attention-rank-collapse/plot.py at main · anothersin/attention-rank-collapse · 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
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
from __future__ import unicode_literals

import os.path as osp
import pathlib
import pdb

import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import scipy

import arguments


matplotlib.use("agg")

"""
Visualization functions.
"""

res_dir = "results"
pathlib.Path(res_dir).mkdir(exist_ok=True, parents=True)


class PlotArg:
    def __init__(self, x, y, std=None):
        """
        x has shape d
        y has shape k x d
        """
        self.title = ""
        self.x_label = "x"
        self.y_label = "y"
        self.x = x
        self.y = y
        self.std = np.zeros(tuple(y.shape)) if std is None else std
        self.legend = [""] * len(y)


def compute_rep(args):
    """
    Compute path representation as function of k
    """
    rep_ar = np.zeros((args.depth,))
    alpha = args.alpha
    for i, k in enumerate(range(args.depth)):
        m_choose_k = scipy.special.comb(args.depth, k)
        rep_ar[i] = (
            m_choose_k
            * args.width ** k
            * args.alpha ** (args.depth - k)
            * (1 - alpha) ** k
            * args.R0
            / (args.c ** (3 ** k - 1))
        )

        print(k, rep_ar[i])


##
def compute_rep_diff_alpha(args):
    """
    Compute path representation as function of k, diff alpha
    """
    rep_ar = [[] for _ in range(len(args.alpha_l))]  # np.zeros((args.depth,))
    compute_rep_rec(args, rep_ar, 0, 0, 1)

    print([sum(l) for l in rep_ar])
    """
    alpha = args.alpha
    for i, k in enumerate(range(args.depth)):
        m_choose_k = scipy.special.comb(args.depth, k)
        rep_ar[i] = m_choose_k*args.width**k *args.alpha**(args.depth -k) * (1-alpha)**k * args.R0 / (args.c**(3**k-1))

        print(k, rep_ar[i])
    """


##
def compute_rep_rec(args, rep_ar, cur_path_len, cur_idx, cur_alpha_prod):
    """
    rep_ar is list of lists
    """
    if cur_idx == len(args.alpha_l):
        if cur_path_len == 0:
            return
        try:
            cur_rep = args.R0 / (args.c ** (3 ** cur_path_len - 1)) * cur_alpha_prod
        except Exception as e:
            print(e)
            pdb.set_trace()
        rep_ar[cur_path_len - 1].append(cur_rep)
        return
    # skip
    compute_rep_rec(args, rep_ar, cur_path_len, cur_idx + 1, cur_alpha_prod * args.alpha_l[cur_idx])
    # non skip
    compute_rep_rec(args, rep_ar, cur_path_len + 1, cur_idx + 1, cur_alpha_prod * (1 - args.alpha_l[cur_idx]))


def plot_scatter(plot_arg, fname, xlim=None, ylim=None, loc=None, bbox=None):
    # plt.plot(data_ar[0], data_ar[i], marker=markers[i-1], label=legend_l[i-1])
    plt.clf()
    markers = ["^", "o", "x", ".", "1", "3", "+", "4", "5"]
    plt.rcParams.update({"font.size": 14.5})

    for i, y in enumerate(plot_arg.y):
        if len(plot_arg.x.shape) > 1:
            x_ar = plot_arg.x[i]
        else:
            x_ar = plot_arg.x
        if plot_arg.std[i].sum() == 0:
            plt.plot(x_ar, y, label=plot_arg.legend[i])  # ,  linestyle="None")
        else:
            plt.errorbar(
                x_ar, y, yerr=plot_arg.std[i], marker=markers[i], label=plot_arg.legend[i]
            )  # ,  linestyle="None")

    plt.grid(True)
    if loc is not None:
        plt.legend(loc=loc, bbox_to_anchor=bbox)
    else:
        plt.legend()
    if xlim is not None:
        plt.xlim(xlim)  # ([-.5, .5])
    if ylim is not None:
        plt.ylim(ylim)  # [-.5, .5])
    plt.xlabel(plot_arg.x_label)
    plt.ylabel(plot_arg.y_label)
    plt.title(plot_arg.title)

    fig_path = osp.join(
        res_dir, "plot_{}.pdf".format(fname)
    )  #'baselines_{}{}{}.jpg'.format(opt.type, name, fname_append))
    plt.savefig(fig_path)
    print("figure saved under {}".format(fig_path))


def plot_plot(plot_arg, fname, xlim=None, ylim=None, loc=None, bbox=None):
    # plt.plot(data_ar[0], data_ar[i], marker=markers[i-1], label=legend_l[i-1])
    plt.clf()
    markers = ["^", "o", "x", ".", "1", "3", "+", "4", "5"]
    plt.rcParams.update({"font.size": 14.5})

    for i, y in enumerate(plot_arg.y):
        if len(plot_arg.x.shape) > 1:
            x_ar = plot_arg.x[i]
        else:
            x_ar = plot_arg.x
        plt.errorbar(
            x_ar, y, yerr=plot_arg.std[i], marker=markers[i], label=plot_arg.legend[i]
        )  # ,  linestyle="None")

    plt.grid(True)
    if loc is not None:
        plt.legend(loc=loc, bbox_to_anchor=bbox)
    else:
        plt.legend()
    if xlim is not None:
        plt.xlim(xlim)  # ([-.5, .5])
    if ylim is not None:
        plt.ylim(ylim)  # [-.5, .5])
    plt.xlabel(plot_arg.x_label)
    plt.ylabel(plot_arg.y_label)
    plt.title(plot_arg.title)

    fig_path = osp.join(
        res_dir, "plot_{}.pdf".format(fname)
    )  #'baselines_{}{}{}.jpg'.format(opt.type, name, fname_append))
    plt.savefig(fig_path)
    print("figure saved under {}".format(fig_path))


if __name__ == "__main__":
    args = arguments.parse_args()
    args.c = 1.0491046  # 0.000491046376 #1.3
    args.R0 = 400  # 48 #3
    args.alpha = 0.6
    args.alpha_l = [0.90140117, 0.90491132, 0.90742852, 0.88882977, 0.85639907, 0.77844384]
    # [0.88690354 0.90293881 0.66461398 0.79351534 0.6367262  0.82008575  0.81519226 0.81471389 0.87243967 0.57299129]
    print(args)
    compute_rep_diff_alpha(args)