2. 程式人生 > >Tensorflow的變數和模型儲存以及模型應用

Tensorflow的變數和模型儲存以及模型應用

阿新 • • 發佈：2019-01-09

Table of Contents

一、模型部分(成功)

1.儲存的模型

import tensorflow as tf
import numpy as np
# To plot pretty figures
%matplotlib inline
import matplotlib
import matplotlib.pyplot as plt
plt.rcParams['axes.labelsize'] = 14
plt.rcParams['xtick.labelsize'] = 12
plt.rcParams['ytick.labelsize'] =
 
 12

t_min, t_max = 0, 30
resolution = 0.1

def time_series(t):
    return t * np.sin(t) / 3 + 2 * np.sin(t*5)

def next_batch(batch_size, n_steps):
    t0 = np.random.rand(batch_size, 1) * (t_max - t_min - n_steps * resolution)
    Ts = t0 + np.arange(0., n_steps + 1) * resolution
    ys = time_series(Ts)
    return ys[:, :-1].reshape(-1, n_steps, 1), ys[:, 1:].reshape(-1, n_steps, 1)

t = np.linspace(t_min, t_max, int((t_max - t_min) / resolution))

n_steps = 20
t_instance = np.linspace(12.2, 12.2 + resolution * (n_steps + 1), n_steps + 1)

n_steps = 20
n_inputs =1
n_neurons = 100

X = tf.placeholder(tf.float32, [None, n_steps, n_inputs],name='X')
y = tf.placeholder(tf.float32, [None, n_steps, n_outputs],name='y')

cell = tf.contrib.rnn.BasicRNNCell(num_units=n_neurons, activation=tf.nn.relu )
rnn_outputs, states = tf.nn.dynamic_rnn(cell=cell, inputs=X, dtype=tf.float32)

n_outputs = 1
learning_rate = 0.001

stacked_rnn_outputs = tf.reshape(tensor=rnn_outputs, shape=[-1, n_neurons])
stacked_outputs = tf.layers.dense(inputs=stacked_rnn_outputs, units=n_outputs)
outputs = tf.reshape(tensor=stacked_outputs, shape=[-1, n_steps, n_outputs],name='outputs')

loss = tf.reduce_mean(tf.square(outputs - y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
training_op = optimizer.minimize(loss)

init = tf.global_variables_initializer()
saver = tf.train.Saver()

n_iterations = 1500
batch_size = 50

with tf.Session() as sess:
    init.run()
    for iteration in range(n_iterations):
        X_batch, y_batch = next_batch(batch_size, n_steps)
        sess.run(training_op, feed_dict={X:X_batch, y:y_batch})
        if iteration %100 ==0:
            mse = loss.eval(feed_dict={X:X_batch, y:y_batch})
            print(iteration, "\tMSE:", mse)
            
    X_new = time_series(np.array(t_instance[:-1].reshape(-1, n_steps, n_inputs)))
    y_pred = sess.run(outputs, feed_dict={X: X_new})
    
    saver.save(sess, "./my_time_series_model")

2.載入模型並用於預測

import tensorflow as tf
import numpy as np

def reset_graph(seed=42):
    tf.reset_default_graph()
    tf.set_random_seed(seed)
    np.random.seed(seed)
reset_graph()

1.載入圖結構和引數

sess = tf.Session()
saver = tf.train.import_meta_graph('./my_time_series_model.meta')
saver.restore(sess, tf.train.latest_checkpoint('./'))


INFO:tensorflow:Restoring parameters from ./my_time_series_model

2.獲取圖

graph = tf.get_default_graph()

3.獲取tensor

X = graph.get_tensor_by_name('X:0')
y = graph.get_tensor_by_name("y:0")
outputs = graph.get_tensor_by_name("outputs:0")

4.新的input準備

X_new = time_series(np.array(t_instance[:-1].reshape(-1, 20, 1)))

y.shape

TensorShape([Dimension(None), Dimension(20), Dimension(1)])

5.應用與預測

y_pred = sess.run(outputs, feed_dict={X: X_new})

print(y_pred)

