FAMOS/main.py at main · alibabacloud-observability/FAMOS · 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
import argparse
import os.path
import time
import yaml
import json
import torch
from torch_geometric.data import DataLoader
from torch.nn.utils.rnn import pack_padded_sequence
from model.base_model import base_model
from tqdm import tqdm
import torch.nn as nn
import torch.optim as optim
import pandas as pd
from utils import set_seed, save_result, output_metrics, get_dataset, print_label


def get_mutimodal_feature(batch_data):
    node_length = [len(list_x) for list_x in batch_data.x]
    batch_trace_feature = torch.stack([x['trace_feature'] for list_x in batch_data.x for x in list_x])
    batch_metrics_feature = torch.stack(
        [x['metrics_feature'][:39, ] for list_x in batch_data.x for x in list_x]).permute(0, 2, 1)
    # Logs are variable-length time series and need padding
    batch_logs_feature = [x['logs_feature'] for list_x in batch_data.x for x in list_x]
    sequence_lengths = [len(seq) for seq in batch_logs_feature]
    # Sort by length in descending order and save the indices to restore the original order
    sequence_lengths, perm_idx = torch.sort(torch.LongTensor(sequence_lengths), descending=True)
    batch_logs_feature = [batch_logs_feature[i] for i in perm_idx]
    # Perform padding
    max_length = max(sequence_lengths)
    padded_sequences = torch.zeros(len(batch_logs_feature), max_length, batch_logs_feature[0].shape[1])
    for i, seq in enumerate(batch_logs_feature):
        end = sequence_lengths[i]
        padded_sequences[i, :end, :] = seq
    # Pack the sequences
    log_packed_input = pack_padded_sequence(padded_sequences, sequence_lengths, batch_first=True)
    return batch_trace_feature.cuda(), log_packed_input.cuda(), batch_metrics_feature.cuda(), perm_idx, batch_data.edge_index.cuda(), node_length


def testing(model):
    # Testing
    model.eval()
    y_pred = []
    y_true = []
    with torch.no_grad():
        for batch_data in test_loader:
            batch_trace_feature, log_packed_input, batch_metrics_feature, perm_idx, edge_index, node_length = get_mutimodal_feature(
                batch_data)
            label = (batch_data.label).cuda()
            output = model(batch_trace_feature, log_packed_input, batch_metrics_feature, perm_idx, edge_index,
                           node_length)
            y_pred += list(torch.argsort(output, dim=1, descending=True).cpu().numpy())
            y_true += list(label.cpu().numpy())
    result_df = pd.DataFrame(
        {
            "y_pred": [list(item) for item in y_pred],
            "y_true": y_true
        }
    )
    return output_metrics(result_df)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="FAMOS")
    parser.add_argument("--config", type=str, default="train-ticket.yaml")
    args = parser.parse_args()
    path = "configs/" + args.config
    with open(path, "r") as f:
        config = yaml.safe_load(f)
    print(json.dumps(config, indent=4))
    if not os.path.exists('./result'): os.makedirs('./result')
    if not os.path.exists('./dataset'): os.makedirs('./dataset')
    set_seed(config['seed'])
    model = base_model(embedding_size=config['word_embedding_size'], metric_feature_size=config['metric_feature_size'],
                       num_classes=config['num_classes'])
    model = model.cuda()
    # Create DataLoader
    X_train, X_test = get_dataset(config)
    y_train = [data.label for data in X_train]
    y_test = [data.label for data in X_test]
    print_label(y_train, 'Training set')
    print_label(y_test, 'Testing set')
    train_loader = DataLoader(X_train, batch_size=config['batch_size'], shuffle=True)
    test_loader = DataLoader(X_test, batch_size=config['batch_size'], shuffle=True)
    criterion = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=float(config['lr']), weight_decay=5e-6)
    average_loss_dict = {}
    iter = tqdm(range(config['epoch']), desc="Training: ")
    for epoch in iter:
        model.train()
        total_loss = 0.0
        for batch_data in train_loader:
            batch_trace_feature, log_packed_input, batch_metrics_feature, perm_idx, edge_index, node_length = get_mutimodal_feature(
                batch_data)
            label = (batch_data.label).cuda()
            output = model(batch_trace_feature, log_packed_input, batch_metrics_feature, perm_idx, edge_index,
                           node_length)
            loss = criterion(output, label)
            # Backward and optimize
            optimizer.zero_grad()
            loss.backward()
            total_loss += loss.item()
            optimizer.step()
        average_loss = total_loss / len(train_loader)
        average_loss_dict[epoch] = average_loss
        iter.set_postfix({'avg_loss': average_loss})
    model.eval()
    y_pred = []
    y_true = []
    s_t = time.time()
    with torch.no_grad():
        for batch_data in test_loader:
            batch_trace_feature, log_packed_input, batch_metrics_feature, perm_idx, edge_index, node_length = get_mutimodal_feature(
                batch_data)
            label = (batch_data.label).cuda()
            output = model(batch_trace_feature, log_packed_input, batch_metrics_feature, perm_idx, edge_index,
                           node_length)
            y_pred += list(torch.argsort(output, dim=1, descending=True).cpu().numpy())
            y_true += list(label.cpu().numpy())
    e_t = time.time()
    i_pre_time = (e_t - s_t) / len(X_test)
    print(f'Inference time: {i_pre_time}')
    result_df = pd.DataFrame(
        {
            "y_pred": [list(item) for item in y_pred],
            "y_true": y_true
        }
    )
    save_result(result_df, config)