singularflow/plotting.py at main · martinjm97/singularflow · 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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
import jax
import jax.numpy as jnp
import matplotlib.pyplot as plt
import numpy as np
from matplotlib import rcParams

from training import load_model


def plot_losses(train_losses, test_losses, args, ground_truth_fun=None, extra_text: str = ""):
    for losses, title in [(train_losses, "Train"), (test_losses, "Test")]:
        if args.save_plot:
            rcParams["text.usetex"] = True
            rcParams["font.family"] = "libertine"
            rcParams["pdf.fonttype"] = 42
            rcParams["ps.fonttype"] = 42
        if args.load_files == "All" or len(losses) == 2:
            plt.figure(figsize=(7, 5))
            plt.plot(losses[0][: args.num_epochs_to_plot], label="Ours", color="green", linewidth=4)
            plt.plot(losses[1][: args.num_epochs_to_plot], label="Standard", color="blue", linestyle="--", linewidth=4)
            if ground_truth_fun is not None:
                len_losses = len(losses[0][: args.num_epochs_to_plot])
                plt.plot(
                    [0, len_losses],
                    [ground_truth_fun(args), ground_truth_fun(args)],
                    label=args.truth_title,
                    color="purple",
                    linestyle="dotted",
                    linewidth=4,
                )

            if args.plot_log_loss:
                plt.yscale("log")

            for spine in plt.gca().spines.values():
                spine.set_visible(False)
            plt.xticks(fontsize=20)
            plt.yticks(fontsize=20)
            plt.xlabel("Iteration", fontsize=32)
            plt.ylabel(f"{'Log ' if args.plot_log_loss else ''}{title} Loss", fontsize=32)
            plt.legend(frameon=False, fontsize=26)
            plt.tight_layout()
        elif args.load_files == "AllSeeds":
            if ground_truth_fun is not None:
                ground_truth_losses = []
                for seed in range(args.num_seeds):
                    args.seed = seed if title == "Train" else seed + args.num_seeds
                    ground_truth_losses.append(ground_truth_fun(args, train=(title == "Train")))
            if args.plot_log_loss:
                losses = np.log(losses)
                if ground_truth_fun is not None:
                    ground_truth_losses = np.log(ground_truth_losses)
            plt.figure(figsize=(7, 5))

            if ground_truth_fun is not None:
                len_losses = len(losses[0::2][0])
                ground_truth_mean_loss = np.mean(ground_truth_losses, axis=0)
                ground_truth_std_loss = np.std(ground_truth_losses, axis=0)
                plt.fill_between(
                    [0, len_losses],
                    2 * [ground_truth_mean_loss - ground_truth_std_loss],
                    2 * [ground_truth_mean_loss + ground_truth_std_loss],
                    color="purple",
                    alpha=0.2,
                )
                (ground_truth_handle,) = plt.plot(
                    [0, len_losses],
                    [ground_truth_mean_loss, ground_truth_mean_loss],
                    label=args.truth_title,
                    color="purple",
                    linestyle="dotted",
                    linewidth=4,
                )
            standard = losses[1::2]
            ours = losses[0::2]
            standard_mean = np.mean(standard, axis=0)
            ours_mean = np.mean(ours, axis=0)
            standard_std = np.std(standard, axis=0)
            ours_std = np.std(ours, axis=0)
            xs = np.arange(len(standard_mean))
            plt.fill_between(
                xs,
                standard_mean - standard_std,
                standard_mean + standard_std,
                color="blue",
                alpha=0.3,
            )
            plt.fill_between(xs, ours_mean - ours_std, ours_mean + ours_std, color="green", alpha=0.3)
            (ours_handle,) = plt.plot(ours_mean, label="Ours", color="green", linewidth=4)
            (standard_handle,) = plt.plot(standard_mean, label="Standard", color="blue", linestyle="--", linewidth=4)

            for spine in plt.gca().spines.values():
                spine.set_visible(False)
            plt.xticks(fontsize=20)
            plt.yticks(fontsize=20)
            plt.xlabel("Iteration", fontsize=32)
            plt.ylabel(f"{'Log ' if args.plot_log_loss else ''}{title} Loss", fontsize=32)
            legend_labels = ["Ours", "Standard"]
            legend_handles = [ours_handle, standard_handle]
            if ground_truth_fun is not None:
                legend_labels.append(args.truth_title)
                legend_handles.append(ground_truth_handle)
            plt.legend(frameon=False, fontsize=26, labels=legend_labels, handles=legend_handles, loc="upper right")
            plt.tight_layout()
        else:
            plt.plot(losses)
            plt.title("Loss over iterations")
        if args.save_plot:
            plt.savefig(
                f"{args.plot_path}/{args.plot_title.lower().replace(' ', '_')}{extra_text}_{title.lower()}_loss_plot.pdf"
            )
            plt.clf()
        else:
            plt.show()


