statistical_test.py

import os
import numpy as np
import pandas as pd

import matplotlib
import matplotlib.pyplot as plt

import math


def graph_ranks(avranks, names, cd=None, cdmethod=None, lowv=None, highv=None,
                width=6, textspace=1, reverse=False, filename=None, **kwargs):
    try:
        import matplotlib.pyplot as plt
        from matplotlib.backends.backend_agg import FigureCanvasAgg
    except ImportError:
        raise ImportError("Function graph_ranks requires matplotlib.")

    width = float(width)
    textspace = float(textspace)

    def nth(l, n):
        n = lloc(l, n)
        return [a[n] for a in l]

    def lloc(l, n):
        if n < 0:
            return len(l[0]) + n
        else:
            return n

    def mxrange(lr):
        if not len(lr):
            yield ()
        else:
            # it can work with single numbers
            index = lr[0]
            if isinstance(index, int):
                index = [index]
            for a in range(*index):
                for b in mxrange(lr[1:]):
                    yield tuple([a] + list(b))

    def print_figure(fig, *args, **kwargs):
        canvas = FigureCanvasAgg(fig)
        canvas.print_figure(*args, **kwargs)

    sums = avranks

    tempsort = sorted([(a, i) for i, a in enumerate(sums)], reverse=reverse)
    ssums = nth(tempsort, 0)
    sortidx = nth(tempsort, 1)
    nnames = [names[x] for x in sortidx]

    if lowv is None:
        lowv = min(1, int(math.floor(min(ssums))))
    if highv is None:
        highv = max(len(avranks), int(math.ceil(max(ssums))))

    cline = 0.4

    k = len(sums)

    lines = None

    linesblank = 0
    scalewidth = width - 2 * textspace

    def rankpos(rank):
        if not reverse:
            a = rank - lowv
        else:
            a = highv - rank
        return textspace + scalewidth / (highv - lowv) * a

    distanceh = 0.25

    if cd and cdmethod is None:
        # get pairs of non significant methods

        def get_lines(sums, hsd):
            # get all pairs
            lsums = len(sums)
            allpairs = [(i, j) for i, j in mxrange([[lsums], [lsums]]) if j > i]
            # remove not significant
            notSig = [(i, j) for i, j in allpairs
                      if abs(sums[i] - sums[j]) <= hsd]
            # keep only longest

            def no_longer(ij_tuple, notSig):
                i, j = ij_tuple
                for i1, j1 in notSig:
                    if (i1 <= i and j1 > j) or (i1 < i and j1 >= j):
                        return False
                return True

            longest = [(i, j) for i, j in notSig if no_longer((i, j), notSig)]

            return longest

        lines = get_lines(ssums, cd)
        linesblank = 0.2 + 0.2 + (len(lines) - 1) * 0.1

        # add scale
        distanceh = 0.25
        cline += distanceh

    # calculate height needed height of an image
    minnotsignificant = max(2 * 0.2, linesblank)
    height = cline + ((k + 1) / 2) * 0.2 + minnotsignificant

    fig = plt.figure(figsize=(width, height))
    fig.set_facecolor('white')
    ax = fig.add_axes([0, 0, 1, 1])  # reverse y axis
    ax.set_axis_off()

    hf = 1. / height  # height factor
    wf = 1. / width

    def hfl(l):
        return [a * hf for a in l]

    def wfl(l):
        return [a * wf for a in l]


    # Upper left corner is (0,0).
    ax.plot([0, 1], [0, 1], c="w")
    ax.set_xlim(0, 1)
    ax.set_ylim(1, 0)

    def line(l, color='k', **kwargs):
        """
        Input is a list of pairs of points.
        """
        ax.plot(wfl(nth(l, 0)), hfl(nth(l, 1)), color=color, **kwargs)

    def text(x, y, s, *args, **kwargs):
        ax.text(wf * x, hf * y, s, size=18, *args, **kwargs)

    line([(textspace, cline), (width - textspace, cline)], linewidth=1.7)

    bigtick = 0.1
    smalltick = 0.05

    tick = None
    for a in list(np.arange(lowv, highv, 0.5)) + [highv]:
        tick = smalltick
        if a == int(a):
            tick = bigtick
        line([(rankpos(a), cline - tick / 2),
              (rankpos(a), cline)],
             linewidth=1.7)

    for a in range(lowv, highv + 1):
        text(rankpos(a), cline - tick / 2 - 0.05, str(a),
                ha="center", va="bottom")

    k = len(ssums)

    a = 0.6
    b = a + 0.2

    for i in range(math.ceil(k / 2)):
        chei = cline + minnotsignificant + i * 0.24
        line([(rankpos(ssums[i]), cline),
              (rankpos(ssums[i]), chei),
              (textspace + b, chei)],
             linewidth=1.7)
        if rankpos(ssums[i]) < 1.1:
            text(textspace +4.2*a, chei, nnames[i], ha="right", va="center")
        else:
            text(textspace + a, chei, nnames[i], ha="right", va="center")

    for i in range(math.ceil(k / 2), k):
        chei = cline + minnotsignificant + (k - i - 1) * 0.24
        line([(rankpos(ssums[i]), cline),
              (rankpos(ssums[i]), chei),
              (textspace + scalewidth - b, chei)],
             linewidth=1.7)
        text(textspace + scalewidth - a, chei, nnames[i],
             ha="left", va="center")

    if cd and cdmethod is None:
        # upper scale
        if not reverse:
            begin, end = rankpos(lowv), rankpos(lowv + cd)
        else:
            begin, end = rankpos(highv), rankpos(highv - cd)

        # no-significance lines
        def draw_lines(lines, side=0.05, height=0.1):
            start = cline + 0.2
            for l, r in lines:
                line([(rankpos(ssums[l]) - side, start),
                      (rankpos(ssums[r]) + side, start)],
                     linewidth=2.5)
                start += height

        draw_lines(lines)

    elif cd:
        begin = rankpos(avranks[cdmethod] - cd)
        end = rankpos(avranks[cdmethod] + cd)
        line([(begin, cline), (end, cline)],
             linewidth=2.5)
        line([(begin, cline + bigtick / 2),
              (begin, cline - bigtick / 2)],
             linewidth=2.5)
        line([(end, cline + bigtick / 2),
              (end, cline - bigtick / 2)],
             linewidth=2.5)

    if filename:
        print_figure(fig, filename, **kwargs)

    return fig