-
Notifications
You must be signed in to change notification settings - Fork 31
/
run_cla_toy.py
223 lines (196 loc) · 8.08 KB
/
run_cla_toy.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
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
# import matplotlib
# matplotlib.use('Agg')
import numpy as np
import matplotlib as mpl
# mpl.use('pgf')
def figsize(scale):
# Get this from LaTeX using \the\textwidth
fig_width_pt = 397.4849
inches_per_pt = 1.0 / 72.27 # Convert pt to inch
# Aesthetic ratio (you could change this)
golden_mean = (np.sqrt(5.0) - 1.0) / 6
fig_width = fig_width_pt * inches_per_pt * scale # width in inches
fig_height = fig_width * golden_mean # height in inches
fig_size = [fig_width, fig_height]
return fig_size
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",
# blank entries should cause plots to inherit fonts from the document
"font.serif": [],
"font.sans-serif": [],
"font.monospace": [],
"axes.labelsize": 7, # LaTeX default is 10pt font.
"font.size": 7,
"legend.fontsize": 6, # Make the legend/label fonts a little smaller
"xtick.labelsize": 6,
"ytick.labelsize": 6,
"figure.figsize": figsize(0.9), # default fig size of 0.9 textwidth
"pgf.preamble": [
# use utf8 fonts becasue your computer can handle it :)
r"\usepackage[utf8x]{inputenc}",
# plots will be generated using this preamble
r"\usepackage[T1]{fontenc}",
# plots will be generated using this preamble
r"\usepackage{amsmath}",
]
}
mpl.rcParams.update(pgf_with_latex)
grey = '#808080'
mpl.rcParams['axes.linewidth'] = 0.3
mpl.rcParams['axes.edgecolor'] = grey
mpl.rcParams['xtick.color'] = grey
mpl.rcParams['ytick.color'] = grey
mpl.rcParams['axes.labelcolor'] = "black"
# These are the "Tableau 20" colors as RGB.
tableau20 = [(31, 119, 180), (174, 199, 232), (255, 127, 14), (255, 187, 120),
(44, 160, 44), (152, 223, 138), (214, 39, 40), (255, 152, 150),
(148, 103, 189), (197, 176, 213), (140, 86, 75), (196, 156, 148),
(227, 119, 194), (247, 182, 210), (127, 127, 127), (199, 199, 199),
(188, 189, 34), (219, 219, 141), (23, 190, 207), (158, 218, 229)]
# Scale the RGB values to the [0, 1] range, which is the format matplotlib
# accepts.
for i in range(len(tableau20)):
r, g, b = tableau20[i]
tableau20[i] = (r / 255., g / 255., b / 255.)
import gpflow
import osvgpc
import matplotlib.pyplot as plt
def init_Z(cur_Z, new_X, use_old_Z=True, first_batch=True):
if use_old_Z:
Z = np.copy(cur_Z)
else:
M = cur_Z.shape[0]
M_old = int(0.8 * M)
M_new = M - M_old
old_Z = cur_Z[np.random.permutation(M)[0:M_old], :]
new_Z = new_X[np.random.permutation(new_X.shape[0])[0:M_new], :]
Z = np.vstack((old_Z, new_Z))
return Z
def get_data(iid):
X = np.loadtxt('../data/banana_train_x.txt', delimiter=',')
y = np.loadtxt('../data/banana_train_y.txt', delimiter=',')
y = y.reshape((y.shape[0], 1))
y[y==-1] = 0
Xtest = np.loadtxt('../data/banana_test_x.txt', delimiter=',')
ytest = np.loadtxt('../data/banana_test_y.txt', delimiter=',')
ytest = ytest.reshape((ytest.shape[0], 1))
ytest[ytest==-1] = 0
if not iid:
# todo sort
idxs = np.argsort(X[:, 0])
X = X[idxs, :]
y = y[idxs, :]
return X, y, Xtest, ytest
def gridParams():
mins = [-3.2, -2.5]
maxs = [3.2, 2.5]
nGrid = 50
xspaced = np.linspace(mins[0], maxs[0], nGrid)
yspaced = np.linspace(mins[1], maxs[1], nGrid)
xx, yy = np.meshgrid(xspaced, yspaced)
Xplot = np.vstack((xx.flatten(), yy.flatten())).T
return mins, maxs, xx, yy, Xplot
def plot_model(ax, m, cur_x, cur_y, seen_x=None, seen_y=None, test_x=None, test_y=None):
col1 = '#0172B2'
