informer-/vis.py at main · WJLGraph/informer- · 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
# utils/visualization.py - 中文命名+自动路径
import os
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.font_manager as fm
from datetime import datetime

# 中文乱码修复
zh_font = fm.FontProperties(fname=r'C:\Windows\Fonts\simhei.ttf')
plt.rcParams['font.family'] = 'sans-serif'
plt.rcParams['font.sans-serif'] = [zh_font.get_name(), 'DejaVu Sans']
plt.rcParams['axes.unicode_minus'] = False

# 保存根目录
SAVE_ROOT = r'results/png'
os.makedirs(SAVE_ROOT, exist_ok=True)

class InformerVisualizer:
    def __init__(self, figsize=(12, 8)):
        self.figsize = figsize

    # ---------- 公共保存方法 ----------
    def _save(self, fig, name):
        path = os.path.join(SAVE_ROOT, f'{name}.png')
        fig.savefig(path, dpi=300, bbox_inches='tight')
        print(f'✅ 已保存：{path}')
        plt.close(fig)

    # ---------- 时间序列 ----------
    def plot_time_series(self, data, timestamps=None, labels=None,
                         title="时间序列数据", xlabel="时间", ylabel="数值"):
        data = [data] if not isinstance(data, list) else data
        labels = labels or [f'序列{i+1}' for i in range(len(data))]

        fig, ax = plt.subplots(figsize=self.figsize)
        for series, label in zip(data, labels):
            x = timestamps if timestamps is not None else range(len(series))
            ax.plot(x, series, label=label, linewidth=2, alpha=0.8)

        ax.set_title(title, fontproperties=zh_font, fontsize=16)
        ax.set_xlabel(xlabel, fontproperties=zh_font)
        ax.set_ylabel(ylabel, fontproperties=zh_font)
        ax.legend(prop=zh_font)
        ax.grid(True, alpha=0.3)
        if timestamps is not None:
            plt.xticks(rotation=45)
        self._save(fig, '时间序列对比图')

    # ---------- 详细对比 ----------
    def plot_prediction_comparison(self, true_values, pred_values, timestamps=None,
                                 feature_names=None, n_samples=3):
        n_samples = min(n_samples, len(true_values))
        n_features = true_values.shape[-1]
        pred_len = true_values.shape[1]
        sample_indices = np.linspace(0, len(true_values) - 1, n_samples, dtype=int)

        fig, axes = plt.subplots(n_features, n_samples,
                                figsize=(5 * n_samples, 4 * n_features), squeeze=False)

        for feat_idx in range(n_features):
            for sample_idx, data_idx in enumerate(sample_indices):
                ax = axes[feat_idx, sample_idx]
                true_line = true_values[data_idx, :, feat_idx]
                pred_line = pred_values[data_idx, :, feat_idx]
                x_axis = timestamps[data_idx] if timestamps is not None else range(pred_len)

                ax.plot(x_axis, true_line, 'b-', label='真实值', linewidth=2)
                ax.plot(x_axis, pred_line, 'r--', label='预测值', linewidth=2)

                mae = np.mean(np.abs(true_line - pred_line))
                mse = np.mean((true_line - pred_line) ** 2)
                ax.set_title(f'特征{feat_idx+1} - 样本{data_idx}\nMAE:{mae:.4f}  MSE:{mse:.4f}',
                           fontproperties=zh_font, fontsize=10)
                ax.set_xlabel('时间步', fontproperties=zh_font)
                ax.set_ylabel('数值', fontproperties=zh_font)
                ax.legend(prop=zh_font)
                ax.grid(True, alpha=0.3)
                if timestamps is not None:
                    ax.tick_params(axis='x', rotation=45)

        plt.suptitle('预测结果详细对比', fontproperties=zh_font, fontsize=16)
        plt.tight_layout()
        self._save(fig, '预测结果详细对比图')

    # ---------- 误差分析 ----------
    def plot_error_analysis(self, true_values, pred_values):
        errors = pred_values - true_values
        # 1. 整体统计（1维）
        flat_errors = errors.reshape(-1)
        mean_err, std_err = np.mean(flat_errors), np.std(flat_errors)

        fig, axes = plt.subplots(2, 2, figsize=(15, 10))

