mnist for ML Beginnersについて

https://qiita.com/kkk3H/items/c3eb0d868170b29b0b87

こちらの記事を読ませていただいて自分も同じことをしてみたいと思ってやってみたのですが、うまく行きませんでした。
最初の
data = data_set.read_data_sets(one_hot=True)
でエラーが出てしまっていると思うのですがなぜエラーが出てしまうのかわかりません。
よろしくお願いします。

Jupyter Notebook
Python 3.6.4
TensorFlow 1.3.0
NumPy 1.13.3

python
1import tensorflow as tf
2import dataset
3
4data = data_set.read_data_sets(one_hot=True)
5
6x = tf.placeholder(tf.float32, [None, 784])
7y_ = tf.placeholder(tf.float32, [None, 3])
8
9W = tf.Variable(tf.zeros([784, 3]))
10b = tf.Variable(tf.zeros([3]))
11
12y = tf.nn.softmax(tf.matmul(x, W) + b)
13
14cross_entropy = -tf.reduce_sum(y_ * tf.log(y))
15
16train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)
17
18init = tf.initialize_all_variables()
19sess = tf.Session()
20sess.run(init)
21
22for i in range(100):
23    batch_xs, batch_ys = data.train.next_batch(10)
24    sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
25
26correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
27accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
28
29sess.run(accuracy, feed_dict={x: data.test.images, y_: data.test.labels})
30

python
1"data_set.py"
2import numpy
3from tensorflow.python.framework import dtypes
4import collections
5
6Datasets = collections.namedtuple('datasets', ['train', 'validation', 'test'])
7
8
9def dense_to_one_hot(labels_dense, num_classes):
10    num_labels = labels_dense.shape[0]
11    index_offset = numpy.arange(num_labels) * num_classes
12    labels_one_hot = numpy.zeros((num_labels, num_classes))
13    labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1
14
15    return labels_one_hot
16
17
18class DataSet(object):
19
20    def __init__(self,
21                 images,
22                 labels):
23
24        self._images = images
25        self._labels = labels
26        self._num_examples = images.shape[0]
27        self._epochs_completed = 0
28        self._index_in_epoch = 0
29
30        self._epoch_test = 0
31        self._index_in_epoch_test = 0
32
33    @property
34    def images(self):
35        return self._images
36
37    @property
38    def labels(self):
39        return self._labels
40
41    @property
42    def num_examples(self):
43        return self._num_examples
44
45    @property
46    def epochs_completed(self):
47        return self._epochs_completed
48
49    @property
50    def epoch_test(self):
51        return self._epoch_test
52
53    def next_batch(self, batch_size, shuffle=True):
54        start = self._index_in_epoch
55        # Shuffle for the first epoch
56        if self._epochs_completed == 0 and start == 0 and shuffle:
57            perm0 = numpy.arange(self._num_examples)
58            numpy.random.shuffle(perm0)
59            self._images = self.images[perm0]
60            self._labels = self.labels[perm0]
61        # Go the the next epoch
62        if start + batch_size > self._num_examples:
63            # Finished epoch
64            self._epochs_completed += 1
65            # Get the rest examples in this epoch
66            rest_num_examples = self._num_examples - start
67            data_rest_part = self._images[start:self._num_examples]
68            label_rest_part = self._labels[start:self._num_examples]
69            # Shuffle the data
70            if shuffle:
71                perm = numpy.arange(self._num_examples)
72                numpy.random.shuffle(perm)
73                self._images = self.images[perm]
74                self._labels = self.labels[perm]
75            # Start next epoch
76            start = 0
77            self._index_in_epoch = batch_size - rest_num_examples
78            end = self._index_in_epoch
79            data_new_part = self._images[start:end]
80            label_new_part = self._labels[start:end]
81
82            return numpy.concatenate((data_rest_part, data_new_part), axis = 0), numpy.concatenate((label_rest_part, label_new_part), axis = 0)
83        else:
84            self._index_in_epoch += batch_size
85            end = self._index_in_epoch
86            return self._images[start:end], self._labels[start:end]
87
88def read_data_sets(one_hot=False,
89                   dtype=dtypes.float32,
90                   reshape=True,
91                   validation_size=1000,
92                   test_size=500):
93    """Construct a data set from a given directory path of the data"""
94
95    images = numpy.load('./imgs.npy')
96    labels = numpy.load('./labels.npy')
97
98    if reshape:
99        images = numpy.reshape(images, (-1, numpy.prod(images.shape[1:])))
100    if one_hot:
101        labels = dense_to_one_hot(labels, 3)
102
103    validation_images = images[:validation_size]
104    validation_labels = labels[:validation_size]
105    test_images = images[validation_size:validation_size+test_size]
106    test_labels = labels[validation_size:validation_size+test_size]
107    train_images = images[validation_size+test_size:]
108    train_labels = labels[validation_size+test_size:]
109
110    train = DataSet(train_images,
111                    train_labels)
112
113    validation = DataSet(validation_images,
114                         validation_labels)
115
116    test = DataSet(test_images,
117                   test_labels)
118
119    return Datasets(train=train, validation=validation, test=test)
120
121

以下エラー文です

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
~/anaconda3/envs/tf/lib/python3.6/site-packages/numpy/core/fromnumeric.py in _wrapfunc(obj, method, *args, **kwds)
     56     try:
---> 57         return getattr(obj, method)(*args, **kwds)
     58 

TypeError: 'numpy.float64' object cannot be interpreted as an integer

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
<ipython-input-23-9ceed1aa7dd7> in <module>()
      2 import dataset
      3 
----> 4 data = data_set.read_data_sets(one_hot=True)

~/lab/data_set/data_set.py in read_data_sets(one_hot, dtype, reshape, validation_size, test_size)
     96 
     97     if reshape:
---> 98         images = numpy.reshape(images, (-1, numpy.prod(images.shape[1:])))
     99     if one_hot:
    100         labels = dense_to_one_hot(labels, 3)

~/anaconda3/envs/tf/lib/python3.6/site-packages/numpy/core/fromnumeric.py in reshape(a, newshape, order)
    230            [5, 6]])
    231     """
--> 232     return _wrapfunc(a, 'reshape', newshape, order=order)
    233 
    234 

~/anaconda3/envs/tf/lib/python3.6/site-packages/numpy/core/fromnumeric.py in _wrapfunc(obj, method, *args, **kwds)
     65     # a downstream library like 'pandas'.
     66     except (AttributeError, TypeError):
---> 67         return _wrapit(obj, method, *args, **kwds)
     68 
     69 

~/anaconda3/envs/tf/lib/python3.6/site-packages/numpy/core/fromnumeric.py in _wrapit(obj, method, *args, **kwds)
     45     except AttributeError:
     46         wrap = None
---> 47     result = getattr(asarray(obj), method)(*args, **kwds)
     48     if wrap:
     49         if not isinstance(result, mu.ndarray):

TypeError: 'numpy.float64' object cannot be interpreted as an integer

行動規範の内容に同意します

回答2件

ベストアンサー

Python
1images = numpy.reshape(images, (-1, numpy.prod(images.shape[1:])))

Errorを見るとこの部分がおかしいんだと思います。
内容としては、numpy.reshapeの引数にはint型しか受け付けないようになっているのに、numpy.float64型が入っているからでしょう。images.shape[1:]の中身が分からないので何とも言えませんが、prodの計算でimage[1:]の中にfloat型が一つでも入っていたら返り値はnumpy.float64になるようなので、データの中身を洗ってみてはいかがでしょう。

投稿2018/01/22 01:44