[[[-3.44828582]
  [-2.48405623]
  [-1.13649726]
  [ 0.71962416]
  [ 2.01745081]
  [ 3.13937259]
  [ 3.54828739]
  [ 3.36234236]
  [ 2.77184248]
  [ 2.10781217]
  [ 1.64527285]
  [ 1.5579648 ]
  [ 1.87219918]
  [ 2.7233479 ]
  [ 3.85228252]
  [ 5.06193066]
  [ 6.07513857]
  [ 6.63054752]
  [ 6.59069633]
  [ 5.9993453 ]]]

plt.title("Testing the model", fontsize=14)
plt.plot(t_instance[:-1], time_series(t_instance[:-1]), "bo", markersize=10, label="instance")
plt.plot(t_instance[1:], time_series(t_instance[1:]), "w*", markersize=10, label="target")
plt.plot(t_instance[1:], y_pred[0,:,0], "r.", markersize=10, label="prediction")
plt.legend(loc="upper left")
plt.xlabel("Time")

# save_fig("time_series_pred_plot")
plt.show()

在這裡插入圖片描述

n_steps=20
sequence1 = [0.]*20
for iteration in range(len(t) - n_steps):
    X_batch = np.array(sequence1[-n_steps:]).reshape(1, 20, 1)
    y_red = sess.run(outputs, feed_dict={X:X_batch})
    sequence1.append(y_pred[0,-1,0])

sequence2 = [time_series(i*resolution+t_min+(t_max-t_min/3)) for i in range(20)]
for iteration in range(len(t) - n_steps):
    X_batch = np.array(sequence2[-n_steps:]).reshape(1, n_steps, 1)
    y_pred = sess.run(outputs, feed_dict={X:X_batch})
    sequence2.append(y_pred[0,-1,0])


