plot rank

plot_rank.py

import numpy as np
from matplotlib import pyplot as plt

supernet = [0.524, 0.569, 0.588, 0.633, 0.438] 
retrain = [0.734, 0.562, 0.556, 0.581, 0.373]

def make_patch_spines_invisible(ax):
    ax.set_frame_on(True)
    ax.patch.set_visible(False)
    for sp in ax.spines.values():
        sp.set_visible(False)

def pplot(x, y, title):
    assert isinstance(x, list)
    assert isinstance(y, list)

    x = np.array(x)
    y = np.array(y)
    arg_x = x.argsort()[::-1] # reverse order
    arg_y = y.argsort()[::-1] 
    nums = len(x)

    # raw x,y 
    fig, ax = plt.subplots(sharey=True)
    ax1 = ax.twinx()
    ax2 = ax.twinx()
    ax2.spines["right"].set_position(("axes", 1.2))
    make_patch_spines_invisible(ax2)
    ax2.spines["right"].set_visible(True)
    ax2.set_ylabel("Retrain")

    ax.set_ylim(0.5,nums+0.5)
    ax.set_xlim([-0.05,1.05])
    ax.set_xticklabels([])
    


    assert len(x) == len(y)
    for i in range(len(x)):
        ax1.plot([0, 1], [arg_x[i]+1, arg_y[i]+1])
    ax1.set_xlim([-0.05,1.05])
    ax1.set_ylim(0.5,nums+0.5)
    ax1.set_title(title)
    # ax1.set_ylabel("SuperNet")


    plt.show()


if __name__ == '__main__':
    pplot(supernet, retrain, 'RANK')
    # Top-1  means best