):
        """Cut out a sub-hypercube from the array object. """
        # TODO: garr(lat,lon,alt).cut(('lat', 10., 20)) -> return garr object
        #       which only contains those points with 'lat' dimension between
        #       10 and 20. Should work for multiple dimensions simultaneously.
        # TODO: garr.cut(0, -8, 8)
        # TODO: garr.cut('x', -8, 8)
        # TODO: garr.cut((0, -8, 8), ('y', -10, 0))
        pass

    def test_advanced_indexing(self):
        """Indexing via arbitrary expressions of type bool"""
        pass


    # Optional functionality
    # ------------------------------------------------------------------------

    def test_timeaxis(self):
        """Ideally, ``gridded_array`` would behave like a
        ``pandas.core.series.Series`` object when it comes to temporal
        manipulations."""
        pass

    def test_lt_gt(self):
        """No need to implement less-than and greater-than, IMO"""
        pass

    def test_axis_arg(self):
        """It would be great if all those class methods / numpy functions,
        which take an ``axis`` parameter would work with the dimension's
        name as argument instead of the numerical order of the dimension"""
        pass


    # Convenience functions to call something from an external lib
    # ------------------------------------------------------------------------

    def test_downsample(self):
        """Support generating a new ``gridded_array`` by down-sampling"""
        # TODO: downsample in one dim
        # TODO: dowmsample in two dims
        # TODO: downsample in three dims
        pass

    def test_interpolate(self):
        """Interpolation to a finer grid. Should be a conveniance function
        to call something appropriate from ``scipy.interpolate``."""
        pass