代码之家 › 专栏 › 技术社区 › user1563247

带Tensorflow的MNIST,值错误

mnist tensorflow python

0

user1563247 · 技术社区 · 8 年前

我正在尝试使用Tensorflow遍历MNIST数据集。这是我的密码

import tensorflow as tf
import numpy as np

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

X_train = np.array(mnist.train.images, 'float')
X_test = np.array(mnist.test.images, 'float')
y_train = np.array(mnist.train.images, 'int32')
y_test = np.array(mnist.test.images, 'int32')

# Specify feature
feature_columns = [tf.contrib.layers.real_valued_column('', dimension=784)]

# Build 3 layer DNN with 10, 20, 10 units respectively.
classifier = tf.contrib.learn.DNNClassifier(feature_columns=feature_columns, 
                                            hidden_units=[200, 100, 60, 30], 
                                            n_classes=10,
                                            model_dir="./output"
                                            )

# Fit model.
classifier.fit(X_train, y_train, batch_size=100, steps=1000)

# Evaluate accuracy.
accuracy_score = classifier.evaluate(X_test, y_test)["accuracy"]

print('Accuracy: {0:f}'.format(accuracy_score))

然而,我不断发现错误:

ValueError:无法挤压尺寸[1],预期尺寸为1,获得784 对于“dnn/multi\u class\u head/softmax\u cross\u entropy\u loss/Squence”(操作: “挤压”),输入形状:[?,784]。

回溯让我注意到第31行,在那里我在分类器上调用fit(),但我不知道为什么。

1 回复 | 直到 8 年前

1

0

jrjames83 8 年前

这应该可以正常工作。版本 =“1.1.0”,并使用python 3.6。输入数据的维度可能存在一些问题,但请从这个角度进行反向操作。

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import numpy as np

mnist_data = "save/here/mnist"

MNIST_DATASET = input_data.read_data_sets(mnist_data)

train_data = np.array(MNIST_DATASET.train.images, 'float32')
train_target = np.array(MNIST_DATASET.train.labels, 'int64')

test_data = np.array(MNIST_DATASET.test.images, 'float32')
test_target = np.array(MNIST_DATASET.test.labels, 'int64')

feature_columns = [tf.contrib.layers.real_valued_column("", dimension=784)]

classifier = tf.contrib.learn.DNNClassifier(
    feature_columns=feature_columns
    ,n_classes=10
    ,hidden_units=[128, 32]
)

classifier.fit(train_data, train_target, steps=5)
accuracy_score = classifier.evaluate(test_data, test_target, steps=5)['accuracy']
print("accuracy: ", 100*accuracy_score,"%")

输出:

WARNING:tensorflow:Skipping summary for global_step, must be a float or np.float32.
accuracy:  44.3899989128 %