Question

我创建并训练了一个神经网络，但我希望能够输入测试点并查看其结果（而不是使用eval函数）。

模型运行良好，成本降低了每个纪元，但我只想在末尾添加一条线来传递一些输入坐标并让它告诉我预测的变换坐标。

import tensorflow as tf
import numpy as np

def coordinate_transform(size, angle):
    input = np.random.rand(size, 2)
    output = np.zeros((size, 2))
    noise = 0.05*(np.add(np.random.rand(size) * 2, -1))
    theta = np.add(np.add(np.arctan(input[:,1] / input[:,0]) , angle) , noise)
    radii = np.sqrt(np.square(input[:,0]) + np.square(input[:,1]))
    output[:,0] = np.multiply(radii, np.cos(theta))
    output[:,1] = np.multiply(radii, np.sin(theta))
    return input, output

#Data
input, output = coordinate_transform(2000, np.pi/2)
train_in = input[:1000]
train_out = output[:1000]
test_in = input[1000:]
test_out = output[1000:]

# Parameters
learning_rate = 0.001
training_epochs = 15
batch_size = 1
display_step = 1

# Network Parameters
n_hidden_1 = 100 # 1st layer number of features
n_input = 2 # [x,y]
n_classes = 2 # output x,y coords

# tf Graph input
x = tf.placeholder("float", [1,n_input])
y = tf.placeholder("float", [1, n_input])

# Create model
def multilayer_perceptron(x, weights, biases):
    # Hidden layer with RELU activation
    layer_1 = tf.add(tf.matmul(x, weights['h1']), biases['b1'])
    layer_1 = tf.nn.relu(layer_1)
    # Output layer with linear activation
    out_layer = tf.matmul(layer_1, weights['out']) + biases['out']
    return out_layer

# Store layers weight & bias
weights = {
    'h1': tf.Variable(tf.random_normal([n_input, n_hidden_1])),
    'out': tf.Variable(tf.random_normal([n_hidden_1, n_classes]))
}
biases = {
    'b1': tf.Variable(tf.random_normal([n_hidden_1])),
    'out': tf.Variable(tf.random_normal([n_classes]))
}

# Construct model
pred = multilayer_perceptron(x, weights, biases)

# Define loss and optimizer
#cost = tf.losses.mean_squared_error(0, (tf.slice(pred, 0, 1) - x)**2 + (tf.slice(pred, 1, 1) - y)**2)
cost = tf.losses.mean_squared_error(y, pred)
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
optimizer = optimizer.minimize(cost)

# Initializing the variables
#init = tf.global_variables_initializer()
init = tf.initialize_all_variables()

# Launch the graph
with tf.Session() as sess:
    sess.run(init)

    # Training cycle
    for epoch in range(training_epochs):
        avg_cost = 0.
        total_batch = 1000#int(len(train_in)/batch_size)
        # Loop over all batches
        for i in range(total_batch):
            batch_x = train_in[i].reshape((1,2))
            batch_y = train_out[i].reshape((1,2))

            #print(batch_x.shape)
            #print(batch_y.shape)
            #print(batch_y, batch_x)
            # Run optimization op (backprop) and cost op (to get loss value)
            _, c = sess.run([optimizer, cost], feed_dict={x: batch_x,
                                                          y: batch_y})
            # Compute average loss
            avg_cost += c / total_batch
        # Display logs per epoch step
        if epoch % display_step == 0:
            print ("Epoch:", '%04d' % (epoch+1), "cost=", \
                "{:.9f}".format(avg_cost))
    print("Optimization Finished!")

    # Test model
    correct_prediction = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))
    # Calculate accuracy
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))

    #Make predictions

Answer 1

'pred'操作是你的实际结果（因为它在计算损失时用于与y比较），所以类似下面的内容应该可以解决这个问题：

print(sess.run([pred], feed_dict={x: _INPUT_GOES_HERE_ })

显然 _INPUT_GOES_HERE _ 需要用实际输入代替。

Answer 2

您还可以使用tensorflow.python.saved_model libs以TensorFlow服务可以提供的格式保存您的模型。

TensorFlow服务最近变得更容易安装和设置：

https://github.com/tensorflow/serving/blob/master/tensorflow_serving/g3doc/setup.md#installing-using-apt-get

下面是一些示例代码（您需要调整用例的提要/输入和提取/输出）。

为您的模型创建SignatureDef：

from tensorflow.python.saved_model import utils
from tensorflow.python.saved_model import signature_constants
from tensorflow.python.saved_model import signature_def_utils

graph = tf.get_default_graph()

x_observed = graph.get_tensor_by_name('x_observed:0')
y_pred = graph.get_tensor_by_name('add:0')

tensor_info_x_observed = utils.build_tensor_info(x_observed)
print(tensor_info_x_observed)

tensor_info_y_pred = utils.build_tensor_info(y_pred)
print(tensor_info_y_pred)

prediction_signature = signature_def_utils.build_signature_def(inputs = 
                {'x_observed': tensor_info_x_observed}, 
                outputs = {'y_pred': tensor_info_y_pred}, 
                method_name = signature_constants.PREDICT_METHOD_NAME)

使用SaveModelBuilder使用上面定义的SignatureDef资产保存模型：

from tensorflow.python.saved_model import builder as saved_model_builder
from tensorflow.python.saved_model import tag_constants

unoptimized_saved_model_path = '/root/models/linear_unoptimized/cpu/%s' % version
print(unoptimized_saved_model_path)

builder = saved_model_builder.SavedModelBuilder(unoptimized_saved_model_path)
builder.add_meta_graph_and_variables(sess, 
                                     [tag_constants.SERVING],
                                     signature_def_map={'predict':prediction_signature,                                     
signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:prediction_signature}, 
                                     clear_devices=True,
)

builder.save(as_text=False)

此处引用的github和docker repos中的更多详细信息：http://pipeline.ai

如何使用训练有素的Tensorflow模型进行预测

2 个答案: