-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathplot_drifting.py
178 lines (152 loc) · 8.11 KB
/
plot_drifting.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
import torch
from data.GP_data_sampler import GPCurvesReader
from data.NIFTY_data_sampler import NIFTYReader
from module.utils import to_numpy, normalize
from module.CNP import ConditionalNeuralProcess as CNP
from module.NP import NeuralProcess as NP
from module.convCNP import ConvCNP, UNet
from module.NP_PROV import ConvCNP as NP_PROV, UNet as PROV_UNet
import matplotlib.pyplot as plt
class Comparison:
def __init__(self, kernel, device):
self.kernel = kernel
self.device = device
self.load_model()
def load_model(self):
kernel = self.kernel
device = self.device
cnp = CNP(input_dim=1, latent_dim=128, output_dim=1).to(device)
# cnp.load_state_dict(torch.load('saved_model/' + kernel + '_CNP.pt', map_location=device))
self.cnp = cnp
# load module parameters
MODELNAME = 'NP'
np = NP(input_dim=1, latent_dim=128, output_dim=1, use_attention=MODELNAME == 'ANP').to(device)
# np.load_state_dict(torch.load('saved_model/' + kernel + '_' + MODELNAME + '.pt' , map_location=device))
self.np = np
# load module parameters
MODELNAME = 'ANP'
anp = NP(input_dim=1, latent_dim=128, output_dim=1, use_attention=MODELNAME == 'ANP').to(device)
# anp.load_state_dict(torch.load('saved_model/' + kernel + '_' + MODELNAME + '.pt', map_location=device))
self.anp = anp
convcnp = ConvCNP(rho=UNet(), points_per_unit=64 if kernel != 'NIFTY50' else 32, device=device).to(device)
convcnp.load_state_dict(torch.load('saved_model/' + kernel + '_ConvCNP.pt', map_location=device))
self.convcnp = convcnp
npprov = NP_PROV(rho=PROV_UNet(), points_per_unit=64 if kernel != 'NIFTY50' else 32, device=device).to(device)
npprov.load_state_dict(torch.load('saved_model/' + kernel + '_NP_PROV.pt', map_location=device))
self.npprov = npprov
self.modellist = [self.cnp, self.np, self.anp, self.convcnp, self.npprov]
self.colorlist = ['limegreen','royalblue','orange', 'purple', 'green' ]
self.namelist = ['CNP', 'NP', 'ANP', 'ConvCNP', 'SCANP']
# self.cololist = ['purple', 'limegreen', 'chocolate', 'cornflowerblue', 'deeppink']
def plot_sample(self, x_context, y_context, x_target, y_target, xlim=(-2, 2), len_legend=2, scale=1):
fig = plt.figure(figsize=(30, 20))
ax = plt.subplot()
# Plot context set and model predictions.
plt.scatter(to_numpy(x_context), to_numpy(y_context), label='Context', color='red', s= 500, zorder = 2)
plt.scatter(to_numpy(x_target), to_numpy(y_target), label='Target', marker='P', c='k', s = 500, zorder = 2)
y_context_norm, mu, std = normalize(y_context)
y_target_norm, _, _ = normalize(y_target, mu, std)
# plt.scatter(to_numpy(x_context), to_numpy(y_context_norm), label='Normalized context', marker='P', color='olive', s=500, alpha=0.7)
# plt.scatter(to_numpy(x_target), to_numpy(y_target_norm), label='Normalized target', c='m',alpha=0.7, s=500)
# plt.vlines([-2, 2], -3, 3, linestyles='dashed')
# plt.ylim(-3., 13)
# plt.xlim(xlim[0], xlim[1])
x_all = torch.linspace(xlim[0], xlim[-1], 200)[None, :, None]
for i, model in enumerate(self.modellist):
name = self.namelist[i]
if name not in ['ConvCNP', 'SCANP']:
continue
color = self.colorlist[i]
with torch.no_grad():
# Make predictions with model.
if name not in ['NP', 'ANP']:
y_mean, y_std = model(x_context.to(device), y_context_norm.to(device), x_all.to(device))
else:
(y_mean, y_std), _, _ = model(x_context.to(device), y_context_norm.to(device), x_all.to(device))
