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 。这是我的模型功能:
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(参见上文的追溯):logits 和标签必须具有相同的第一维,得到 logits 形状 [3,2] 和标签形状 [1]
filenames_array 是一个 16 个字符串的数组
["file1.png", "file2.png", "file3.png", ...]
labels_array 是一个 16 个整数的数组
[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
编辑
经过更多时间搜索,在我的模型函数的第一行
input_layer = tf.reshape(features, [-1, 40, 40, 1])
表示将动态计算 batch_size 维度的“-1”在此处具有值“3”。与我的错误中相同的“3”:logits 和 labels 必须具有相同的第一维,得到 logits 形状 [3,2] 和标签形状 [1]
如果我将值强制为“1”,我会遇到这个新错误:
reshape 的输入是一个有 4800 个值的张量,但请求的形状有 1600 个
也许我的功能有问题?
编辑2:
完整的代码在这里:https ://gist.github.com/geoffreyp/cc8e97aab1bff4d39e10001118c6322e
编辑3
我更新了要点
logits = tf.layers.dense(inputs=dropout, units=1)
https://gist.github.com/geoffreyp/cc8e97aab1bff4d39e10001118c6322e
但是我不完全理解您对批量大小的回答,这里的批量大小如何可以为 3 而我选择批量大小为 1 ?
如果我选择一个 batch_size = 3 我有这个错误: logits 和 labels 必须具有相同的第一维,得到 logits 形状 [9,1] 和标签形状 [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...