-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_statistical_metrics.py
198 lines (172 loc) · 10.7 KB
/
main_statistical_metrics.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
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
#!/usr/bin/env python
# ------------------------------------------------------------------------------------------------------%
# Created by "Thieu" at 14:05, 28/01/2021 %
# %
# Email: [email protected] %
# Homepage: https://www.researchgate.net/profile/Nguyen_Thieu2 %
# Github: https://github.com/thieu1995 %
# ------------------------------------------------------------------------------------------------------%
### Reading all results files to find True pareto-fronts (Reference Fronts)
from time import time
from pathlib import Path
from copy import deepcopy
from config import Config, OptExp, OptParas
from pandas import read_csv, DataFrame, to_numeric
from numpy import array, vstack, hstack, std
from utils.io_util import load_tasks, load_nodes
from utils.metric_util import *
from utils.visual.scatter import visualize_front_3d
def inside_loop(my_model, n_trials, n_timebound, epoch, fe, end_paras):
for pop_size in OptExp.POP_SIZE:
if Config.TIME_BOUND_KEY:
path_results = f'{Config.RESULTS_DATA}/{n_timebound}s/task_{my_model["problem"]["n_tasks"]}/{Config.METRICS}/{my_model["name"]}/{n_trials}'
else:
path_results = f'{Config.RESULTS_DATA}/no_time_bound/task_{my_model["problem"]["n_tasks"]}/{Config.METRICS}/{my_model["name"]}/{n_trials}'
name_paras = f'{epoch}_{pop_size}_{end_paras}'
file_name = f'{path_results}/experiment_results/{name_paras}-results.csv'
df = read_csv(file_name, usecols=["Power", "Latency", "Cost"])
return df.values
def getting_results_for_task(models):
matrix_fit = zeros((1, 6))
for n_task in OptExp.N_TASKS:
for my_model in models:
tasks = load_tasks(f'{Config.INPUT_DATA}/tasks_{n_task}.json')
problem = deepcopy(my_model['problem'])
problem["tasks"] = tasks
problem["n_tasks"] = n_task
problem["shape"] = [len(problem["clouds"]) + len(problem["fogs"]), n_task]
my_model['problem'] = problem
for n_trials in range(OptExp.N_TRIALS):
if Config.TIME_BOUND_KEY:
for n_timebound in OptExp.TIME_BOUND_VALUES:
if Config.MODE == "epoch":
for epoch in OptExp.EPOCH:
end_paras = f"{epoch}"
df_matrix = inside_loop(my_model, n_trials, n_timebound, epoch, None, end_paras)
df_name = array([[n_task, my_model["name"], n_trials], ] * len(df_matrix))
matrix = hstack((df_name, df_matrix))
matrix_fit = vstack((matrix_fit, matrix))
else:
if Config.MODE == "epoch":
for epoch in OptExp.EPOCH:
end_paras = f"{epoch}"
df_matrix = inside_loop(my_model, n_trials, None, epoch, None, end_paras)
df_name = array([[n_task, my_model["name"], n_trials], ] * len(df_matrix))
matrix = hstack((df_name, df_matrix))
matrix_fit = vstack((matrix_fit, matrix))
return matrix_fit[1:]
starttime = time()
clouds, fogs, peers = load_nodes(f'{Config.INPUT_DATA}/nodes_4_10_7.json')
problem = {
"clouds": clouds,
"fogs": fogs,
"peers": peers,
"n_clouds": len(clouds),
"n_fogs": len(fogs),
"n_peers": len(peers),
}
models = [