def plot_functions(bounds, params, args, ground_truth=None, extra_text: str = ""):
    # if args.save_plot:
    rcParams["text.usetex"] = True
    rcParams["font.family"] = "libertine"
    rcParams["pdf.fonttype"] = 42
    rcParams["ps.fonttype"] = 42
    if args.load_files == "All":
        a, b = bounds
        singularp, standardp = params
        xs = np.linspace(a, b, 100)[1:]
        singular_model, standard_model = load_model(singularp, args), load_model(standardp, args)
        singular_nn_at_xs = jax.vmap(lambda x: singular_model(jnp.array([x])))(xs)
        standard_nn_at_xs = jax.vmap(lambda x: standard_model(jnp.array([x])))(xs)

        plt.figure(figsize=(7, 5))

        plt.plot(xs, singular_nn_at_xs, label="Ours", color="green", linewidth=4)
        plt.plot(xs, standard_nn_at_xs, label="Standard", color="blue", linestyle="--", linewidth=4)

        if ground_truth is not None:
            fun_at_xs = [ground_truth(x) for x in xs]
            if fun_at_xs[0] == float("NaN"):
                fun_at_xs[0] = None
            plt.plot(xs, fun_at_xs, label=args.truth_title, color="purple", linestyle="dotted", linewidth=4)

        for spine in plt.gca().spines.values():
            spine.set_visible(False)
        plt.xticks(fontsize=20)
        plt.yticks(fontsize=20)
        plt.xlabel(args.plot_function_xlabel, fontsize=32)
        plt.ylabel(args.plot_function_ylabel, fontsize=32)
        plt.legend(frameon=False, fontsize=26)
        plt.tight_layout()
        if args.save_plot:
            plt.savefig(f"{args.plot_path}/{args.plot_title.lower().replace(' ', '_')}{extra_text}_function.pdf")
            plt.clf()
        else:
            plt.show()
    elif args.load_files == "AllSeeds":
        a, b = bounds
        singularps, standardps = params[0::2], params[1::2]
        xs = np.linspace(a, b, 100)[1:]
        singular_nn_at_xss, standard_nn_at_xss = [], []
        for singularp, standardp in zip(singularps, standardps):
            singular_model, standard_model = load_model(singularp, args), load_model(standardp, args)
            singular_nn_at_xss.append(jax.vmap(lambda x: singular_model(jnp.array([x])))(xs))
            standard_nn_at_xss.append(jax.vmap(lambda x: standard_model(jnp.array([x])))(xs))
        singular_nn_at_xs = np.mean(singular_nn_at_xss, axis=0).flatten()
        standard_nn_at_xs = np.mean(standard_nn_at_xss, axis=0).flatten()
        singular_std = np.std(singular_nn_at_xss, axis=0).flatten()
        standard_std = np.std(standard_nn_at_xss, axis=0).flatten()

        plt.figure(figsize=(7, 5))

        plt.fill_between(
            xs,
            singular_nn_at_xs - singular_std,
            singular_nn_at_xs + singular_std,
            color="green",
            alpha=0.3,
        )
        plt.plot(xs, singular_nn_at_xs, label="Ours", color="green", linewidth=4)

        plt.fill_between(
            xs,
            standard_nn_at_xs - standard_std,
            standard_nn_at_xs + standard_std,
            color="blue",
            alpha=0.3,
        )
        plt.plot(xs, standard_nn_at_xs, label="Standard", color="blue", linestyle="--", linewidth=4)

        if ground_truth is not None:
            fun_at_xs = [ground_truth(x, None) for x in xs]
            if fun_at_xs[0] == float("NaN"):
                fun_at_xs[0] = None
            plt.plot(xs, fun_at_xs, label=args.truth_title, color="purple", linestyle="dotted", linewidth=4)

        for spine in plt.gca().spines.values():
            spine.set_visible(False)
        plt.xticks(fontsize=20)
        plt.yticks(fontsize=20)
        plt.xlabel(args.plot_function_xlabel, fontsize=32)
        plt.ylabel(args.plot_function_ylabel, fontsize=32)
        plt.legend(frameon=False, fontsize=26)
        plt.tight_layout()
        if args.save_plot:
            plt.savefig(f"{args.plot_path}/{args.plot_title.lower().replace(' ', '_')}{extra_text}_function.pdf")
            plt.clf()
        else:
            plt.show()