sequence2

[-11.310069100186947,
 -10.293466647143219,
 -9.0351304228581899,
 -7.7791526184875721,
 -6.7463786174128,
 -6.0810586339718817,
 -5.8164786703312812,
 -5.8679774217158975,
 -6.0550484641549618,
 -6.1471037973674534,
 -5.9216796338876341,
 -5.2208171081907668,
 -3.991795656706794,
 -2.3022262397260698,
 -0.32578778096384303,
 1.6979234284416602,
 3.5196245383171041,
 4.938914081326276,
 5.8508622581105945,
 6.2692415768875769,
 6.2414274,
 6.0063138,
 5.7208457,
 5.6004872,
 5.7733955,
 6.2796745,
 7.0539942,
 7.9423361,
 8.7371979,
 9.2194023,
 9.2210941,
 8.6577063,
 7.55442,
 6.0555,
 4.37221,
 2.7345319,
 1.3436384,
 0.32472184,
 -0.3159388,
 -0.6737811,
 -0.92999107,
 -1.3043951,
 -1.9887228,
 -3.10217,
 -4.6800175,
 -6.6327167,
 -8.7743168,
 -10.866323,
 -12.663467,
 -13.970881,
 -14.69599,
 -14.907141,
 -14.78536,
 -14.409622,
 -14.090124,
 -14.002066,
 -14.257333,
 -15.155252,
 -16.459118,
 -18.0779,
 -19.749088,
 -21.18313,
 -22.134678,
 -22.438381,
 -22.068893,
 -21.052273,
 -19.576031,
 -18.089518,
 -16.549288,
 -15.274453,
 -14.365784,
 -13.75396,
 -13.394633,
 -13.047565,
 -12.466007,
 -11.425198,
 -9.8138237,
 -7.6150599,
 -4.9504156,
 -2.0713785,
 0.75190848,
 3.2330799,
 5.1569152,
 6.4215879,
 7.0548859,
 7.2106261,
 7.115716,
 7.0375557,
 7.2040081,
 7.7478371,
 8.6662283,
 9.8395824,
 11.034622,
 11.977497,
 12.413702,
 12.166514,
 11.191109,
 9.5760918,
 7.5540438,
 5.3992167,
 3.4150782,
 1.812041,
 0.70379823,
 0.057163272,
 -0.30561826,
 -0.6333279,
 -1.2131691,
 -2.2672105,
 -3.9158158,
 -6.1540804,
 -8.8191996,
 -11.635813,
 -14.286608,
 -16.462812,
 -17.943993,
 -18.685251,
 -18.793633,
 -18.618328,
 -18.118195,
 -17.951168,
 -18.098774,
 -18.851744,
 -20.435009,
 -22.447792,
 -24.747829,
 -26.917685,
 -28.524897,
 -29.264297,
 -29.117596,
 -28.16255,
 -26.49226,
 -24.614424,
 -23.007265,
 -21.564766,
 -20.551498,
 -19.873325,
 -19.364033,
 -18.779448,
 -17.821594,
 -16.258081,
 -13.970452,
 -10.981444,
 -7.4684205,
 -3.7501016,
 -0.15635261,
 2.9772096,
 5.4395175,
 7.132719,
 8.1294727,
 8.6363735,
 8.9281454,
 9.2941914,
 9.9697638,
 11.079341,
 12.562345,
 14.237216,
 15.809559,
 16.950453,
 17.381609,
 16.938213,
 15.613298,
 13.565508,
 11.076948,
 8.5601444,
 6.3773327,
 4.7868357,
 3.8481448,
 3.5138502,
 3.4309714,
 3.2506099,
 2.566304,
 1.0682412,
 -1.3450102,
 -4.6130099,
 -8.4637041,
 -12.477572,
 -16.190722,
 -19.199072,
 -21.232229,
 -22.2255,
 -22.393307,
 -22.079338,
 -21.793861,
 -21.614162,
 -22.374886,
 -23.66847,
 -25.927404,
 -28.822844,
 -31.882042,
 -34.618858,
 -36.46328,
 -37.181564,
 -36.713959,
 -35.274097,
 -32.990501,
 -30.3675,
 -28.14781,
 -26.146204,
 -24.562925,
 -23.419697,
 -22.426777,
 -21.289103,
 -19.713984,
 -17.391224,
 -14.248542,
 -10.339687,
 -5.8647647,
 -1.2366034,
 3.1689858,
 6.9699464,
 9.9272709,
 11.941613,
 13.105474,
 13.669283,
 13.953389,
 14.285711,
 14.963319,
 16.072838,
 17.565983,
 19.225279,
 20.71352,
 21.667736,
 21.790447,
 20.899879,
 19.012274,
 16.317251,
 13.150064,
 9.8907776,
 7.0072351,
 4.6846075,
 3.029146,
 1.946041,
 1.0919217,
 0.098311812,
 -1.4406949,
 -3.7629437,
 -7.0077753,
 -11.050987,
 -15.585717,
 -20.164085,
 -24.306299,
 -27.590773,
 -29.736839,
 -30.848682,
 -31.069016,
 -31.061485,
 -30.686995,
 -31.03178,
 -31.748112,
 -33.299461,
 -35.919754,
 -38.907295,
 -42.115463,
 -44.881657,
 -46.655895,
 -47.255661,
 -46.659004,
 -44.970089,
 -42.384876,
 -39.614048,
 -37.0881,
 -34.543285,
 -32.361893,
 -30.367081,
 -28.412891,
 -26.267321,
 -23.673107,
 -20.30426,
 -16.241322,
 -11.44988,
 -6.1824598,
 -0.85328394,
 4.2204714,
 8.6730185,
 12.300264,
 15.019421,
 16.92049,
 18.221352,
 19.190708,
 20.100592,
 21.194742,
 22.507717,
 23.986784,
 25.428883,
 26.544825,
 27.042883,
 26.692993,
 25.375931,
 23.142214,
 20.169025,
 16.782326,
 13.413164,
 10.326519,
 7.692627,
 5.611692,
 3.8167367,
 2.0629253,
 -0.033325758,
 -2.7715981,
 -6.3110862]

6.其他內容

6.1 檢視tensor、node等
方法一：pywrap_tensorflow
tf.train.get_checkpoint_state(checkpoint_dir='checkpoint路徑')  # checkpoint路徑比如放在C:\Users\Administrator\Documents\checkpoint，這裡填寫r'C:\Users\Administrator\Documents\'即可

import os

logdir='./'

from tensorflow.python import pywrap_tensorflow
ckpt = tf.train.get_checkpoint_state(logdir)
reader = pywrap_tensorflow.NewCheckpointReader(ckpt.model_checkpoint_path)
var_to_shape_map = reader.get_variable_to_shape_map()
for key in var_to_shape_map:
    print("tensor_name: ", key)  # tensor的名稱
#     print(reader.get_tensor(key))