{"name": "NSGA-II", "class": "BaseNSGA_II", "param_grid": OptParas.NSGA_II, "problem": problem},
{"name": "NSGA-III", "class": "BaseNSGA_III", "param_grid": OptParas.NSGA_III, "problem": problem},
{"name": "NSGAII_SDE", "class": "BaseNSGAII_SDE", "param_grid": OptParas.NSGAII_SDE, "problem": problem},
{"name": "MO-PSO-RI", "class": "BaseMOPSORI", "param_grid": OptParas.MOPSORI, "problem": problem},
{"name": "MO_ILCO", "class": "MO_ILCO", "param_grid": OptParas.MO_ILCO, "problem": problem},
{"name": "MO_ILCO_2", "class": "MO_ILCO_2", "param_grid": OptParas.MO_ILCO, "problem": problem},
{"name": "MO-ALO", "class": "BaseMO_ALO", "param_grid": OptParas.MO_ALO, "problem": problem},
{"name": "MO-SSA", "class": "BaseMO_SSA", "param_grid": OptParas.MO_SSA, "problem": problem},
#{"name": "NS-SSA", "class": "BaseNS_SSA", "param_grid": OptParas.MO_SSA, "problem": problem},
#{"name": "Improved-NSGA-III", "class": "Improved_NSGA_III", "param_grid": OptParas.IMPROVED_NSGA_III, "problem": problem},
]
## Load all results of all trials
matrix_results = getting_results_for_task(models)
# df_full = DataFrame(matrix_results, columns=["Task", "Model", "Trial", "Fit1", "Fit2", "Fit3"])
data = {'Task': matrix_results[:, 0],
'Model': matrix_results[:, 1],
'Trial': matrix_results[:, 2],
'Fit1': matrix_results[:, 3],
'Fit2': matrix_results[:, 4],
'Fit3': matrix_results[:, 5],
}
df_full = DataFrame(data)
df_full["Task"] = to_numeric(df_full["Task"])
df_full["Trial"] = to_numeric(df_full["Trial"])
df_full["Fit1"] = to_numeric(df_full["Fit1"])
df_full["Fit2"] = to_numeric(df_full["Fit2"])
df_full["Fit3"] = to_numeric(df_full["Fit3"])
for n_task in OptExp.N_TASKS:
performance_results = []
performance_results_mean = []
## Find matrix results for each problem
df_task = df_full[df_full["Task"] == n_task]
matrix_task = df_task[['Fit1', 'Fit2', 'Fit3']].values
hyper_point = max(matrix_task, axis=0)
## Find non-dominated matrix for each problem
reference_fronts = zeros((1, 3))
dominated_list = find_dominates_list(matrix_task)
for idx, value in enumerate(dominated_list):
if value == 0:
reference_fronts = vstack((reference_fronts, matrix_task[idx]))
reference_fronts = reference_fronts[1:]
## For each model and each trial, calculate its performance metrics
for model in models:
er_list = zeros(OptExp.N_TRIALS)
gd_list = zeros(OptExp.N_TRIALS)
igd_list = zeros(OptExp.N_TRIALS)
ste_list = zeros(OptExp.N_TRIALS)
hv_list = zeros(OptExp.N_TRIALS)
har_list = zeros(OptExp.N_TRIALS)
for trial in range(OptExp.N_TRIALS):
df_result = df_task[ (df_task["Model"] == model["name"]) & (df_task["Trial"] == trial) ]
pareto_fronts = array(df_result.values[:, 3:], dtype=float)
er = error_ratio(pareto_fronts, reference_fronts)
gd = generational_distance(pareto_fronts, reference_fronts)
igd = inverted_generational_distance(pareto_fronts, reference_fronts)
ste = spacing_to_extent(pareto_fronts)
hv = hyper_volume(pareto_fronts, reference_fronts, hyper_point, 100)
har = hyper_area_ratio(pareto_fronts, reference_fronts, hyper_point, 100)
