質問編集履歴
1
作成コードとエラー内容の記入
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -1,3 +1,171 @@
|
|
|
1
1
|
現在、DeepLearningの学習をしています。ゼロから作るDeep Learningを参考に学習しています。しかし、モデルの評価やデータセット内の1画像を分類する例は載っているのですが、肝心の実際に自分が解析したい画像をどのように処理して、どのようにモデルに読み込ませるのかという実践的な部分がありませんでした。また、ネットで調べてみても私の力では見つけることが出来ませんでした。
|
|
2
2
|
|
|
3
3
|
作成した学習モデルを実際に使ってみるというような内容のことが書かれている本やサイト等はありませんか？回答よろしくお願いします。
|
|
4
|
+
|
|
5
|
+
|
|
6
|
+
|
|
7
|
+
作成したコードを載せておきます。
|
|
8
|
+
|
|
9
|
+
```python3
|
|
10
|
+
|
|
11
|
+
from sklearn.datasets import fetch_mldata
|
|
12
|
+
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
mnist = fetch_mldata('MNIST original', data_home=".")
|
|
16
|
+
|
|
17
|
+
X = mnist.data / 255
|
|
18
|
+
|
|
19
|
+
y = mnist.target
|
|
20
|
+
|
|
21
|
+
import matplotlib.pyplot as plt
|
|
22
|
+
|
|
23
|
+
|
|
24
|
+
|
|
25
|
+
|
|
26
|
+
|
|
27
|
+
plt.imshow(X[0].reshape(28, 28), cmap='gray')
|
|
28
|
+
|
|
29
|
+
print("This is {:.0f}".format(y[0]))
|
|
30
|
+
|
|
31
|
+
import torch
|
|
32
|
+
|
|
33
|
+
from torch.utils.data import TensorDataset, DataLoader
|
|
34
|
+
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
from sklearn.model_selection import train_test_split
|
|
38
|
+
|
|
39
|
+
X_train, X_test, y_train, y_test = train_test_split(
|
|
40
|
+
|
|
41
|
+
X, y, test_size=1/7, random_state=0)
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
X_train = torch.Tensor(X_train)
|
|
46
|
+
|
|
47
|
+
X_test = torch.Tensor(X_test)
|
|
48
|
+
|
|
49
|
+
y_train = torch.LongTensor(y_train)
|
|
50
|
+
|
|
51
|
+
y_test = torch.LongTensor(y_test)
|
|
52
|
+
|
|
53
|
+
|
|
54
|
+
|
|
55
|
+
ds_train = TensorDataset(X_train, y_train)
|
|
56
|
+
|
|
57
|
+
ds_test = TensorDataset(X_test, y_test)
|
|
58
|
+
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
loader_train = DataLoader(ds_train, batch_size=64, shuffle=True)
|
|
62
|
+
|
|
63
|
+
loader_test = DataLoader(ds_test, batch_size=64, shuffle=False)
|
|
64
|
+
|
|
65
|
+
|
|
66
|
+
|
|
67
|
+
from torch import nn
|
|
68
|
+
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
model = nn.Sequential()
|
|
72
|
+
|
|
73
|
+
model.add_module('fc1', nn.Linear(28*28, 100))
|
|
74
|
+
|
|
75
|
+
model.add_module('relu1', nn.ReLU())
|
|
76
|
+
|
|
77
|
+
model.add_module('fc2', nn.Linear(100, 100))
|
|
78
|
+
|
|
79
|
+
model.add_module('relu2', nn.ReLU())
|
|
80
|
+
|
|
81
|
+
model.add_module('fc3', nn.Linear(100, 10))
|
|
82
|
+
|
|
83
|
+
|
|
84
|
+
|
|
85
|
+
print(model)
|
|
86
|
+
|
|
87
|
+
from torch import optim
|
|
88
|
+
|
|
89
|
+
|
|
90
|
+
|
|
91
|
+
loss_fn = nn.CrossEntropyLoss()
|
|
92
|
+
|
|
93
|
+
|
|
94
|
+
|
|
95
|
+
optimizer = optim.Adam(model.parameters(), lr=0.01)
|
|
96
|
+
|
|
97
|
+
|
|
98
|
+
|
|
99
|
+
from torch.autograd import Variable
|
|
100
|
+
|
|
101
|
+
|
|
102
|
+
|
|
103
|
+
def train(epoch):
|
|
104
|
+
|
|
105
|
+
model.train()
|
|
106
|
+
|
|
107
|
+
for data, target in loader_train:
|
|
108
|
+
|
|
109
|
+
data, target = Variable(data), Variable(target)
|
|
110
|
+
|
|
111
|
+
optimizer.zero_grad()
|
|
112
|
+
|
|
113
|
+
output = model(data)
|
|
114
|
+
|
|
115
|
+
loss = loss_fn(output, target)
|
|
116
|
+
|
|
117
|
+
loss.backward()
|
|
118
|
+
|
|
119
|
+
optimizer.step()
|
|
120
|
+
|
|
121
|
+
print("epoch{}:end\n".format(epoch))
|
|
122
|
+
|
|
123
|
+
|
|
124
|
+
|
|
125
|
+
def test():
|
|
126
|
+
|
|
127
|
+
model.eval()
|
|
128
|
+
|
|
129
|
+
correct = 0
|
|
130
|
+
|
|
131
|
+
for data, target in loader_test:
|
|
132
|
+
|
|
133
|
+
data, target = Variable(data), Variable(target)
|
|
134
|
+
|
|
135
|
+
output = model(data)
|
|
136
|
+
|
|
137
|
+
pred = output.data.max(1, keepdim=True)[1]
|
|
138
|
+
|
|
139
|
+
correct += pred.eq(target.data.view_as(pred)).sum()
|
|
140
|
+
|
|
141
|
+
data_num = len(loader_test.dataset)
|
|
142
|
+
|
|
143
|
+
print('\n answer: {}/{} ({:.0f}%)\n'.format(correct,data_num, 100. * correct / data_num))
|
|
144
|
+
|
|
145
|
+
|
|
146
|
+
|
|
147
|
+
for epoch in range(3):
|
|
148
|
+
|
|
149
|
+
train(epoch)
|
|
150
|
+
|
|
151
|
+
|
|
152
|
+
|
|
153
|
+
test()
|
|
154
|
+
|
|
155
|
+
import tensorflow as tf
|
|
156
|
+
|
|
157
|
+
img_r = tf.read_file("8.png")
|
|
158
|
+
|
|
159
|
+
read_image = tf.image.decode_image(img_r, channels=3)
|
|
160
|
+
|
|
161
|
+
data = Variable(read_image)
|
|
162
|
+
|
|
163
|
+
output = model(data)
|
|
164
|
+
|
|
165
|
+
pred = output.data.max(0, keepdim=True)[1]
|
|
166
|
+
|
|
167
|
+
print("I think {}".format(pred))
|
|
168
|
+
|
|
169
|
+
```
|
|
170
|
+
|
|
171
|
+
error:variable data has o be a tensor, but got Tensor
|