TensorFlowでデータ型のエラーを解消したい。

＜1/26追記＞

最下部の内容から進展があったので、記事を更新します。
問題があるであろう、グラフ作成の部分と実行部分を抜粋します（いくつか変更してます）。
実行部ではfor文でエポック数を増やしたいのですが、2周目からエラーが出るようです。実行部のfor文をfor i in range(1)とした場合はエラーは生じません。
エラー内容は「sess.runにndarrayを入れるな。テンソルを入れてくれ」ということだと思いますが、何故、forの1周目でテンソルだったものが2周目からndarrayに変化してしまうのかがわかりません。

python
1グラフの作成
2g = tf.Graph()
3with g.as_default():
4  tf.set_random_seed(123)
5  
6  tf_x = tf.placeholder(tf.float32, shape=(None, 4), name='tf_x')
7  tf_y = tf.placeholder(tf.int32, shape=(None), name='tf_y')
8  
9  oh_y = tf.one_hot(tf_y, 3, dtype=tf.float32, name='oh_y')
10
11  w = tf.Variable(tf.random_normal((4, 3)), name='weight')
12  b = tf.Variable(tf.zeros(3), name='bias')
13  
14  logits = tf.add(tf.matmul(tf_x, w), b, name='logits')
15  
16  prediction = {'probabilities': tf.nn.softmax(logits, name='probabilities'),
17                'labels': tf.argmax(logits,1)}
18  
19  cost = tf.losses.softmax_cross_entropy(logits=logits, onehot_labels=oh_y)
20#   cost = -tf.reduce_sum(oh_y * tf.log(prediction['probabilities']))
21  optimizer = tf.train.AdamOptimizer()
22  train = optimizer.minimize(cost)
23  correct_predictions = tf.equal(prediction['labels'], tf.argmax(oh_y, 1))
24#   tf.reduce_meanの中身はfloatにしてあげないと小数点の計算できない
25  accuracy = tf.reduce_mean(tf.cast(correct_predictions, tf.float32), name='accuracy')
26  init = tf.global_variables_initializer()
27
28実行
29with tf.Session(graph=g) as sess:
30  sess.run(init)
31# ↓のfor文のrange(1)にすると、エラーが生じません。
32  for step in range(2):
33    _, cost= sess.run([train, cost], feed_dict={tf_x: X_train, tf_y: y_train})
34    if (step + 1) % 10 == 0:
35      print('Epoch%2d Cost:%.2f' %(step+1, cost))
36  
37  print('Prediction Accuracy: %.2f' %(sess.run(accuracy, feed_dict={tf_x: X_test, tf_y: y_test}) * 100))
38

生じているエラー

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in __init__(self, fetches, contraction_fn)
    299         self._unique_fetches.append(ops.get_default_graph().as_graph_element(
--> 300             fetch, allow_tensor=True, allow_operation=True))
    301       except TypeError as e:

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py in as_graph_element(self, obj, allow_tensor, allow_operation)
   3489     with self._lock:
-> 3490       return self._as_graph_element_locked(obj, allow_tensor, allow_operation)
   3491 

/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/ops.py in _as_graph_element_locked(self, obj, allow_tensor, allow_operation)
   3578       raise TypeError("Can not convert a %s into a %s." % (type(obj).__name__,
-> 3579                                                            types_str))
   3580 

TypeError: Can not convert a float32 into a Tensor or Operation.

During handling of the above exception, another exception occurred:

TypeError                                 Traceback (most recent call last)
<ipython-input-134-171c87a28859> in <module>()
      3 # ↓のfor文のrange(1)にすると、エラーが生じません。
      4   for step in range(2):
----> 5     _, cost= sess.run([train, cost], feed_dict={tf_x: X_train, tf_y: y_train})
      6     if (step + 1) % 10 == 0:
      7       print('Epoch%2d Cost:%.2f' %(step+1, cost))

/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in run(self, fetches, feed_dict, options, run_metadata)
    927     try:
    928       result = self._run(None, fetches, feed_dict, options_ptr,
--> 929                          run_metadata_ptr)
    930       if run_metadata:
    931         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in _run(self, handle, fetches, feed_dict, options, run_metadata)
   1135     # Create a fetch handler to take care of the structure of fetches.
   1136     fetch_handler = _FetchHandler(
-> 1137         self._graph, fetches, feed_dict_tensor, feed_handles=feed_handles)
   1138 
   1139     # Run request and get response.

/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in __init__(self, graph, fetches, feeds, feed_handles)
    469     """
    470     with graph.as_default():
--> 471       self._fetch_mapper = _FetchMapper.for_fetch(fetches)
    472     self._fetches = []
    473     self._targets = []

/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in for_fetch(fetch)
    259     elif isinstance(fetch, (list, tuple)):
    260       # NOTE(touts): This is also the code path for namedtuples.
--> 261       return _ListFetchMapper(fetch)
    262     elif isinstance(fetch, collections.Mapping):
    263       return _DictFetchMapper(fetch)

/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in __init__(self, fetches)
    368     """
    369     self._fetch_type = type(fetches)
--> 370     self._mappers = [_FetchMapper.for_fetch(fetch) for fetch in fetches]
    371     self._unique_fetches, self._value_indices = _uniquify_fetches(self._mappers)
    372 

/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in <listcomp>(.0)
    368     """
    369     self._fetch_type = type(fetches)
--> 370     self._mappers = [_FetchMapper.for_fetch(fetch) for fetch in fetches]
    371     self._unique_fetches, self._value_indices = _uniquify_fetches(self._mappers)
    372 