performance_results.append([n_task, model["name"], trial, er, gd, igd, ste, hv, har])
er_list[trial] = er
gd_list[trial] = gd
igd_list[trial] = igd
ste_list[trial] = ste
hv_list[trial] = hv
har_list[trial] = har
er_min, er_max, er_mean, er_std, er_cv = min(er_list), max(er_list), mean(er_list), std(er_list), std(er_list)/mean(er_list)
gd_min, gd_max, gd_mean, gd_std, gd_cv = min(gd_list), max(gd_list), mean(gd_list), std(gd_list), std(gd_list)/mean(gd_list)
igd_min, igd_max, igd_mean, igd_std, igd_cv = min(igd_list), max(igd_list), mean(igd_list), std(igd_list), std(igd_list)/mean(igd_list)
ste_min, ste_max, ste_mean, ste_std, ste_cv = min(ste_list), max(ste_list), mean(ste_list), std(ste_list), std(ste_list)/mean(ste_list)
hv_min, hv_max, hv_mean, hv_std, hv_cv = min(hv_list), max(hv_list), mean(hv_list), std(hv_list), std(hv_list) / mean(hv_list)
har_min, har_max, har_mean, har_std, har_cv = min(har_list), max(har_list), mean(har_list), std(har_list), std(har_list) / mean(har_list)
performance_results_mean.append([n_task, model["name"], er_min, er_max, er_mean, er_std, er_cv, gd_min, gd_max, gd_mean, gd_std, gd_cv,
igd_min, igd_max, igd_mean, igd_std, igd_cv, ste_min, ste_max, ste_mean, ste_std, ste_cv,
hv_min, hv_max, hv_mean, hv_std, hv_cv, har_min, har_max, har_mean, har_std, har_cv])
filepath1 = f'{Config.RESULTS_DATA}/100s/task_{n_task}/{Config.METRICS}/metrics'
Path(filepath1).mkdir(parents=True, exist_ok=True)
df1 = DataFrame(performance_results, columns=["Task", "Model", "Trial", "ER", "GD", "IGD", "STE", "HV", "HAR"])
df1.to_csv(f'{filepath1}/full_trials.csv', index=False)
df2 = DataFrame(performance_results_mean, columns=["Task", "Model", "ER-MIN", "ER-MAX", "ER-MEAN", "ER-STD", "ER-CV",
"GD-MIN", "GD-MAX", "GD-MEAN", "GD-STD", "GD-CV",
"IGD-MIN", "IGD-MAX", "IGD-MEAN", "IGD-STD", "IGD-CV",
"STE-MIN", "STE-MAX", "STE-MEAN", "STE-STD", "STE-CV",
"HV-MIN", "HV-MAX", "HV-MEAN", "HV-STD", "HV-CV",
"HAR-MIN", "HAR-MAX", "HAR-MEAN", "HAR-STD", "HAR-CV"])
df2.to_csv(f'{filepath1}/statistics.csv', index=False)
## Drawing some pareto-fronts founded. task --> trial ---> [modle1, model2, model3, ....]
filepath3 = f'{Config.RESULTS_DATA}/100s/task_{n_task}/{Config.METRICS}/visual/'
Path(filepath3).mkdir(parents=True, exist_ok=True)
print(filepath3)
labels = ["Power Consumption (Wh)", "Service Latency (s)", "Monetary Cost ($)"]
names = ["Reference Front"]
list_color = [Config.VISUAL_FRONTS_COLORS[0]]
list_marker = [Config.VISUAL_FRONTS_MARKERS[0]]
for trial in range(OptExp.N_TRIALS):
list_fronts = [reference_fronts, ]
for idx, model in enumerate(models):
df_result = df_task[ (df_task["Trial"] == trial) & (df_task["Model"] == model["name"]) ]
list_fronts.append(df_result[['Fit1', 'Fit2', 'Fit3']].values)
names.append(model["name"])
list_color.append(Config.VISUAL_FRONTS_COLORS[idx+1])
list_marker.append(Config.VISUAL_FRONTS_MARKERS[idx + 1])
filename = f'pareto_fronts-{n_task}-{trial}'
visualize_front_3d(list_fronts, labels, names, list_color, list_marker, filename, [filepath3, filepath3], inside=False)
print('That took: {} seconds'.format(time() - starttime))