-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathanalyze_tunability.py
139 lines (115 loc) · 6.01 KB
/
analyze_tunability.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
#
# SPDX-FileCopyrightText: 2020 Idiap Research Institute <[email protected]>
#
# Written by Prabhu Teja <[email protected]>,
# Florian Mai <[email protected]>
# Thijs Vogels <[email protected]>
#
# SPDX-License-Identifier: MIT
#
import multiprocessing as mp
import os
import random
import sys
import argparse
import matplotlib as mpl
import matplotlib.gridspec as gridspec
import matplotlib.pyplot as plt
sys.path.append("./DeepOBS/")
from deepobs import analyzer
from deepobs.analyzer import (get_hyperparameter_optimization_performance,
plot_box_hyperparam_optim,
plot_hyperparam_optimization,
print_tunability_to_latex)
def temp(x):
return get_hyperparameter_optimization_performance(x[0], num_shuffle=x[1])
def main():
parser = argparse.ArgumentParser(description='Computes summary metrics and boxplots for any given problem.')
parser.add_argument('-inpath', type=str, required=True,
help='Basepath to all the logs.')
parser.add_argument('-problem', type=str, required=True, choices = ["fmnist_2c2d", "mnist_vae", "quadratic_deep", "fmnist_vae", "cifar100_allcnnc", "cifar10_3c3d", "imdb_bilstm", "svhn_wrn164", "tolstoi_char_rnn"], help = "Problem to plot the boxplot for.")
parser.add_argument('-optimizers', type=str, nargs = "+", required=True, help = "List of optimizers to plot.")
parser.add_argument('-optimizer_labels', type=str, nargs = "+", required=False, default = None, help = "List of labels to display for each optimizer given in 'optimizers'. If None, the optimizer names are taken directly as labels.")
parser.add_argument('-num_shuffle', default=100, type=int, help="Number of times to shuffle for computing expected validation performance via bootstrapping.")
parser.add_argument('-outfile', type=str, default = "boxplot.pdf", help="Path to store the plot.")
parser.add_argument('-x_axis', type=str, default = "trials", choices = ["trials", "wct"], help="What to consider as budget, either 'trials' (number of hyperparameter configurations) or 'wct' (wallclock-time).")
parser.add_argument('-print_metrics', action="store_true", help="If set, print metrics (CPE, CPU, etc...) instead of plotting boxplots.")
args = parser.parse_args()
random.seed(a=12) # fix random seed for when shuffling the random search results.
num_shuffle = args.num_shuffle
mpl.use('pgf')
pgf_with_latex = { # setup matplotlib to use latex for output
"pgf.texsystem": "pdflatex", # change this if using xetex or lautex
"text.usetex": True, # use LaTeX to write all text
"font.family": "serif",
"font.serif": [], # blank entries should cause plots
"font.sans-serif": [], # to inherit fonts from the document
"font.monospace": [],
"font.size": 10,
"legend.fontsize": 10, # Make the legend/label fonts
"xtick.labelsize": 20, # a little smaller
"ytick.labelsize": 10, # default fig size of 0.9 textwidth
"axes.labelsize": 28,
"pgf.preamble": [
r"\usepackage[utf8]{inputenc}", # use utf8 input and T1 fonts
r"\usepackage[T1]{fontenc}", # plots will be generated
] # using this preamble
}
#mpl.use("agg")
mpl.rcParams.update(pgf_with_latex)
included_metrics = ["Cumulative Performance Early", "Cumulative Performance Uniform", "Cumulative Performance Late", "Sharpness", "Avg WCT", "k=0.1", "k=0.01", "k=0.001", "k=0.0001"]
problems = [args.problem]
x_axis = args.x_axis
axis_limits = {
'fmnist_2c2d': (80, 94),
'mnist_vae': (25, 100),
'quadratic_deep': (80, 150),
'fmnist_vae': (20, 80),
'cifar100_allcnnc': (1, 50),
'cifar10_3c3d': (40, 90),
'imdb_bilstm': (60, 90),
'svhn_wrn164': (80, 99),
'tolstoi_char_rnn': (10, 70)
}
plot_labels = {
'fmnist_2c2d': 'FMNIST Classification',
'mnist_vae': 'MNIST VAE',
'quadratic_deep': 'Quadratic Deep',
'fmnist_vae': 'FMNIST VAE',
'cifar100_allcnnc': 'CIFAR 100',
'cifar10_3c3d': 'CIFAR 10',
'imdb_bilstm': 'IMDb LSTM',
'svhn_wrn164': 'WRN 16(4)',
'tolstoi_char_rnn': 'Char RNN'
}
pool = mp.Pool(16)
all_axes = []
all_figs = []
for prob_no, problem in enumerate(problems):
print(f"Doing {problem}")
problem_type = ['acc', 'loss'][('vae' in problem) or ('deep' in problem)]
root_path = './'
prob_dir = args.inpath
dir_path = []
for opt in args.optimizers:
path_to_jsons = find_json_parent(os.path.join(prob_dir, opt))
dir_path.append([path_to_jsons])
opt_labels = args.optimizer_labels or args.optimizers
if args.print_metrics:
hyperparam_perf = analyzer.compute_tunability_metrics(dir_path, opt_labels, obj="max" if problem_type =="acc" else "min", num_shuffle=num_shuffle, x_axis=x_axis)
print_tunability_to_latex(hyperparam_perf, included_metrics, score_type = problem_type)
else:
all_opts_all_logs = pool.map(temp, zip(dir_path, [num_shuffle] * len(dir_path)))
fig, ax = plot_box_hyperparam_optim(all_opts_all_logs, labels=opt_labels, x_type='num-evals', score_type = problem_type, y_limits=axis_limits[problem], do_best=False, plot_box=True, do_legend=False)
plt.suptitle(f'{plot_labels[problem]}', fontsize=15, fontweight='bold')
plt.savefig(args.outfile)
def find_json_parent(rootdir):
for subdir, dirs, files in os.walk(rootdir):
for d in dirs:
d = os.path.join(subdir, d)
for f in os.listdir(d):
is_correct_dir = f.endswith(".json")
if is_correct_dir:
return subdir
if __name__ == "__main__":
main()