|
1
|
|
|
# Author: Simon Blanke |
|
2
|
|
|
# Email: [email protected] |
|
3
|
|
|
# License: MIT License |
|
4
|
|
|
|
|
5
|
|
|
import numpy as np |
|
6
|
|
|
import pandas as pd |
|
7
|
|
|
import plotly as py |
|
8
|
|
|
import plotly.graph_objects as go |
|
9
|
|
|
import plotly.express as px |
|
10
|
|
|
|
|
11
|
|
|
from hyperactive import Hyperactive |
|
12
|
|
|
|
|
13
|
|
|
|
|
14
|
|
|
class Insight: |
|
15
|
|
|
def __init__(self, search_config, X, y): |
|
16
|
|
|
self.search_config = search_config |
|
17
|
|
|
self.X = X |
|
18
|
|
|
self.y = y |
|
19
|
|
|
|
|
20
|
|
|
|
|
21
|
|
|
def plot_performance(self, runs=3, path=None, optimizers="all"): |
|
22
|
|
|
if optimizers == "all": |
|
23
|
|
|
optimizers = [ |
|
24
|
|
|
"HillClimbing", |
|
25
|
|
|
"StochasticHillClimbing", |
|
26
|
|
|
"TabuSearch", |
|
27
|
|
|
"RandomSearch", |
|
28
|
|
|
"RandomRestartHillClimbing", |
|
29
|
|
|
"RandomAnnealing", |
|
30
|
|
|
"SimulatedAnnealing", |
|
31
|
|
|
"StochasticTunneling", |
|
32
|
|
|
"ParallelTempering", |
|
33
|
|
|
"ParticleSwarm", |
|
34
|
|
|
"EvolutionStrategy", |
|
35
|
|
|
"Bayesian"] |
|
36
|
|
|
|
|
37
|
|
|
eval_times = [] |
|
38
|
|
|
total_times = [] |
|
39
|
|
|
for run in range(runs): |
|
40
|
|
|
|
|
41
|
|
|
eval_time = [] |
|
42
|
|
|
total_time = [] |
|
43
|
|
|
for optimizer in optimizers: |
|
44
|
|
|
opt = Hyperactive(self.X, self.y, memory=False) |
|
45
|
|
|
opt.search(self.search_config, n_iter=3, optimizer=optimizer) |
|
46
|
|
|
|
|
47
|
|
|
eval_time.append(opt.eval_time) |
|
48
|
|
|
total_time.append(opt.total_time) |
|
49
|
|
|
|
|
50
|
|
|
eval_time = np.array(eval_time) |
|
51
|
|
|
total_time = np.array(total_time) |
|
52
|
|
|
|
|
53
|
|
|
eval_times.append(eval_time) |
|
54
|
|
|
total_times.append(total_time) |
|
55
|
|
|
|
|
56
|
|
|
eval_times = np.array(eval_times) |
|
57
|
|
|
total_times = np.array(total_times) |
|
58
|
|
|
opt_times = np.subtract(total_times, eval_times) |
|
59
|
|
|
|
|
60
|
|
|
opt_time_mean = opt_times.mean(axis=0) |
|
61
|
|
|
eval_time_mean = eval_times.mean(axis=0) |
|
62
|
|
|
total_time_mean = total_times.mean(axis=0) |
|
63
|
|
|
|
|
64
|
|
|
opt_time_std = opt_times.std(axis=0) |
|
65
|
|
|
eval_time_std = eval_times.std(axis=0) |
|
66
|
|
|
|
|
67
|
|
|
eval_time = eval_time_mean/total_time_mean |
|
68
|
|
|
opt_time = opt_time_mean/total_time_mean |
|
69
|
|
|
|
|
70
|
|
|
fig = go.Figure(data=[ |
|
71
|
|
|
go.Bar(name='Eval time', x=optimizers, y=eval_time), |
|
72
|
|
|
go.Bar(name='Opt time', x=optimizers, y=opt_time) |
|
73
|
|
|
]) |
|
74
|
|
|
fig.update_layout(barmode='stack') |
|
75
|
|
|
py.offline.plot(fig, filename="sampleplot.html") |
|
76
|
|
|
|
|
77
|
|
|
|
|
78
|
|
|
def plot_search_path(self, path=None, optimizers=["HillClimbing"]): |
|
79
|
|
|
for optimizer in optimizers: |
|
80
|
|
|
opt = Hyperactive(self.X, self.y, memory=False, verbosity=10) |
|
81
|
|
|
opt.search(self.search_config, n_iter=20, optimizer=optimizer) |
|
82
|
|
|
|
|
83
|
|
|
pos_list = opt.pos_list |
|
84
|
|
|
score_list = opt.score_list |
|
85
|
|
|
|
|
86
|
|
|
pos_list = np.array(pos_list) |
|
87
|
|
|
score_list = np.array(score_list) |
|
88
|
|
|
|
|
89
|
|
|
pos_list = np.squeeze(pos_list) |
|
90
|
|
|
score_list = np.squeeze(score_list) |
|
91
|
|
|
|
|
92
|
|
|
df = pd.DataFrame( |
|
93
|
|
|
{"n_neighbors": pos_list[:, 0], "leaf_size": pos_list[:, 1], "score": score_list} |
|
94
|
|
|
) |
|
95
|
|
|
|
|
96
|
|
|
layout = go.Layout( |
|
97
|
|
|
xaxis=dict( |
|
98
|
|
|
range=[0, 50] |
|
99
|
|
|
), |
|
100
|
|
|
yaxis=dict( |
|
101
|
|
|
range=[0, 50] |
|
102
|
|
|
) |
|
103
|
|
|
) |
|
104
|
|
|
fig = go.Figure(data=go.Scatter( |
|
105
|
|
|
x = df["n_neighbors"], |
|
106
|
|
|
y = df["leaf_size"], |
|
107
|
|
|
mode='lines+markers', |
|
108
|
|
|
marker=dict( |
|
109
|
|
|
size=10, |
|
110
|
|
|
color=df["score"], |
|
111
|
|
|
colorscale='Viridis', # one of plotly colorscales |
|
112
|
|
|
showscale=True |
|
113
|
|
|
) |
|
114
|
|
|
), |
|
115
|
|
|
layout=layout |
|
116
|
|
|
) |
|
117
|
|
|
|
|
118
|
|
|
py.offline.plot(fig, filename="search_path" + optimizer + ".html") |
SimonBlanke /
Hyperactive
