我是tensoflow的新手,我想用我自己的数据(40x40的图像)调整MNIST教程https://www.tensorflow.org/tutorials/layers . 这是我的模特功能:
def cnn_model_fn(features, labels, mode):
# Input Layer
input_layer = tf.reshape(features, [-1, 40, 40, 1])
# Convolutional Layer #1
conv1 = tf.layers.conv2d(
inputs=input_layer,
filters=32,
kernel_size=[5, 5],
# To specify that the output tensor should have the same width and height values as the input tensor
# value can be "same" ou "valid"
padding="same",
activation=tf.nn.relu)
# Pooling Layer #1
pool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2, 2], strides=2)
# Convolutional Layer #2 and Pooling Layer #2
conv2 = tf.layers.conv2d(
inputs=pool1,
filters=64,
kernel_size=[5, 5],
padding="same",
activation=tf.nn.relu)
pool2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=[2, 2], strides=2)
# Dense Layer
pool2_flat = tf.reshape(pool2, [-1, 10 * 10 * 64])
dense = tf.layers.dense(inputs=pool2_flat, units=1024, activation=tf.nn.relu)
dropout = tf.layers.dropout(
inputs=dense, rate=0.4, training=mode == tf.estimator.ModeKeys.TRAIN)
# Logits Layer
logits = tf.layers.dense(inputs=dropout, units=2)
predictions = {
# Generate predictions (for PREDICT and EVAL mode)
"classes": tf.argmax(input=logits, axis=1),
# Add `softmax_tensor` to the graph. It is used for PREDICT and by the
# `logging_hook`.
"probabilities": tf.nn.softmax(logits, name="softmax_tensor")
}
if mode == tf.estimator.ModeKeys.PREDICT:
return tf.estimator.EstimatorSpec(mode=mode, predictions=predictions)
# Calculate Loss (for both TRAIN and EVAL modes)
loss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits)
# Configure the Training Op (for TRAIN mode)
if mode == tf.estimator.ModeKeys.TRAIN:
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.001)
train_op = optimizer.minimize(
loss=loss,
global_step=tf.train.get_global_step())
return tf.estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op)
# Add evaluation metrics (for EVAL mode)
eval_metric_ops = {
"accuracy": tf.metrics.accuracy(
labels=labels, predictions=predictions["classes"])}
return tf.estimator.EstimatorSpec(
mode=mode, loss=loss, eval_metric_ops=eval_metric_ops)
我在标签和logits之间有一个形状大小错误:
InvalidArgumentError (see above for traceback): logits and labels must have the same first dimension, got logits shape [3,2] and labels shape [1]
filenames_array是一个16字符串的数组
["file1.png", "file2.png", "file3.png", ...]
和labels_array是一个16整数的数组
[0,0,1,1,0,1,0,0,0,...]
主要功能是:
# Create the Estimator
mnist_classifier = tf.estimator.Estimator(model_fn=cnn_model_fn, model_dir="/tmp/test_convnet_model")
# Train the model
cust_train_input_fn = lambda: train_input_fn_custom(
filenames_array=filenames, labels_array=labels, batch_size=1)
mnist_classifier.train(
input_fn=cust_train_input_fn,
steps=20000,
hooks=[logging_hook])
我试图重塑logits但没有成功:
logits = tf.reshape(logits,[1,2])
我需要你的帮助,谢谢
EDIT
经过更长时间的搜索,在我的模型功能的第一行
input_layer = tf.reshape(features, [-1, 40, 40, 1])
"-1"表示将动态计算batch_size维度,此处的值为"3" . 与我的错误相同"3": logits and labels must have the same first dimension, got logits shape [3,2] and labels shape [1]
如果我强制该值为“1”我有这个新错误:
Input to reshape is a tensor with 4800 values, but the requested shape has 1600
也许我的功能有问题?
EDIT2 :
完整的代码在这里:https://gist.github.com/geoffreyp/cc8e97aab1bff4d39e10001118c6322e
EDIT3
我更新了要点
logits = tf.layers.dense(inputs=dropout, units=1)
https://gist.github.com/geoffreyp/cc8e97aab1bff4d39e10001118c6322e
但我不完全理解你对批量大小的答案,这里的批量大小是3,而我选择1的批量大小?
如果我选择batch_size = 3我有这个错误: logits and labels must have the same first dimension, got logits shape [9,1] and labels shape [3]
我试图重塑标签:
labels = tf.reshape(labels, [3, 1])
我更新了功能和标签结构:
filenames_train = [['blackcorner-data/1.png', 'blackcorner-data/2.png', 'blackcorner-data/3.png',
'blackcorner-data/4.png', 'blackcorner-data/n1.png'],
['blackcorner-data/n2.png',
'blackcorner-data/n3.png', 'blackcorner-data/n4.png',
'blackcorner-data/11.png', 'blackcorner-data/21.png'],
['blackcorner-data/31.png',
'blackcorner-data/41.png', 'blackcorner-data/n11.png', 'blackcorner-data/n21.png',
'blackcorner-data/n31.png']
]
labels = [[0, 0, 0, 0, 1], [1, 1, 1, 0, 0], [0, 0, 1, 1, 1]]
但没有成功......
2 回答
我有一个类似的问题,结果是一个池图层没有正确重新整形 . 我错误地使用了我的情况
tf.reshape(pool, shape=[-1, 64 * 7 * 7])
而不是tf.reshape(pool, shape=[-1, 64 * 14 * 14])
,这导致了关于logits和标签的类似错误按摩 . 改变因素,例如tf.reshape(pool, shape=[-1, 64 * 12 * 12])
导致完全不同,误导性较差的错误消息 .也许这也是这种情况 . 我建议通过代码检查节点的形状,以防万一 .
您的logits形状看起来正确,批量大小为3,输出层大小为2,这是您定义的输出层 . 你的标签也应该是形状[3,2] . 批次3,每批有2 [1,0]或[0,1] .
还要注意,当你有一个布尔分类输出时,输出/ logits层上不应该有2个神经元 . 您只需输出一个取值为0或1的值,您可以看到[1,0]和[0,1]的2个输出是多余的,可以表示为简单的[0 | 1]值 . 当你这样做时,你往往会得到更好的结果 .
因此,您的logits应该最终为[3,1],并且您的标签应该是一个包含3个值的数组,每个值对应一批中的每个样本 .