-
Notifications
You must be signed in to change notification settings - Fork 5
/
03_plot_tune.py
176 lines (151 loc) · 5.79 KB
/
03_plot_tune.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
"""
Plot the results of FNN tuning as in Figures 5, 11, 17.
"""
# Author: Arturs Berzins <berzins@cats.rwth-aachen.de>
# License: BSD 3 clause
import config
import utils
import os
import numpy as np
import pandas as pd
from ray import tune
import matplotlib
from matplotlib import pyplot, colors
from matplotlib.ticker import FixedLocator
import warnings
model_key = 'FNN'
preamble = (
r'\usepackage{amsmath}'
r'\usepackage{amssymb}'
r'\newcommand{\vekt}[1]{\mbox{$\boldsymbol{#1}$}}'
)
matplotlib.rcParams.update({
'pgf.texsystem': "pdflatex",
'font.family': 'serif',
'text.latex.preamble': preamble,
'pgf.preamble': preamble,
'text.usetex': True,
'pgf.rcfonts': False,
'font.size': 8,
'legend.fontsize': 8,
'axes.labelsize': 8,
'axes.titlesize': 8,
'xtick.labelsize': 8,
'ytick.labelsize': 8
})
df_all = {}
for component in config.components:
df_all[component] = pd.DataFrame()
# https://ray.readthedocs.io/en/latest/tune-package-ref.html#ray.tune.Analysis
# https://ray.readthedocs.io/en/latest/tune-usage.html#analyzing-results
path_analysis = os.path.join(utils.model_dir, model_key, component)
n = sum(f.endswith('.json') for f in os.listdir(path_analysis))
if n>1:
warnings.warn(F'{path_analysis} contains {n} experimental runs.'+
'Results might not plot correctly, so you might want to remove previous runs.')
analysis = tune.Analysis(path_analysis)
df_temp = analysis.dataframe()
df_all[component] = pd.concat([df_all[component],df_temp], ignore_index=True, sort=False)
### CONTOUR SENSITIVITY
fig, axes = pyplot.subplots(1, len(config.components), sharex=True, sharey=True)
fig.suptitle('')
legends = []
caxes = [None]*len(axes)
w=6.3
h=2.3
fig.set_size_inches(w=w, h=h)
fig.subplots_adjust(bottom=0, top=1, left=0, right=1, wspace=0)
for idx_ax, component in enumerate(config.components):
df = df_all[component]
lrs = df['config/lr']
nhs = df['config/hidden_size']
ls = df['mean_loss']
x = np.unique(lrs)
y = np.unique(nhs)
X, Y = np.meshgrid(x,y)
Z = np.zeros(X.shape)
for ix, vx in enumerate(x):
for iy, vy in enumerate(y):
# BEWARE: this can conflict with other experiments in the same folder
z = df.loc[ (df['config/lr']==vx) &
(df['config/hidden_size']==vy)
]['mean_loss'].values[0] ** 0.5
Z[iy,ix] = z
Z = np.clip(Z, None, 1)
vmin = np.min(Z)
vmax = np.max(Z)
lev_exp = np.linspace( np.log10(Z.min()),
np.log10(Z.max()),
20)
levs = np.power(10, lev_exp)
cf = axes[idx_ax].contourf(X, Y, Z, levs, norm=colors.LogNorm(), cmap='viridis')
axes[idx_ax].set_xlabel(r'$\alpha_0$')
axes[idx_ax].set_title(r'${}$'.format([r'\vekt{u}','p','T'][idx_ax]))
axes[idx_ax].set_xscale('log')
# mark best location
idx = df['mean_loss'].idxmin()
lr_min = df.loc[idx]["config/lr"]
nh_min = df.loc[idx]["config/hidden_size"]
axes[idx_ax].scatter([lr_min], [nh_min], c='red',marker='+', linewidth=0.5)
# force limits of axes because marker might create whitespace
axes[idx_ax].set_xlim([lrs.min(), lrs.max()])
axes[idx_ax].set_ylim([nhs.min(), nhs.max()])
#axes[idx_ax].set_box_aspect(1)
pos_ax = axes[idx_ax].get_position()
#print(pos_ax)
#print([pos_ax.width,pos_ax.height])
margin_top = 0.1
margin_bottom = 0.35
scaler = (1-margin_top-margin_bottom)
new_w = pos_ax.height*h/w*scaler
new_h = pos_ax.height*scaler
c=0.5
wspace = c*new_w
L = len(axes)
left = (1-new_w*(L+(L-1)*c))/2
new_x0 = left + idx_ax*(wspace+new_w)
new_y0 = margin_bottom
pos_ax = [new_x0, new_y0, new_w, new_h]
axes[idx_ax].set_position(pos_ax)
#divider = make_axes_locatable(axes[idx_ax])
#caxes[idx_ax] = divider.append_axes('bottom', size='10%', pad=0)
caxes[idx_ax] = pyplot.axes([new_x0, 0.05, new_w, 0.1*new_h])
#cax.set_title(r'$\varepsilon_{FNN}$', size='medium', position='bottom')
bar = fig.colorbar(cf, cax=caxes[idx_ax], orientation='horizontal')
exp_min = np.log10(Z.min())
exp_max = np.log10(Z.max())
# MAJOR
exps_major = np.arange( np.floor(exp_min), np.ceil(exp_max) +1)
levs_major = 10**exps_major
# linearly interpolate exponent: exp_min-exp_max -> 0-1
#locs_major = (exps_major - exp_min) / (exp_max-exp_min)
caxes[idx_ax].xaxis.set_major_locator(FixedLocator(levs_major))
# NOTE: FixedLocator in old matplotlib versions expected locations in terms of axis
# newer versions seem to use values per default, so we don't have to compute levels
# MINOR
levs_minor = []
# Add minor tick levels ..
# .. below first major level
for f in range(2,9+1):
levs_minor.append(levs_major[0]*0.1*f)
# .. above each major level
for lev_major in levs_major:
for f in range(2,9+1):
levs_minor.append(lev_major*f)
caxes[idx_ax].xaxis.set_minor_locator(FixedLocator(levs_minor))
m, e = '{:.2e}'.format(Z.min()).split('e')
label_min = r'${}{{\times}}10^{{{}}}$'.format(m,int(e))
if Z.max() == 1:
label_max = r'$\ge 1$'
else:
m, e = '{:.2e}'.format(Z.max()).split('e')
label_max = r'${}{{\times}}10^{{{}}}$'.format(m,int(e))
caxes[idx_ax].annotate(label_min, xy=(0, 1.2), xytext=(0, 0),
xycoords='axes fraction', textcoords='offset points',
ha='left', va='bottom')
caxes[idx_ax].annotate(label_max, xy=(1, 1.2), xytext=(0, 0),
xycoords='axes fraction', textcoords='offset points',
ha='right', va='bottom')
# va basline or top seems to work best
axes[0].set_ylabel(r'$N_\nu$')
pyplot.show()