-
Notifications
You must be signed in to change notification settings - Fork 8
/
Copy pathmnist1d.py
150 lines (133 loc) · 7.2 KB
/
mnist1d.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
import argparse
import os
import numpy as np
import matplotlib.pyplot as plt
from torchvision.datasets import MNIST
from redunet import Architecture, Lift1D, Fourier1D
import dataset
import evaluate
import plot
import functionals as F
import utils
# hyperparameters
parser = argparse.ArgumentParser()
parser.add_argument('--samples', type=int, default=10,
help="number of samples for initialization")
parser.add_argument('--time', type=int, default=150,
help="number of timesteps")
parser.add_argument('--channels', type=int, default=9,
help="number of channels")
parser.add_argument('--classes', nargs="+", type=int, default=list(range(10)),
help="Classes to Keep (Example: 0 1)")
parser.add_argument('--outchannels', type=int, default=5,
help="number of output channels for kernel")
parser.add_argument('--ksize', type=int, default=9,
help="kernel size")
parser.add_argument('--layers', type=int, default=10,
help="number of layers")
parser.add_argument('--eta', type=float, default=0.5,
help='learning rate')
parser.add_argument('--eps', type=float, default=0.1,
help='eps squared')
parser.add_argument('--lmbda', type=float, default=500,
help='lambda')
parser.add_argument('--tail', type=str, default='',
help='extra information to add to folder name')
parser.add_argument('--save_dir', type=str, default='./saved_models/',
help='base directory for saving PyTorch model. (default: ./saved_models/)')
parser.add_argument('--data_dir', type=str, default='./data/',
help='base directory for saving PyTorch model. (default: ./saved_models/)')
args = parser.parse_args()
model_dir = os.path.join(args.save_dir,
f"mnist1d-classes{''.join(map(str, args.classes))}",
f"time{args.time}"
f"channels{args.channels}"
f"samples{args.samples}"
f"_layers{args.layers}"
f"_outchannels{args.outchannels}"
f"_ksize{args.ksize}"
f"_eps{args.eps}"
f"_eta{args.eta}"
f"{args.tail}")
os.makedirs(model_dir, exist_ok=True)
utils.save_params(model_dir, vars(args))
print(model_dir)
# data loading
X_train, y_train, X_test, y_test = dataset.load_MNIST(args.data_dir, expand_dims=False)
X_train, y_train = F.filter_class(X_train, y_train, args.classes, args.samples)
X_train, y_train = F.shuffle(X_train, y_train)
X_train = F.convert2polar(X_train, args.channels, args.time)
X_test, y_test = F.filter_class(X_test, y_test, args.classes, args.samples)
X_test = F.convert2polar(X_test, args.channels, args.time)
X_translate_train, y_translate_train = F.translate1d(*F.get_n_each(X_train, y_train, 10), n=2, stride=4)
X_translate_test, y_translate_test = F.translate1d(*F.get_n_each(X_test, y_test, 10), n=2, stride=4)
utils.save_features(model_dir, "X_train", X_train, y_train)
utils.save_features(model_dir, "X_test", X_test, y_test)
utils.save_features(model_dir, "X_translate_train", X_translate_train, y_translate_train)
utils.save_features(model_dir, "X_translate_test", X_translate_test, y_translate_test)
# setup architecture
kernels = F.generate_kernel('gaussian', (args.outchannels, args.channels, args.ksize))
layers = [Lift1D(kernels)] + [Fourier1D(args.layers, eta=args.eta, eps=args.eps, lmbda=args.lmbda)]
model = Architecture(layers, model_dir, len(args.classes))
# train/test pass
print("Forward pass - train features")
Z_train = model(X_train, y_train).real
utils.save_loss(model.loss_dict, model_dir, "train")
