Eff_pattern_matching/evaluate.py at master · SepKfr/Eff_pattern_matching · 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
import argparse
import json
import random

import numpy as np
import pandas as pd
import torch
import torch.nn as nn
import os

from Utils.base_train import batching, ModelData, batch_sampled_data
from data.data_loader import ExperimentConfig
from models.Transformers import Transformer
from models.rnn import RNN

parser = argparse.ArgumentParser(description="preprocess argument parser")
parser.add_argument("--attn_type", type=str, default='basic_attn')
parser.add_argument("--name", type=str, default='basic_attn')
parser.add_argument("--exp_name", type=str, default='covid')
parser.add_argument("--cuda", type=str, default="cuda:0")
parser.add_argument("--pred_len", type=int, default=24)


args = parser.parse_args()

kernel = [1, 3, 6, 9]
n_heads = 8
d_model = [16, 32]
batch_size = 256

config = ExperimentConfig(args.pred_len, args.exp_name)

formatter = config.make_data_formatter()
params = formatter.get_experiment_params()
total_time_steps = params['total_time_steps']
num_encoder_steps = params['num_encoder_steps']
column_definition = params["column_definition"]
pred_len = args.pred_len

device = torch.device(args.cuda if torch.cuda.is_available() else "cpu")
data_csv_path = "{}.csv".format(args.exp_name)
raw_data = pd.read_csv(data_csv_path)

data = formatter.transform_data(raw_data)
train_max, valid_max = formatter.get_num_samples_for_calibration(num_train=batch_size)
max_samples = (train_max, valid_max)

train, valid, test = batch_sampled_data(data, 0.8, max_samples, params['total_time_steps'],
                                        params['num_encoder_steps'], pred_len,
                                        params["column_definition"],
                                        device)

test_batching = batching(batch_size, test.enc, test.dec, test.y_true, test.y_id)
test = ModelData(test_batching[0], test_batching[1], test_batching[2], test_batching[3], device)

device = torch.device(args.cuda if torch.cuda.is_available() else "cpu")
model_path = "models_{}_{}".format(args.exp_name, pred_len)
model_params = formatter.get_default_model_params()

src_input_size = test.enc.shape[3]
tgt_input_size = test.dec.shape[3]

predictions = np.zeros((3, test.y_true.shape[0], test.y_true.shape[1], test.y_true.shape[2]))
n_batches_test = test.enc.shape[0]
y_true = test.y_true.squeeze(-1).detach().cpu()

mse = nn.MSELoss()
mae = nn.L1Loss()
stack_size = 1


for i, seed in enumerate([7835, 5129, 9468]):
    try:
        for d in d_model:
            for k in kernel:
                d_k = int(d / n_heads)

                if args.name == "lstm":
                    model = RNN(n_layers=stack_size,
                                hidden_size=d,
                                src_input_size=src_input_size,
                                tgt_input_size=tgt_input_size,
                                rnn_type="lstm",
                                device=device,
                                d_r=0,
                                seed=seed,
                                pred_len=pred_len)
                else:

                    model = Transformer(src_input_size=src_input_size,
                                        tgt_input_size=tgt_input_size,
                                        pred_len=pred_len,
                                        d_model=d,
                                        d_ff=d * 4,
                                        d_k=d_k, d_v=d_k, n_heads=n_heads,
                                        n_layers=stack_size, src_pad_index=0,
                                        tgt_pad_index=0, device=device,
                                        attn_type=args.attn_type,
                                        seed=seed, kernel=k)

                checkpoint = torch.load(os.path.join("models_{}_{}".format(args.exp_name, args.pred_len),
                                        "{}_{}".format(args.name, seed)))
                model.load_state_dict(checkpoint['model_state_dict'])
                model.eval()
                model.to(device)

                for j in range(n_batches_test):
                    output = model(test.enc[j], test.dec[j])
                    predictions[i, j] = output.squeeze(-1).cpu().detach().numpy()

    except RuntimeError:
        pass

predictions = torch.from_numpy(np.mean(predictions, axis=0))

results = torch.zeros(2, args.pred_len)
normaliser = y_true.abs().mean()

test_loss = mse(predictions, y_true).item() / normaliser
mae_loss = mae(predictions, y_true).item() / normaliser

for j in range(args.pred_len):

    results[0, j] = mse(predictions[:, :, j], y_true[:, :, j]).item()
    results[1, j] = mae(predictions[:, :, j], y_true[:, :, j]).item()

df = pd.DataFrame(results.detach().cpu().numpy())
df.to_csv("{}_{}_{}.csv".format(args.exp_name, args.name, args.pred_len))

erros = dict()
erros["{}".format(args.name)] = list()
erros["{}".format(args.name)].append(float("{:.5f}".format(test_loss)))
erros["{}".format(args.name)].append(float("{:.5f}".format(mae_loss)))

error_path = "final_errors_{}_{}.json".format(args.exp_name, pred_len)

if os.path.exists(error_path):
    with open(error_path) as json_file:
        json_dat = json.load(json_file)
        if json_dat.get("{}".format(args.name)) is None:
            json_dat["{}".format(args.name)] = list()
        json_dat["{}".format(args.name)].append(float("{:.5f}".format(test_loss)))
        json_dat["{}".format(args.name)].append(float("{:.5f}".format(mae_loss)))

    with open(error_path, "w") as json_file:
        json.dump(json_dat, json_file)
else:
    with open(error_path, "w") as json_file:
        json.dump(erros, json_file)