# Plot model predictions.
# plt.plot(to_numpy(x_all.squeeze()), to_numpy(y_mean.squeeze()), linestyle = '--', label='Model: normed %s' % name,
# color= 'm' if name == 'ConvCNP' else 'olive', linewidth =7, zorder =1)
# plt.fill_between(to_numpy(x_all.squeeze()),
# to_numpy(y_mean.squeeze() + 2 * y_std.squeeze()),
# to_numpy(y_mean.squeeze() - 2 * y_std.squeeze()),
# color= 'm' if name == 'ConvCNP' else 'olive', alpha=0.1)
#
y_all_mean = y_mean*std + mu
y_all_std = y_std *std
plt.plot(to_numpy(x_all.squeeze()), to_numpy(y_all_mean.squeeze()), label='Model: %s' % name, color=color, linewidth =7,
zorder = 1)
plt.fill_between(to_numpy(x_all.squeeze()),
to_numpy(y_all_mean.squeeze() + 2 * y_all_std.squeeze()),
to_numpy(y_all_mean.squeeze() - 2 * y_all_std.squeeze()),
color=color, alpha=0.2)
# plt.axis('off')
# plt.legend()
plt.xlabel("Location", size=60)
plt.ylabel("Function value", size=60)
plt.legend(ncol=len_legend, prop={'size': 50})
plt.grid("on", linewidth=3)
ax.tick_params(axis="x", labelsize=50)
ax.tick_params(axis="y", labelsize=50)
plt.savefig("saved_fig/"+self.kernel+"_"+str(scale)+".png")
plt.show()
def save_data(data):
import pandas as pd
import numpy as np
(x_context, y_context), x_target = data.query
x_context = to_numpy(x_context[0])
y_context = to_numpy(y_context[0])
x_target = to_numpy(x_target[0])
y_target = to_numpy(data.y_target[0])
context = pd.DataFrame(np.concatenate([x_context, y_context], axis=1))
target = pd.DataFrame(np.concatenate([x_target, y_target], axis=1))
context.to_csv("context.csv", index=False, header=False)
target.to_csv("target.csv", index=False, header=False)
print("saving succeed!")
def load_data():
import pandas as pd
import numpy as np
import collections
NPRegressionDescription = collections.namedtuple(
"NPRegressionDescription",
("query", "y_target", "num_total_points", "num_context_points"))
scale = 7
context = pd.read_csv("context.csv", header=None).values
target = pd.read_csv("target.csv", header=None).values
x_context = torch.from_numpy(context[:,0]).float()[None, :, None].to(device)
y_context = torch.from_numpy(context[:,1]*scale).float()[None, :, None].to(device)
x_target = torch.from_numpy(target[:,0]).float()[None, :, None].to(device)
y_target = torch.from_numpy(target[:,1]*scale).float()[None, :, None].to(device)
query = ((x_context, y_context), x_target)
data = NPRegressionDescription(query=query,
y_target=y_target,
num_total_points= 100,
num_context_points=50)
return data, scale
def main_GP():
kernel = 'EQ' # EQ or period or NIFTY50
models = Comparison(kernel, device)
# dataset = GPCurvesReader(kernel, batch_size=1, max_num_context=50, device=device)
# data = dataset.generate_curves(include_context=False)
# save_data(data)
data, scale = load_data()
(x_context, y_context), x_target = data.query
models.plot_sample(x_context, y_context, x_target, data.y_target, scale=scale)
def main_realworld():
kernel = 'NIFTY50' # EQ or period or NIFTY50
models = Comparison(kernel, device)
dataset = NIFTYReader(batch_size=1, max_num_context=50, device=device)
train_loader = dataset.train_dataloader()
val_loader = dataset.val_dataloader()
test_loader = dataset.test_dataloader()
for i, data in enumerate(test_loader): # 50 stocks per epoch, 1 batch is enough
(x_context, y_context), x_target = data.query
x_min = min(torch.min(x_context).cpu().numpy(),
torch.min(x_target).cpu().numpy(), 0.) - 0.1
x_max = max(torch.max(x_context).cpu().numpy(),
torch.max(x_target).cpu().numpy(), 2.) + 0.1
models.plot_sample(x_context.to(device), y_context.to(device), x_target.to(device), data.y_target.to(device), xlim=(x_min, x_max),
scale=4)
if __name__ == '__main__':
device = torch.device('cuda:7' if torch.cuda.is_available() else 'cpu')
main_GP()
# main_realworld()