在 Python 中部署 TensorFlow 模型

Question

在实时实施 Tensorflow 模型方面需要帮助。 当我在训练时，一切都运行良好，但是当我继续进行实时预测或预测时，我收到的输出不及格。 我不知道为什么会这样。 我使用了这里的代码参考：https://www.kaggle.com/raoulma/ny-stock-price-prediction-rnn-lstm-gru/notebook 并尝试使用相同的代码来实现或部署，只需少量更改。

见以下代码：

import numpy as np
import pandas as pd
import sklearn
import sklearn.preprocessing
import datetime
import os
import tensorflow as tf

df = pd.read_csv("Realtime_Values.csv", index_col = 0)
df.info()
def load_data(stock,seq_len):

    data_raw = stock.as_matrix() # convert to numpy array
    data = []

    for index in range(len(data_raw) - seq_len): 
        data.append(data_raw[index: index + seq_len])
    #print(len(data))
    data = np.array(data);

    x_forecast = data[:,:-1,:]
    return x_forecast

def normalize_data(df):
    cols = list(df.columns.values)
    min_max_scaler = sklearn.preprocessing.MinMaxScaler()
    df = pd.DataFrame(min_max_scaler.fit_transform(df.values))
    df.columns = cols
    return df
model_path ="modelsOHLC"
seq_len = 9
# parameters
n_steps = seq_len-1 
n_inputs = 4
n_neurons = 100 
n_outputs = 4
n_layers = 4
learning_rate = 0.01
batch_size = 10
n_epochs = 1000
tf.reset_default_graph()

X = tf.placeholder(tf.float32, [None, n_steps, n_inputs])
y = tf.placeholder(tf.float32, [None, n_outputs])
layers = [tf.contrib.rnn.BasicRNNCell(num_units=n_neurons, activation=tf.nn.elu)
          for layer in range(n_layers)]
multi_layer_cell = tf.contrib.rnn.MultiRNNCell(layers)
rnn_outputs, states = tf.nn.dynamic_rnn(multi_layer_cell, X, dtype=tf.float32)

stacked_rnn_outputs = tf.reshape(rnn_outputs, [-1, n_neurons]) 
stacked_outputs = tf.layers.dense(stacked_rnn_outputs, n_outputs)
outputs = tf.reshape(stacked_outputs, [-1, n_steps, n_outputs])
outputs = outputs[:,n_steps-1,:] # keep only last output of sequence

loss = tf.reduce_mean(tf.square(outputs - y)) # loss function = mean squared error 
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate) 
training_op = optimizer.minimize(loss)
saver = tf.train.Saver()
sess  =tf.Session()
sess.run(tf.global_variables_initializer())    
if(tf.train.checkpoint_exists(tf.train.latest_checkpoint(model_path))):
        saver.restore(sess, tf.train.latest_checkpoint(model_path))
df = normalize_data(df)
x_forecast = load_data(df,seq_len)
y_forecast_pred = sess.run(outputs, feed_dict={X: x_forecast})
print(y_forecast_pred)

谁能帮助我实时运行上述代码而不会出现任何问题？

Answer 1

程序训练模型时，代码可能找不到保存的权重；因此，预测是在未经训练的状态下生成的。您的训练模型代码是：

if (tf.train.checkpoint_exists(tf.train.latest_checkpoint(model_path))):
    saver.restore(sess, tf.train.latest_checkpoint(model_path))

解决这个问题：

• 添加调试代码如print("checkpoint exists!")

• 通过调试器在save.restore(...) 之前或之后放置断点以查找要从中恢复的检查点。

• 查看model_path 以确保您的检查点已正确保存。