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
|
import numpy as np
from models.MetaModel import MetaModel
def accuracy_evaluator(
real_data,
multi_model,
compute_mape=True,
compute_nad=True,
compute_rmsle=True,
rmsle_hyperparameter=0.5,
only_metamodel=False
):
"""
:param real_data: the real-world data of the simulation
:param multi_model: the Multi-Model, containing individual models (possibly also a Meta-Model, with id=101)
:param MAPE: whether to calculate Mean Absolute Percentage Error (MAPE)
:param NAD: whether to calculate Normalized Absolute Differences (NAD)
:param RMSLE: whether to calculate Root Mean Square Logarithmic Error (RMSLE)
:param rmsle_hyperparameter: the hyperparameter that balances the ration underestimations:overestimations
- default is 0.5 (balanced penalty)
- < 0.5: more penalty for overestimations
- > 0.5: more penalty for underestimations
e.g., RMSLE_hyperparameter=0.3 -> 30% penalty for overestimations, 70% penalty for underestimations (3:7 ratio)
:return: None, but prints the accuracy metrics
"""
meta_model = MetaModel(multimodel=multi_model)
multi_model.models.append(meta_model.meta_model) # metamodel
# multi_model.models.append(Model(raw_host_data=real_data, id=-1, path=None)) # real-world data
with open(multi_model.output_folder_path + "/accuracy_report.txt", "a") as f:
f.write("====================================\n")
f.write("Accuracy Report, against ground truth\n")
for model in multi_model.models:
if only_metamodel and model.id != 101:
continue
if model.id == -1:
f.write("Real-World data")
elif model.id == 101:
f.write(
f"Meta-Model, meta-function: {multi_model.user_input['meta_function']}, window_size: {meta_model.multi_model.window_size}")
else:
f.write(f"Model {model.id}")
simulation_data = model.raw_sim_data
min_len = min(len(real_data), len(simulation_data))
real_data = real_data[:min_len]
simulation_data = simulation_data[:min_len]
if compute_mape:
accuracy_mape = mape(
real_data=real_data,
simulation_data=simulation_data
)
f.write(f"\nMean Absolute Percentage Error (MAPE): {accuracy_mape}%")
if compute_nad:
accuracy_nad = nad(
real_data=real_data,
simulation_data=simulation_data
)
f.write(f"\nNormalized Absolute Differences (NAD): {accuracy_nad}%")
if compute_rmsle:
accuracy_rmsle = rmsle(
real_data=real_data,
simulation_data=simulation_data,
alpha=rmsle_hyperparameter
)
f.write(
f"\nRoot Mean Square Logarithmic Error (RMSLE), alpha={rmsle_hyperparameter}:{accuracy_rmsle}\n\n")
f.write("====================================\n")
def mape(real_data, simulation_data):
"""
Calculate Mean Absolute Percentage Error (MAPE)
:param real_data: Array of real values
:param simulation_data: Array of simulated values
:return: MAPE value
"""
real_data = np.array(real_data)
simulation_data = np.array(simulation_data)
return round(np.mean(np.abs((real_data - simulation_data) / real_data)) * 100, 3)
def nad(real_data, simulation_data):
"""
Calculate Normalized Absolute Differences (NAD)
:param real_data: Array of real values
:param simulation_data: Array of simulated values
:return: NAD value
"""
real_data = np.array(real_data)
simulation_data = np.array(simulation_data)
return round(np.sum(np.abs(real_data - simulation_data)) / np.sum(real_data) * 100, 3)
def rmsle(real_data, simulation_data, alpha=0.5):
"""
Calculate Root Mean Square Logarithmic Error (RMSLE) with an adjustable alpha parameter
:param real_data: Array of real values
:param simulation_data: Array of simulated values
:param alpha: Hyperparameter that balances the penalty between underestimations and overestimations
:return: RMSLE value
"""
real_data = np.array(real_data)
simulation_data = np.array(simulation_data)
log_diff = alpha * np.log(real_data) - (1 - alpha) * np.log(simulation_data)
return round(np.sqrt(np.mean(log_diff ** 2)) * 100, 3)
|
