-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathanalyse_result.py
160 lines (146 loc) · 8.8 KB
/
analyse_result.py
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
"""
python analyse_result.py --model BDCRNNVariableDecoder --dataset METR_LA --exp_id 89148
python analyse_result.py --model BDCRNNVariableDecoderShared --dataset PEMS_BAY --exp_id 67654
"""
import argparse
import os.path
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import torch
from analyse_testing_result import get_datetime
from libcity.config import ConfigParser
from libcity.model import loss
from libcity.utils import get_logger, ensure_dir, get_local_time
if __name__ == '__main__':
plt.rc('font', family='Times New Roman', size=20)
parser = argparse.ArgumentParser()
parser.add_argument('--task', type=str, default='traffic_state_pred')
parser.add_argument('--model', type=str, required=True)
parser.add_argument('--exp_id', type=str, required=True)
parser.add_argument('--dataset', type=str, required=True)
args = parser.parse_args()
task, model_name, dataset_name, exp_id = args.task, args.model, args.dataset, args.exp_id
# load config
config = ConfigParser(task, model_name, dataset_name, saved_model=False, train=False)
config['exp_id'] = exp_id
config['batch_size'] = 288 # 24h=24*60/5
# logger
logger = get_logger(config)
logger.info('Begin analyzing result, task={}, model_name={}, dataset_name={}, exp_id={}'.
format(str(task), str(model_name), str(dataset_name), str(exp_id)))
logger.info(config.config)
batch_size = config['batch_size']
evaluate_cache_dir = './libcity/cache/{}/evaluate_cache'.format(exp_id)
analyze_cache_dir = './libcity/cache/{}/analyze_cache/{}'.format(exp_id, get_local_time())
images_cache_dir = '{}/images'.format(analyze_cache_dir)
assert os.path.exists(evaluate_cache_dir)
ensure_dir(analyze_cache_dir)
ensure_dir(images_cache_dir)
visited = False
for filename in os.listdir(evaluate_cache_dir):
if filename[-4:] == '.npz':
assert not visited
arr = np.load(f'{evaluate_cache_dir}/{filename}')
visited = True
assert visited
prediction, truth, outputs, sigmas = arr['prediction'], arr['truth'], arr['outputs'], arr['sigmas']
logger.info(f'prediction shape: {prediction.shape}') # (6912, 12, 207, 1) (10432, 12, 325, 1)
logger.info(f'truth shape: {truth.shape}') # (6912, 12, 207, 1) (10432, 12, 325, 1)
logger.info(f'outputs shape: {outputs.shape}') # (30, 6912, 12, 207, 1) (30, 10432, 12, 325, 1)
logger.info(f'sigmas shape: {sigmas.shape}') # (30, 6912, 12, 207, 1) (30, 10432, 12, 325, 1)
assert outputs.shape == sigmas.shape
evaluate_rep, num_data, output_window, num_nodes, output_dim = outputs.shape
prediction, truth, outputs, sigmas = prediction[..., 0], truth[..., 0], outputs[..., 0], sigmas[..., 0]
if dataset_name == 'METR_LA':
nodes_list = [0, 149, 181]
elif dataset_name == 'PEMS_BAY':
nodes_list = [2, 4, 8, 57, 188, 113, 212, 258]
else:
raise NotImplementedError(f'No such dataset of {dataset_name}.')
metrics = ['masked_MAE', 'masked_MSE', 'masked_RMSE', 'masked_MAPE', 'MAE', 'MSE', 'RMSE', 'MAPE']
for i in nodes_list:
logger.info(f'Analyzing node {i}...')