col2 = '#CC6600'
mins, maxs, xx, yy, Xplot = gridParams()
# p = m.predict_y(Xplot)[0]
mf, vf = m.predict_f(Xplot)
mf = mf.numpy()
vf = vf.numpy()
ax.plot(
cur_x[:, 0][cur_y[:, 0] == 1],
cur_x[:, 1][cur_y[:, 0] == 1],
'o', color=col1, mew=0, alpha=0.6)
ax.plot(
cur_x[:, 0][cur_y[:, 0] == 0],
cur_x[:, 1][cur_y[:, 0] == 0],
'o', color=col2, mew=0, alpha=0.6)
if seen_x is not None:
ax.plot(
seen_x[:, 0][seen_y[:, 0] == 1],
seen_x[:, 1][seen_y[:, 0] == 1],
'o', color=col1, mew=0, alpha=0.05)
ax.plot(
seen_x[:, 0][seen_y[:, 0] == 0],
seen_x[:, 1][seen_y[:, 0] == 0],
'o', color=col2, mew=0, alpha=0.05)
if hasattr(m, 'inducing_variable'):
Z = m.inducing_variable.Z
ax.plot(Z[:, 0], Z[:, 1], 'ko', mew=0, ms=3, alpha=0.8)
ax.contour(xx, yy, mf.reshape(*xx.shape),
[0], colors='k', linewidths=1.4, zorder=100)
# plt.contour(xx, yy, p.reshape(*xx.shape), [0.5],
# colors='k', linewidths=1.8, zorder=100)
if test_x is not None:
mf, _ = m.predict_f(test_x)
mf = mf.numpy()
pred_y = 1.0 * (mf > 0)
err = np.sum(np.abs(pred_y - test_y)) / mf.shape[0]
ax.set_title('error=%.2f'%err)
def run_vfe(no_batches, M, use_old_Z, iid):
X, y, Xtest, ytest = get_data(iid)
N = X.shape[0]
mb_size = int(np.floor(N / no_batches))
fig, axs = plt.subplots(1, no_batches+1, figsize=figsize(1), sharey=True)
maxiter = 2000
for i in range(no_batches):
Xi = X[i * mb_size:(i + 1) * mb_size, :]
yi = y[i * mb_size:(i + 1) * mb_size, :]
if i == 0:
Z1 = Xi[np.random.permutation(Xi.shape[0])[0:M], :]
model = gpflow.models.SVGP(gpflow.kernels.RBF(lengthscales=np.ones(2)),
gpflow.likelihoods.Bernoulli(), Z1)
gpflow.optimizers.Scipy().minimize(
model.training_loss_closure((Xi, yi)), model.trainable_variables,
options=dict(disp=1, maxiter=maxiter))
else:
Zinit = init_Z(Zopt, Xi, use_old_Z)
model = osvgpc.OSVGPC((Xi, yi), gpflow.kernels.RBF(lengthscales=np.ones(2)),
gpflow.likelihoods.Bernoulli(),
mu, Su, Kaa, Zopt, Zinit)
gpflow.optimizers.Scipy().minimize(
model.training_loss, model.trainable_variables,
options=dict(disp=1, maxiter=maxiter))
Zopt = model.inducing_variable.Z.numpy()
mu, Su = model.predict_f(Zopt, full_cov=True)
if len(Su.shape) == 3:
Su = Su[0, :, :] + 1e-4 * np.eye(mu.shape[0])
Kaa = model.kernel(model.inducing_variable.Z)
if i == 0:
seen_x = None
seen_y = None
else:
seen_x = X[:i * mb_size, :]
seen_y = y[:i * mb_size, :]
plot_model(axs[i], model, Xi, yi, seen_x, seen_y, Xtest, ytest)
# run sparse GP
Z = X[np.random.permutation(X.shape[0])[0:M], :]
model = gpflow.models.SVGP(gpflow.kernels.RBF(lengthscales=np.ones(2)),
gpflow.likelihoods.Bernoulli(), Z)
gpflow.optimizers.Scipy().minimize(
model.training_loss_closure((X, y)), model.trainable_variables,
options=dict(disp=1, maxiter=maxiter))
plot_model(axs[-1], model, X, y, None, None, Xtest, ytest)
for i in range(no_batches+1):
axs[i].locator_params(nbins=5, axis='y')
axs[i].locator_params(nbins=5, axis='x')
axs[i].tick_params('both', length=3, width=0.5, which='minor')
axs[i].tick_params('both', length=3, width=0.5, which='major')
axs[i].tick_params('both', length=3, width=0.5, which='minor')
axs[i].tick_params('both', length=3, width=0.5, which='major')
axs[i].set_xlabel(r'$x_1$')
axs[0].set_ylabel(r'$x_2$')
plt.subplots_adjust(wspace=0.01, hspace=0.01)
fig.savefig('../tmp/cla_VFE_M_%d_iid_%r.png' % (M, iid), bbox_inches='tight', pad_inches=0)
if __name__ == '__main__':
use_old_Z = False
no_batches = 3
np.random.seed(42)
M = 30
iid = True
run_vfe(no_batches, M, use_old_Z, iid)
iid = False
run_vfe(no_batches, M, use_old_Z, iid)