# 參考：https://blog.csdn.net/wc781708249/article/details/78040735

tensor_name:  dense/kernel
tensor_name:  beta1_power
tensor_name:  beta2_power
tensor_name:  dense/bias
tensor_name:  rnn/basic_rnn_cell/bias
tensor_name:  dense/bias/Adam_1
tensor_name:  dense/bias/Adam
tensor_name:  dense/kernel/Adam_1
tensor_name:  dense/kernel/Adam
tensor_name:  rnn/basic_rnn_cell/bias/Adam
tensor_name:  rnn/basic_rnn_cell/bias/Adam_1
tensor_name:  rnn/basic_rnn_cell/kernel
tensor_name:  rnn/basic_rnn_cell/kernel/Adam
tensor_name:  rnn/basic_rnn_cell/kernel/Adam_1
方法二：inspect_checkpoint
inspect_checkpoint.print_tensors_in_checkpoint_file(file_name=,tensor_name=,all_tensors=)  # file_name引數填寫路徑，比如checkpoint等四個模型檔案存放在C:\Users\Administrator\Documents，其中meta檔案C:\Users\Administrator\Documents\my_time_series_model.meta，所以file_name=r'C:\Users\Administrator\Documents\my_time_series_model'

help(chkp.print_tensors_in_checkpoint_file)

Help on function print_tensors_in_checkpoint_file in module tensorflow.python.tools.inspect_checkpoint:

print_tensors_in_checkpoint_file(file_name, tensor_name, all_tensors)
    Prints tensors in a checkpoint file.
    
    If no `tensor_name` is provided, prints the tensor names and shapes
    in the checkpoint file.
    
    If `tensor_name` is provided, prints the content of the tensor.
    
    Args:
      file_name: Name of the checkpoint file.
      tensor_name: Name of the tensor in the checkpoint file to print.
      all_tensors: Boolean indicating whether to print all tensors.

#使用inspect_checkpoint來檢視ckpt裡的內容
from tensorflow.python.tools import inspect_checkpoint as chkp

chkp.print_tensors_in_checkpoint_file(file_name="./my_time_series_model",
                                      tensor_name=None, # 如果為None,則預設為ckpt裡的所有變數
                                      all_tensors=False, # bool 是否列印所有的tensor，這裡打印出的是tensor的值，一般不推薦這裡設定為False
                                     ) # bool 是否列印所有的tensor的name

#上print_tensors_in_checkpoint_file其實是用NewCheckpointReader實現的。

beta1_power (DT_FLOAT) []
beta2_power (DT_FLOAT) []
dense/bias (DT_FLOAT) [1]
dense/bias/Adam (DT_FLOAT) [1]
dense/bias/Adam_1 (DT_FLOAT) [1]
dense/kernel (DT_FLOAT) [100,1]
dense/kernel/Adam (DT_FLOAT) [100,1]
dense/kernel/Adam_1 (DT_FLOAT) [100,1]
rnn/basic_rnn_cell/bias (DT_FLOAT) [100]
rnn/basic_rnn_cell/bias/Adam (DT_FLOAT) [100]
rnn/basic_rnn_cell/bias/Adam_1 (DT_FLOAT) [100]
rnn/basic_rnn_cell/kernel (DT_FLOAT) [101,100]
rnn/basic_rnn_cell/kernel/Adam (DT_FLOAT) [101,100]
rnn/basic_rnn_cell/kernel/Adam_1 (DT_FLOAT) [101,100]
方法三：檢視所node的名稱