print("Forward pass - test features")
Z_test = model(X_test).real
utils.save_loss(model.loss_dict, model_dir, "test")
print("Forward pass - translated train features")
Z_translate_train = model(X_translate_train).real
utils.save_loss(model.loss_dict, model_dir, "translate_train")
print("Forward pass - translated test features")
Z_translate_test = model(X_translate_test).real
utils.save_loss(model.loss_dict, model_dir, "translate_test")
# save features
utils.save_features(model_dir, "Z_train", Z_train, y_train)
utils.save_features(model_dir, "Z_test", Z_test, y_test)
utils.save_features(model_dir, "Z_translate_train", Z_translate_train, y_translate_train)
utils.save_features(model_dir, "Z_translate_test", Z_translate_test, y_translate_test)
np.save(os.path.join(model_dir, 'features', 'kernel.npy'), kernels)
# normalize
X_train = F.normalize(X_train.reshape(X_train.shape[0], -1))
X_test = F.normalize(X_test.reshape(X_test.shape[0], -1))
X_translate_train = F.normalize(X_translate_train.reshape(X_translate_train.shape[0], -1))
X_translate_test = F.normalize(X_translate_test.reshape(X_translate_test.shape[0], -1))
Z_train = F.normalize(Z_train.reshape(Z_train.shape[0], -1))
Z_test = F.normalize(Z_test.reshape(Z_test.shape[0], -1))
Z_translate_train = F.normalize(Z_translate_train.reshape(Z_translate_train.shape[0], -1))
Z_translate_test = F.normalize(Z_translate_test.reshape(Z_translate_test.shape[0], -1))
# plot
plot.plot_combined_loss(model_dir)
plot.plot_heatmap(X_train, y_train, "X_train", model_dir)
plot.plot_heatmap(X_test, y_test, "X_test", model_dir)
plot.plot_heatmap(X_translate_train, y_translate_train, "X_translate_train", model_dir)
plot.plot_heatmap(X_translate_test, y_translate_test, "X_translate_test", model_dir)
plot.plot_heatmap(Z_train, y_train, "Z_train", model_dir)
plot.plot_heatmap(Z_test, y_test, "Z_test", model_dir)
plot.plot_heatmap(Z_translate_train, y_translate_train, "Z_translate_train", model_dir)
plot.plot_heatmap(Z_translate_test, y_translate_test, "Z_translate_test", model_dir)
# evaluation train
_, acc_svm = evaluate.svm(Z_train, y_train, Z_train, y_train)
acc_knn = evaluate.knn(Z_train, y_train, Z_train, y_train, k=5)
acc_svd = evaluate.nearsub(Z_train, y_train, Z_train, y_train, n_comp=10)
acc = {"svm": acc_svm, "knn": acc_knn, "nearsub-svd": acc_svd}
utils.save_params(model_dir, acc, name="acc_train.json")
# evaluation test
_, acc_svm = evaluate.svm(Z_train, y_train, Z_test, y_test)
acc_knn = evaluate.knn(Z_train, y_train, Z_test, y_test, k=5)
acc_svd = evaluate.nearsub(Z_train, y_train, Z_test, y_test, n_comp=10)
acc = {"svm": acc_svm, "knn": acc_knn, "nearsub-svd": acc_svd}
utils.save_params(model_dir, acc, name="acc_test.json")
# evaluation translate train
_, acc_svm = evaluate.svm(Z_train, y_train, Z_translate_train, y_translate_train)
acc_knn = evaluate.knn(Z_train, y_train, Z_translate_train, y_translate_train, k=5)
acc_svd = evaluate.nearsub(Z_train, y_train, Z_translate_train, y_translate_train, n_comp=10)
acc = {"svm": acc_svm, "knn": acc_knn, "nearsub-svd": acc_svd}
utils.save_params(model_dir, acc, name="acc_translate_train.json")
# evaluation translate
_, acc_svm = evaluate.svm(Z_train, y_train, Z_translate_test, y_translate_test)
acc_knn = evaluate.knn(Z_train, y_train, Z_translate_test, y_translate_test, k=5)
acc_svd = evaluate.nearsub(Z_train, y_train, Z_translate_test, y_translate_test, n_comp=10)
acc = {"svm": acc_svm, "knn": acc_knn, "nearsub-svd": acc_svd}
utils.save_params(model_dir, acc, name="acc_translate_test.json")