Value By Alpha specification

[ljwolf]

@darribas and I talked, and we think it makes the most sense for value-by-alpha maps to be implemented as follows:

def value_by_alpha('attribute', 'alpha_attribute', ..., data=geodataframe):
    rgb = make_RGBA_array_from_attribute(data.attribute)
    alpha_channel = data['alpha_attribute'] / data['alpha_attribute'].max()
    rgba = np.vstack((rgb, np.asarray(alpha_channel).reshape(-1,1)))
    return geodataframe.plot('attribute', color=rgba)

[ljwolf]

also, we would need to check if the cmap is divergent. If it is, we may offer a diverging_alpha=True flag that lets the inside of the colormap become transparent, rather than the extreme in the specification shown above.

Essentially, we need to offer all options for the rgb matrix for the alpha channel as well.

[darribas]

Here's an example on how to pass RGBA tuples to geopandas for fully customisation (provided by @jorisvandenbossche on Gitter):

In [35]: import geopandas

In [36]: from shapely.geometry import Polygon

In [37]: import random

In [26]: import matplotlib

In [30]: cols = matplotlib.colors.to_rgba_array(['r', 'g', 'k', 'b'])

In [31]: cols
Out[31]: 
array([[ 1. ,  0. ,  0. ,  1. ],
       [ 0. ,  0.5,  0. ,  1. ],
       [ 0. ,  0. ,  0. ,  1. ],
       [ 0. ,  0. ,  1. ,  1. ]])

In [32]: cols[:, -1] = [0.5, 0.7, 0.9, 0.2]

In [33]: s = geopandas.GeoSeries([Polygon([(random.random(), random.random()) for i in range(3)]) for _ in range(4)])

In [34]: s.plot(color=cols)
Out[34]: <matplotlib.axes._subplots.AxesSubplot at 0x7fc0d783cd68>

[slumnitz]

pr #28 is a start dealing with value_by_alpha functionality. Matplotlib actually offers a great way to work with rgba cmaps:

def value_by_alpha_cmap(x, y, cmap='GnBu'):
    """
    Calculates Value by Alpha rgba values
    
    Parameters
    ----------
    x : array
        Variable determined by color
    y : array
        Variable determining alpha value
    cmap : str or list of str
        Matplotlib Colormap or list of colors used
        to create vba_layer
    
    Returns
    -------
    rgba : ndarray (n,4)
        RGBA colormap, where the alpha channel represents one
        attribute (x) and the rgb color the other attribute (y)
    """
    # option for cmap or colorlist input
    if isinstance(cmap, str):
        cmap = cm.get_cmap(cmap)
    elif isinstance(cmap, collections.Sequence):
        cmap = colors.LinearSegmentedColormap.from_list('newmap', cmap)
    rgba = cmap(x)
    rgba[:, 3] = y/y.max()
    return rgba

[slumnitz]

@ljwolf @darribas @sjsrey do you think it makes sense to allow mapclassifyed data to also be included into value by alpha data? E.g. x and y are classified by 'Quantiles` and then displayed with y in the alpha layer? I am unsure at the moment how to best connect all different options for choropleth visualisations together and was wondering if you had ideas? Maybe a topic to discuss on Friday.

as to say do we try and squeeze everything in one function eg.
choropleth(gdf, x, vba=y, divergent_vba=True, mapclassify='Quantiles', ..., ax=None)

or do we create a class and build something like:

choro = choropleth(gdf, x)
choro.Quantiles(x, y)
choro.vba(y, divergent)

[slumnitz]

closing this issue, vba functionality will be added with #28 and #30