EmbeddingComponents/method1_vs_method2_scatterplot.py at main · rrevoid/EmbeddingComponents · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import scipy.stats as sps

from argparse import ArgumentParser


def get_scatterplot(
        fi1: np.ndarray, # fi stands for feature imporatnces
        fi2: np.ndarray,
        name1: str="",
        name2: str="",
        task: str="",
        logy: bool=False,
        plot95: bool=False,
        save_path: str="./output.png",
) -> None:
    outlier_dimentions = [61, 77, 82, 97, 217, 219, 240, 330, 361, 453, 494, 496, 498, 551, 570, 588, 656, 731, 749]
    c = np.zeros((768, 3))
    c[:, 2] = 1
    data = pd.DataFrame(np.stack((fi1, fi2), axis=1), columns=[name1, name2])
    data['color'] = data.index.isin(outlier_dimentions).astype(int)

    ax = sns.scatterplot(x=name1, y=name2, data=data, hue='color', label=rf'$r_s$={sps.spearmanr(fi1, fi2).statistic:.2f}', alpha=0.2, legend=False)
    for dim in outlier_dimentions:
        #ax.text(fi1[dim], fi2[dim], dim, horizontalalignment='left', verticalalignment='top', size='small', color='black')
        ax.text(fi1[dim], fi2[dim], dim, size='small', color='black')

    if plot95:
        plt.axhline(np.percentile(fi2, 95), linestyle='--', color='red')

    if logy:
        plt.yscale("log")

    plt.xlabel(name1)
    plt.ylabel(name2)
    plt.title(task)
    plt.legend()

    plt.savefig(save_path)


if __name__ == "__main__":
    parser = ArgumentParser(
        description="""
        Draw scatterplot: feature importances of method1 vs feature importances of method2.
        Note: under given fi_path should lie a np.ndarray (i -> feature imortance of i-th feature).
        """
    )
    parser.add_argument("--fi1_path", type=str, help="path to feature importances of method1")
    parser.add_argument("--fi2_path", type=str, help="path to feature importances of method2")
    parser.add_argument("--fi1_name", type=str, default="method1", help="name of method1")
    parser.add_argument("--fi2_name", type=str, default="method2", help="name of method2")
    parser.add_argument("--output_path", type=str, default="./output.png")
    parser.add_argument("--task", type=str, default="", help="probing task")
    parser.add_argument("--logy", action="store_true", help="make yscale logarithmic")
    parser.add_argument("--plot95", action="store_true", help="plot 95-th percentile")
    args = parser.parse_args()

    with open(args.fi1_path, "rb") as file:
        fi1 = np.load(file)

    with open(args.fi2_path, "rb") as file:
        fi2 = np.load(file)

    get_scatterplot(
        fi1, fi2,
        name1=args.fi1_name,
        name2=args.fi2_name,
        task=args.task,
        logy=args.logy,
        plot95=args.plot95,
        save_path=args.output_path,
    )