FCDO-Humanitarian-Forecasting-Presentation/utils.py at main · devinit/FCDO-Humanitarian-Forecasting-Presentation · 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
import torch
from sklearn.metrics import roc_auc_score, mean_squared_error
import numpy as np
from augmentations import embed_data_mask
import torch.nn as nn

def make_default_mask(x):
    mask = np.ones_like(x)
    mask[:,-1] = 0
    return mask

def tag_gen(tag,y):
    return np.repeat(tag,len(y['data']))


def count_parameters(model):
    return sum(p.numel() for p in model.parameters() if p.requires_grad)

def get_scheduler(args, optimizer):
    if args.scheduler == 'cosine':
        scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, args.epochs)
    elif args.scheduler == 'linear':
        scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,
                                      milestones=[args.epochs // 2.667, args.epochs // 1.6, args.epochs // 1.142], gamma=0.1)
    return scheduler

def imputations_acc_justy(model,dloader,device):
    model.eval()
    m = nn.Softmax(dim=1)
    y_test = torch.empty(0).to(device)
    y_pred = torch.empty(0).to(device)
    prob = torch.empty(0).to(device)
    with torch.no_grad():
        for i, data in enumerate(dloader, 0):
            x_categ, x_cont, cat_mask, con_mask = data[0].to(device), data[1].to(device),data[2].to(device),data[3].to(device)
            _ , x_categ_enc, x_cont_enc = embed_data_mask(x_categ, x_cont, cat_mask, con_mask,model)
            reps = model.transformer(x_categ_enc, x_cont_enc)
            y_reps = reps[:,model.num_categories-1,:]
            y_outs = model.mlpfory(y_reps)
            # import ipdb; ipdb.set_trace()
            y_test = torch.cat([y_test,x_categ[:,-1].float()],dim=0)
            y_pred = torch.cat([y_pred,torch.argmax(m(y_outs), dim=1).float()],dim=0)
            prob = torch.cat([prob,m(y_outs)[:,-1].float()],dim=0)

    correct_results_sum = (y_pred == y_test).sum().float()
    acc = correct_results_sum/y_test.shape[0]*100
    auc = roc_auc_score(y_score=prob.cpu(), y_true=y_test.cpu())
    return acc, auc


def multiclass_acc_justy(model,dloader,device):
    model.eval()
    vision_dset = True
    m = nn.Softmax(dim=1)
    y_test = torch.empty(0).to(device)
    y_pred = torch.empty(0).to(device)
    prob = torch.empty(0).to(device)
    with torch.no_grad():
        for i, data in enumerate(dloader, 0):
            x_categ, x_cont, cat_mask, con_mask = data[0].to(device), data[1].to(device),data[2].to(device),data[3].to(device)
            _ , x_categ_enc, x_cont_enc = embed_data_mask(x_categ, x_cont, cat_mask, con_mask,model,vision_dset)
            reps = model.transformer(x_categ_enc, x_cont_enc)
            y_reps = reps[:,model.num_categories-1,:]
            y_outs = model.mlpfory(y_reps)
            # import ipdb; ipdb.set_trace()
            y_test = torch.cat([y_test,x_categ[:,-1].float()],dim=0)
            y_pred = torch.cat([y_pred,torch.argmax(m(y_outs), dim=1).float()],dim=0)

    correct_results_sum = (y_pred == y_test).sum().float()
    acc = correct_results_sum/y_test.shape[0]*100
    return acc, 0


def classification_scores(model, dloader, device, task,vision_dset):
    model.eval()
    m = nn.Softmax(dim=1)
    y_test = torch.empty(0).to(device)
    y_pred = torch.empty(0).to(device)
    prob = torch.empty(0).to(device)
    with torch.no_grad():
        for i, data in enumerate(dloader, 0):
            x_categ, x_cont, y_gts, cat_mask, con_mask = data[0].to(device), data[1].to(device),data[2].to(device),data[3].to(device),data[4].to(device)
            _ , x_categ_enc, x_cont_enc = embed_data_mask(x_categ, x_cont, cat_mask, con_mask,model,vision_dset)
            reps = model.transformer(x_categ_enc, x_cont_enc)
            y_reps = reps[:,0,:]
            y_outs = model.mlpfory(y_reps)
            # import ipdb; ipdb.set_trace()
            y_test = torch.cat([y_test,y_gts.squeeze().float()],dim=0)
            y_pred = torch.cat([y_pred,torch.argmax(y_outs, dim=1).float()],dim=0)
            if task == 'binary':
                prob = torch.cat([prob,m(y_outs)[:,-1].float()],dim=0)

    correct_results_sum = (y_pred == y_test).sum().float()
    acc = correct_results_sum/y_test.shape[0]*100
    auc = 0
    if task == 'binary':
        auc = roc_auc_score(y_score=prob.cpu(), y_true=y_test.cpu())
    return acc.cpu().numpy(), auc

def mean_sq_error(model, dloader, device, vision_dset):
    model.eval()
    y_test = torch.empty(0).to(device)
    y_pred = torch.empty(0).to(device)
    with torch.no_grad():
        for i, data in enumerate(dloader, 0):
            x_categ, x_cont, y_gts, cat_mask, con_mask = data[0].to(device), data[1].to(device),data[2].to(device),data[3].to(device),data[4].to(device)
            _ , x_categ_enc, x_cont_enc = embed_data_mask(x_categ, x_cont, cat_mask, con_mask,model,vision_dset)
            reps = model.transformer(x_categ_enc, x_cont_enc)
            y_reps = reps[:,0,:]
            y_outs = model.mlpfory(y_reps)
            y_test = torch.cat([y_test,y_gts.squeeze().float()],dim=0)
            y_pred = torch.cat([y_pred,y_outs],dim=0)
        # import ipdb; ipdb.set_trace()
        rmse = mean_squared_error(y_test.cpu(), y_pred.cpu(), squared=False)
        return rmse


def generate_outputs(model, dloader, device, task,vision_dset):
    model.eval()
    m = nn.Softmax(dim=1)
    y_test = torch.empty(0).to(device)
    y_pred = torch.empty(0).to(device)
    prob = torch.empty(0).to(device)
    with torch.no_grad():
        for i, data in enumerate(dloader, 0):
            x_categ, x_cont, y_gts, cat_mask, con_mask = data[0].to(device), data[1].to(device),data[2].to(device),data[3].to(device),data[4].to(device)
            _ , x_categ_enc, x_cont_enc = embed_data_mask(x_categ, x_cont, cat_mask, con_mask,model,vision_dset)
            reps = model.transformer(x_categ_enc, x_cont_enc)
            y_reps = reps[:,0,:]
            y_outs = model.mlpfory(y_reps)
            # import ipdb; ipdb.set_trace()
            y_test = torch.cat([y_test,y_gts.squeeze().float()],dim=0)
            if task == 'regression':
                y_pred = torch.cat([y_pred,y_outs],dim=0)
            else:
                y_pred = torch.cat([y_pred,torch.argmax(y_outs, dim=1).float()],dim=0)
            if task == 'binary':
                prob = torch.cat([prob,m(y_outs)[:,-1].float()],dim=0)
    if task == 'binary':
        return prob.cpu(), y_test.cpu()
    return y_pred.cpu(), y_test.cpu()