/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in for_fetch(fetch)
    269         if isinstance(fetch, tensor_type):
    270           fetches, contraction_fn = fetch_fn(fetch)
--> 271           return _ElementFetchMapper(fetches, contraction_fn)
    272     # Did not find anything.
    273     raise TypeError('Fetch argument %r has invalid type %r' % (fetch,

/usr/local/lib/python3.6/dist-packages/tensorflow/python/client/session.py in __init__(self, fetches, contraction_fn)
    302         raise TypeError('Fetch argument %r has invalid type %r, '
    303                         'must be a string or Tensor. (%s)' %
--> 304                         (fetch, type(fetch), str(e)))
    305       except ValueError as e:
    306         raise ValueError('Fetch argument %r cannot be interpreted as a '

TypeError: Fetch argument 5.690171 has invalid type <class 'numpy.float32'>, must be a string or Tensor. (Can not convert a float32 into a Tensor or Operation.)

＜以下、更新前の原文＞
GitHubの記事を参考にTensorFlowでアヤメの分類問題をやっているのですが、恐らくデータ型のエラーがどうしても解消できません。
コスト関数の最適化のとこでエラーが出ていますが、上の記事と見合わせても、データ型的にどこが間違えているのかがわかりません。
解決の方法がわかる方、是非ご教授くださると助かります。
試行錯誤して色々書き換えると様々なエラーが生じるので、もしかしたら多くの個所がおかしいのかもしれません。

書いたコード

python
1from urllib.request import urlretrieve
2import pandas as pd
3import numpy as np
4import tensorflow as tf
5from sklearn.model_selection import train_test_split
6
7# データの呼び出し
8url = '上記記事のコードのＵＲＬ'
9urlretrieve(url, 'iris.csv')
10df = pd.read_csv('iris.csv', encoding='utf-8')
11
12# データの準備
13names = sorted(set(df.Name.values))
14name2num = {w:i for i, w in enumerate(names)}
15df['label'] = df['Name'].map(name2num)
16X = df.iloc[:,:4].values
17y = df.label.values
18X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, stratify=y, random_state=0)
19
20
21# グラフの作成
22g = tf.Graph()
23with g.as_default():
24  tf.set_random_seed(123)
25  
26  tf_x = tf.placeholder(tf.float32, shape=(None, 4), name='tf_x')
27  tf_y = tf.placeholder(tf.int32, shape=(None), name='tf_y')
28  
29  oh_y = tf.one_hot(tf_y, 3, dtype=tf.float32, name='oh_y')
30  
31  w = tf.Variable(tf.random_normal((4, 3)), name='weight')
32  b = tf.Variable(tf.zeros(3), name='bias')
33  
34  logits = tf.add(tf.matmul(tf_x, w), b, name='logits')
35  
36  prediction = {'probabilities': tf.nn.softmax(logits, name='probabilities'),
37                'labels': tf.cast(tf.argmax(logits,1), tf.int32)}
38  
39#   cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(logits=logits, labels=oh_y), name='cost')
40  cost = -tf.reduce_sum(oh_y * tf.log(prediction['probabilities']))
41  optimizer = tf.train.AdamOptimizer()
42  train = optimizer.minimize(cost)
43  correct_predictions = tf.equal(prediction['labels'], tf.cast(tf.argmax(oh_y, 1), tf.int32))
44  accuracy = tf.reduce_mean(tf.cast(correct_predictions, tf.float32), name='accuracy')  
45  init = tf.global_variables_initializer()
46  
47  
48# 実行  
49with tf.Session(graph=g) as sess:
50  sess.run(init)
51
52  for step in range(300):
53    _, cost, accuracy = sess.run([train, cost, accuracy], feed_dict={tf_x: X_train, tf_y: y_train})
54    if (step + 1) % 10 == 0:
55      print('Epoch%2d Cost:%.2f Accuracy:%.2f%%' %(step+1, cost, accuracy*100))
56  
57  print('Prediction Accuracy: %.2f' %(sess.run(accuracy, feed_dict={tf_x: X_test, tf_y: y_test}) * 100))

生じているエラー（追記時に字数が足りなくなったので削りました）

---------------------------------------------------------------------------
TypeError: Fetch argument 682.8205 has invalid type <class 'numpy.float32'>, must be a string or Tensor. (Can not convert a float32 into a Tensor or Operation.)

環境など

Windows 10
Python 3.6.5

行動規範の内容に同意します

回答1件

ベストアンサー

コードを見た限りだと、変数costの前が重複しているので、発生しているエラーだと思います。
1週目は、tensorのcostですが、````_, cost = ....で変数costが上書きされ、numpy.float32型のndarrayで出力されてしまい、2周目は、sess.run(...)```を計算できないと考えられます。

  cost = tf.losses.softmax_cross_entropy(logits=logits, onehot_labels=oh_y)
  .....
実行
with tf.Session(graph=g) as sess:
  sess.run(init)
# ↓のfor文のrange(1)にすると、エラーが生じません。
  for step in range(2):
    ### ここで変数が上書きされてしまっている。
    _, cost= sess.run([train, cost], feed_dict={tf_x: X_train, tf_y: y_train})
    if (step + 1) % 10 == 0:
      print('Epoch%2d Cost:%.2f' %(step+1, cost))

そのため、for文で回す変数costをcost_valなどに変更するとエラーは消えると思います。

投稿2019/01/31 02:31