t_dir = '{}/{}'.format(images_cache_dir, i)
ensure_dir(t_dir)
# (num_data, output_window), (evaluate_rep, num_data, output_window)
prediction_node, truth_node, outputs_node, sigmas_node = prediction[:, :, i], truth[:, :, i], \
outputs[:, :, :, i], sigmas[:, :, :, i]
output_res = np.zeros((12, 8))
for j in range(12):
for k, metric in enumerate(metrics):
if metric == 'masked_MAE':
output_res[j, k] = loss.masked_mae_torch(torch.Tensor(prediction_node[:, j - 1]), torch.Tensor(truth_node[:, j - 1]), 0).item()
elif metric == 'masked_MSE':
output_res[j, k] = loss.masked_mse_torch(torch.Tensor(prediction_node[:, j - 1]), torch.Tensor(truth_node[:, j - 1]), 0).item()
elif metric == 'masked_RMSE':
output_res[j, k] = loss.masked_rmse_torch(torch.Tensor(prediction_node[:, j - 1]), torch.Tensor(truth_node[:, j - 1]), 0).item()
elif metric == 'masked_MAPE':
output_res[j, k] = loss.masked_mape_torch(torch.Tensor(prediction_node[:, j - 1]), torch.Tensor(truth_node[:, j - 1]), 0).item()
elif metric == 'MAE':
output_res[j, k] = loss.masked_mae_torch(torch.Tensor(prediction_node[:, j - 1]), torch.Tensor(truth_node[:, j - 1])).item()
elif metric == 'MSE':
output_res[j, k] = loss.masked_mse_torch(torch.Tensor(prediction_node[:, j - 1]), torch.Tensor(truth_node[:, j - 1])).item()
elif metric == 'RMSE':
output_res[j, k] = loss.masked_rmse_torch(torch.Tensor(prediction_node[:, j - 1]), torch.Tensor(truth_node[:, j - 1])).item()
elif metric == 'MAPE':
output_res[j, k] = loss.masked_mape_torch(torch.Tensor(prediction_node[:, j - 1]), torch.Tensor(truth_node[:, j - 1])).item()
df = pd.DataFrame(output_res, columns=metrics)
# 显示所有列
pd.set_option('display.max_columns', None)
# 显示所有行
pd.set_option('display.max_rows', None)
# 设置value的显示长度为100，默认为50
pd.set_option('max_colwidth', 100)
logger.info(df)
sigmas_node_2 = sigmas_node * sigmas_node
outputs_node_2 = outputs_node * outputs_node
prediction_node_2 = prediction_node * prediction_node
writer = pd.ExcelWriter(f'{analyze_cache_dir}/{dataset_name}_node_{i}.xlsx')
for j, time in zip([2, 5, 11], ['15min', '30min', '1h']):
p, t, o, s = prediction_node[:, j], truth_node[:, j], outputs_node[:, :, j], sigmas_node[:, :, j]
p2, s2, o2 = prediction_node_2[:, j], sigmas_node_2[:, :, j], outputs_node_2[:, :, j]
error = p - t
a_uncertainty = 1 / evaluate_rep * np.sum(s, axis=0)
e_uncertainty = 1 / evaluate_rep * np.sum(o2, axis=0) - p2
uncertainty = a_uncertainty + e_uncertainty
logger.info(f'Masked Aleatoric={loss.masked_uncertainty_torch(torch.Tensor(a_uncertainty), torch.Tensor(t), 0).item()}, '
f'Masked Epistemic={loss.masked_uncertainty_torch(torch.Tensor(e_uncertainty), torch.Tensor(t), 0).item()}, '
f'Masked Uncertainty={loss.masked_uncertainty_torch(torch.Tensor(uncertainty), torch.Tensor(t), 0).item()}.')
res = np.stack((p, t, error, uncertainty, a_uncertainty, e_uncertainty), axis=1)
res = np.concatenate((res, o[:5].T, s[:5].T), axis=1)
columns_name = ['pred', 'truth', 'error', 'uncertainty', 'a_uncertainty', 'e_uncertainty']
# columns_name.extend(['output_{}'.format(i) for i in range(evaluate_rep)])
# columns_name.extend(['sigma_{}'.format(i) for i in range(evaluate_rep)])
columns_name.extend(['output_{}'.format(i) for i in range(5)])
columns_name.extend(['sigma_{}'.format(i) for i in range(5)])
pd_data = pd.DataFrame(res, columns=columns_name)
pd_data.to_excel(writer, sheet_name=time, float_format='%.4f')
if j == 2:
start = 235 if dataset_name == 'METR_LA' else 60 # 2012-06-05 00:00:00 / 2017-05-26 00:00:00
elif j == 5:
start = 232 if dataset_name == 'METR_LA' else 57
elif j == 11:
start = 226 if dataset_name == 'METR_LA' else 51
else:
raise NotImplementedError(f'No such timestamp of {j}')
# for k in range(start, num_data, batch_size):
# t_num = min(batch_size + 1, num_data - k)
# if t_num < batch_size + 1:
# continue
# x = np.arange(k, k + t_num)
# mask = np.where(t[k: k + t_num], 1, 0)
# plt.xlim(k, k + t_num - 1)
# plt.xticks(np.arange(k, k + t_num, 36), [get_datetime(dataset_name, k, j, fmt='%H:%M\n%b-%d')] +
# [f'{_}:00' for _ in range(3, 24, 3)] +
# [get_datetime(dataset_name, k + t_num - 1, j, fmt='%H:%M\n%b-%d')])
# plt.ylabel('mile/h')
# plt.plot(x, np.abs(error[k:k + t_num]) * mask, label='|error|')
# plt.plot(x, a_uncertainty[k:k + t_num] * mask, label='a_uncertainty')
# plt.plot(x, e_uncertainty[k:k + t_num] * mask, label='e_uncertainty')
# plt.plot(x, uncertainty[k:k + t_num] * mask, label='uncertainty')
# plt.legend(labelspacing=0.05)
# plt.savefig(f'{t_dir}/{dataset_name}_node_{i}_batch_{k // batch_size}_{time}.svg',
# bbox_inches='tight')
# plt.close()
writer.close()