        # 1. 直方图
        axes[0, 0].hist(flat_errors, bins=50, alpha=0.7, edgecolor='black')
        axes[0, 0].set_title('误差分布直方图', fontproperties=zh_font)
        axes[0, 0].set_xlabel('误差值', fontproperties=zh_font)
        axes[0, 0].set_ylabel('频次', fontproperties=zh_font)
        axes[0, 0].grid(True, alpha=0.3)

        # 2. 随时间平均误差（必须1维）
        mean_error_t = np.mean(errors, axis=0)   # [pred_len, n_features]
        std_error_t  = np.std(errors, axis=0)    # [pred_len, n_features]

        # 如果是多特征，则对特征维度求均值以得到单条曲线
        if len(mean_error_t.shape) > 1:
            mean_error_t = np.mean(mean_error_t, axis=-1)  # [pred_len]
            std_error_t = np.mean(std_error_t, axis=-1)    # [pred_len]

        time_steps   = np.arange(len(mean_error_t))

        axes[0, 1].plot(time_steps, mean_error_t, 'b-', label='平均误差', linewidth=2)
        axes[0, 1].fill_between(time_steps,
                                 mean_error_t - std_error_t,
                                 mean_error_t + std_error_t,
                                 alpha=0.3, label='标准差范围')
        axes[0, 1].set_title('误差随时间变化', fontproperties=zh_font)
        axes[0, 1].set_xlabel('时间步', fontproperties=zh_font)
        axes[0, 1].set_ylabel('误差', fontproperties=zh_font)
        axes[0, 1].legend(prop=zh_font)
        axes[0, 1].grid(True, alpha=0.3)

        # 3. 散点图
        axes[1, 0].scatter(true_values.flatten(), pred_values.flatten(), alpha=0.5, s=1)
        min_val, max_val = true_values.min(), true_values.max()
        axes[1, 0].plot([min_val, max_val], [min_val, max_val], 'r--', linewidth=2)
        axes[1, 0].set_title('预测值 vs 真实值', fontproperties=zh_font)
        axes[1, 0].set_xlabel('真实值', fontproperties=zh_font)
        axes[1, 0].set_ylabel('预测值', fontproperties=zh_font)
        axes[1, 0].grid(True, alpha=0.3)

        # 4. 热图
        if len(errors.shape) == 3 and errors.shape[0] > 1 and errors.shape[1] > 1:
            im = axes[1, 1].imshow(errors[:50, :], aspect='auto', cmap='RdBu', origin='lower')
            axes[1, 1].set_title('误差热图（前50个样本）', fontproperties=zh_font)
            axes[1, 1].set_xlabel('时间步', fontproperties=zh_font)
            axes[1, 1].set_ylabel('样本索引', fontproperties=zh_font)
            plt.colorbar(im, ax=axes[1, 1])
        else:
            axes[1, 1].text(0.5, 0.5, '数据维度不适合\n绘制热图', ha='center', va='center',
                           transform=axes[1, 1].transAxes, fontproperties=zh_font)
            axes[1, 1].set_title('误差热图', fontproperties=zh_font)

        plt.suptitle('预测误差分析', fontproperties=zh_font, fontsize=16)
        plt.tight_layout()
        self._save(fig, '预测误差分析图')


# ---------------- 快速测试 ----------------
def demo_visualization():
    np.random.seed(42)
    n_samples, n_features, pred_len = 10, 3, 24
    true_values = np.random.randn(n_samples, pred_len, n_features)
    pred_values = true_values + 0.1 * np.random.randn(n_samples, pred_len, n_features)

    vis = InformerVisualizer()
    vis.plot_time_series([true_values[0, :, 0], pred_values[0, :, 0]],
                         labels=['真实值', '预测值'], title='时间序列对比')
    vis.plot_prediction_comparison(true_values, pred_values)
    vis.plot_error_analysis(true_values, pred_values)
    print("✅ 可视化完成，文件已保存至 results/png/ 目录（中文命名）")


if __name__ == "__main__":
    demo_visualization()