这个问题也许是通用的，但是我已经在网上搜索了很长一段时间，却没有以我能理解的方式找到它。
我改编了TensorFlow tutorial上的代码，内容如下：
import numpy as np
import tensorflow as tf
tf.logging.set_verbosity(tf.logging.INFO)
def cnn_model_fn(features, labels, mode):
'''Model function for CNN'''
#Input Layer
#Reshape X to 4-D tensor: [batch_size, width, height, channels]
input_layer = tf.reshape(features['x'],[-1, 28, 28, 1])
#Convolutional Layer
#Computes 3 features maps using 5x5 filter with sigmoid activation
#Input Tensor Shape: [batch_size, 28, 28, 1]
#Output Tensor Shape: [batch_size, 24, 24, 20]
conv = tf.layers.conv2d(
inputs = input_layer,
filters = 20,
kernel_size = [5,5],
activation = tf.nn.sigmoid)
#Pooling Layer
#Max pooling layer with a 2x2 flter and stride of 2
#Input Tensor Shape: [batch_size, 24, 24, 20]
#Output Tensor Shape: [batch_size, 12, 12, 20]
pool = tf.layers.max_pooling2d(
inputs = conv,
pool_size = [2, 2],
strides = 2)
#Flatten tensor into a batch of vectors
#Input Tensor Shape: [batch_size, 12, 12, 20]
#Output Tensor Shape: [batch_size, 12 * 12 * 20]
pool_flat = tf.reshape(pool, [-1, 12 * 12 * 20])
#Fully Connected Layer = dense layer
#
#Input Tensor Shape: [batch_size, 12 * 12 * 20]
#Output Tensor Shape: [batch_size, 100]
fully_cnted = tf.layers.dense(
inputs = pool_flat,
units = 100,
activation = tf.nn.sigmoid)
#Output Layer
#Softmax
#Input Tensor Shape: [batch_size, 100]
#Output Tensor Shape: [batch_size, 10]
logits = tf.layers.dense(
inputs = fully_cnted,
units = 10,
activation = tf.nn.softmax)
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)
#Configura the Training Op (for TRAIN mode)
if mode == tf.estimator.ModeKeys.TRAIN:
optimizer = tf.train.GradientDescentOptimizer(learning_rate = 0.1)
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 evaluations 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)
def main(unused_argv):
#Load training, validation and test data
mnist = tf.contrib.learn.datasets.load_dataset('mnist')
train_data = mnist.train.images #return a np.array
train_labels = np.asarray(mnist.train.labels, dtype = np.int32)
#val_data = mnist.test.images #returns a np.array
#val_labels = np.asarray(mnist.test.labels, dtype = np.int32)
test_data = mnist.test.images #returns a np.array
test_labels = np.asarray(mnist.test.labels, dtype = np.int32)
#Create the Estimator
mnist_classifier = tf.estimator.Estimator(
model_fn = cnn_model_fn, model_dir="/tmp/mnist_convnet_model")
#Set up logging for predictions
#Log the Values in the "softmax" tensor with label "probabilities
tensors_to_log = {"probabilities": "softmax_tensor"}
logging_hook = tf.train.LoggingTensorHook(
tensors=tensors_to_log, every_n_iter=50)
#Train the model
train_input_fn = tf.estimator.inputs.numpy_input_fn(
x={'x':train_data},
y=train_labels,
batch_size=10,
num_epochs=60,
shuffle=True)
mnist_classifier.train(
input_fn=train_input_fn,
hooks=[logging_hook])
#Evaluate the model and print results
eval_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x":test_data},
y=test_labels,
num_epochs=1,
shuffle=False)
eval_results = mnist_classifier.evaluate(input_fn=eval_input_fn)
print(eval_results)
if __name__ == "__main__":
tf.app.run()
使用这种结构，在训练结束时我的准确性为96％，并将其保存为几个model.cpkt文件。
如何使用此模型来预测输入like this，不一定是Android应用，而是像MNIST一样预测28 * 28的jpg之类的输入，我正在寻找任何类型的在这里提供帮助（教程，书籍...）。