PRC/fig/figC.py at main · yrrIsHere/PRC · 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
import pandas as pd
import torch
import torch.nn.functional as F
from data.dataset import SeqDataSet
import einops
from tqdm import trange
import numpy as np
from sklearn.linear_model import Ridge
from sklearn.model_selection import cross_val_score, KFold
from model.PRC import PRL_Net


device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = torch.load('model/models/myNet_1130_142138.pth').to(device)
real_weight = np.linspace(-0.9, 0.9, 10)


def regression(X,y):
    alpha_range = np.logspace(-3, 3, 50)  # 超参数范围
    best_alpha = None
    best_score = -np.inf

    for alpha in alpha_range:
        ridge = Ridge(alpha=alpha)
        kf = KFold(n_splits=10, shuffle=True, random_state=42)
        scores = cross_val_score(ridge, X, y, scoring='neg_mean_squared_error', cv=kf)
        mean_score = np.mean(scores)
        if mean_score > best_score:
            best_score = mean_score
            best_alpha = alpha

    ridge = Ridge(alpha=best_alpha)
    ridge.fit(X, y)

    beta_0 = ridge.intercept_  # 偏置项
    beta = ridge.coef_  # 各形状的权重

    print(f"最佳 alpha: {best_alpha}")
    print(f"偏置 β0: {beta_0}")
    print(f"权重 β1-10: {beta}")

    return beta


def get_choice(t_dataset,bs=1):
    model.eval()
    x,y,end,t_shapes,labels = t_dataset.x,t_dataset.y,t_dataset.shape_end,t_dataset.shapes,t_dataset.choices

    pred_result = []
    label_result = []
    shapes = []

    # 进行推理
    # 获得形状序列对应的决策结果
    for idx in trange(t_dataset.get_len()):
        batch_x, batch_y, shape,label = x[idx],y[idx],t_shapes[idx],labels[idx]
        batch_x,batch_y = torch.from_numpy(batch_x).unsqueeze(dim=0),torch.from_numpy(batch_y).unsqueeze(dim=0)
        batch_x = einops.rearrange(batch_x, 'b c l -> b l c')
        batch_y = einops.rearrange(batch_y, 'b c l -> b l c')
        feats = model.rc_step(batch_x)
        feats = [torch.from_numpy(feat).float().to(device) for feat in feats]
        batch_x, batch_y = batch_x.to(device), batch_y.to(device)

        pred_choice = -1
        label_choice = -1
        #
        for t in range(batch_x.shape[1]):
            if pred_choice >0 and label_choice>0:
                break

            if pred_choice==-1:
                pred = model.linear_process(t, feats[t])
                pred = torch.argmax(F.sigmoid(pred),dim=1).cpu().numpy()[0]
                if pred==1 or pred==2:
                    pred_choice=pred

        if pred_choice !=-1:
            # pred_choice = label_choice
            pred_result.append(pred_choice)
            label_result.append(2-label)
            shapes.append(np.array(shape))

    # 统计每个形状的数量
    shape_num = []
    for i,shape in enumerate(shapes):
        temp = np.zeros(10)
        for j in range(shape.__len__()):
            temp[shape[j]] += 1
        shape_num.append(temp)
    shape_num = np.array(shape_num).astype(int)

    pred_result = np.array(pred_result)
    label_result = np.array(label_result)

    return label_result, shape_num


if __name__ == "__main__":

    test_dataset = SeqDataSet(num=2000,mode=0)
    data_y,data_x = get_choice(test_dataset,bs=1)

    data = np.hstack([data_x, data_y.reshape(-1, 1)])  # 合并成 11维度的 [x, y]

    # 建表，分组
    columns = [f"x{i}" for i in range(10)] + ["y"]
    df = pd.DataFrame(data, columns=columns)

    df['group_key'] = df.iloc[:, :-1].apply(tuple, axis=1)

    grouped = df.groupby('group_key')

    x = []
    y = []
    # P -> Q
    # 论文公式(2)
    for group, sub_df in grouped:
        sub_df = sub_df.to_numpy()
        p = (np.sum(sub_df[:,10]==1))/sub_df.shape[0]
        q = -np.log10(1 / (p+1e-8) - 1 +1e-8 )
        x.append(sub_df[0, :10])
        y.append(q)
    x = np.array(x)
    y = np.array(y)

    import numpy as np
    import matplotlib.pyplot as plt

    # 回归得到模型所学习到的主观权重
    infer_weights = regression(x,y)

    x = real_weight
    y = infer_weights.copy()

    plt.scatter(x,y,c='green')
    plt.xticks(ticks=x, labels=[round(float(item),1) for item in x])
    plt.xlabel("Assigned logLR",fontsize=13)
    plt.ylabel("Subjective logLR",fontsize=13)
    plt.show()
    plt.savefig('figC.png',dpi=600)