先載入模型，獲取圖結構，然後列印圖結構中的node。
(便於獲取變數為基於模型的應用服務)

tf.get_default_graph().as_graph_def().node

# 檢視所有的tensor名稱
[tensor.name for tensor in tf.get_default_graph().as_graph_def().node]

['X',
 'y',
 'Rank',
 'range/start',
 'range/delta',
 'range',
 'concat/values_0',
 'concat/axis',
 'concat',
 'transpose',
 'rnn/Shape',
 'rnn/strided_slice/stack',
 'rnn/strided_slice/stack_1',
 'rnn/strided_slice/stack_2',
 'rnn/strided_slice',
 'rnn/strided_slice_1/stack',
 'rnn/strided_slice_1/stack_1',
 'rnn/strided_slice_1/stack_2',
 'rnn/strided_slice_1',
 'rnn/BasicRNNCellZeroState/ExpandDims/dim',
 'rnn/BasicRNNCellZeroState/ExpandDims',
 'rnn/BasicRNNCellZeroState/Const',
 'rnn/BasicRNNCellZeroState/concat/axis',
 'rnn/BasicRNNCellZeroState/concat',
 'rnn/BasicRNNCellZeroState/ExpandDims_1/dim',
 'rnn/BasicRNNCellZeroState/ExpandDims_1',
 'rnn/BasicRNNCellZeroState/Const_1',
 'rnn/BasicRNNCellZeroState/zeros/Const',
 'rnn/BasicRNNCellZeroState/zeros',
 'rnn/Shape_1',
 'rnn/strided_slice_2/stack',
 'rnn/strided_slice_2/stack_1',
 'rnn/strided_slice_2/stack_2',
 'rnn/strided_slice_2',
 'rnn/strided_slice_3/stack',
 'rnn/strided_slice_3/stack_1',
 'rnn/strided_slice_3/stack_2',
 'rnn/strided_slice_3',
 'rnn/ExpandDims/dim',
 'rnn/ExpandDims',
 'rnn/Const',
 'rnn/concat/axis',
 'rnn/concat',
 'rnn/zeros/Const',
 'rnn/zeros',
 'rnn/time',
 'rnn/TensorArray',
 'rnn/TensorArray_1',
 'rnn/TensorArrayUnstack/Shape',
 'rnn/TensorArrayUnstack/strided_slice/stack',
 'rnn/TensorArrayUnstack/strided_slice/stack_1',
 'rnn/TensorArrayUnstack/strided_slice/stack_2',
 'rnn/TensorArrayUnstack/strided_slice',
 'rnn/TensorArrayUnstack/range/start',
 'rnn/TensorArrayUnstack/range/delta',
 'rnn/TensorArrayUnstack/range',
 'rnn/TensorArrayUnstack/TensorArrayScatter/TensorArrayScatterV3',
 'rnn/while/Enter',
 'rnn/while/Enter_1',
 'rnn/while/Enter_2',
 'rnn/while/Merge',
 'rnn/while/Merge_1',
 'rnn/while/Merge_2',
 'rnn/while/Less/Enter',
 'rnn/while/Less',
 'rnn/while/LoopCond',
 'rnn/while/Switch',
 'rnn/while/Switch_1',
 'rnn/while/Switch_2',
 'rnn/while/Identity',
 'rnn/while/Identity_1',
 'rnn/while/Identity_2',
 'rnn/while/TensorArrayReadV3/Enter',
 'rnn/while/TensorArrayReadV3/Enter_1',
 'rnn/while/TensorArrayReadV3',
 'rnn/basic_rnn_cell/kernel/Initializer/random_uniform/shape',
 'rnn/basic_rnn_cell/kernel/Initializer/random_uniform/min',
 'rnn/basic_rnn_cell/kernel/Initializer/random_uniform/max',
 'rnn/basic_rnn_cell/kernel/Initializer/random_uniform/RandomUniform',
 'rnn/basic_rnn_cell/kernel/Initializer/random_uniform/sub',
 'rnn/basic_rnn_cell/kernel/Initializer/random_uniform/mul',
 'rnn/basic_rnn_cell/kernel/Initializer/random_uniform',
 'rnn/basic_rnn_cell/kernel',
 'rnn/basic_rnn_cell/kernel/Assign',
 'rnn/basic_rnn_cell/kernel/read',
 'rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat/axis',
 'rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat',
 'rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul/Enter',
 'rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul',
 'rnn/basic_rnn_cell/bias/Initializer/Const',
 'rnn/basic_rnn_cell/bias',
 'rnn/basic_rnn_cell/bias/Assign',
 'rnn/basic_rnn_cell/bias/read',
 'rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd/Enter',
 'rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd',
 'rnn/while/rnn/basic_rnn_cell/Relu',
 'rnn/while/TensorArrayWrite/TensorArrayWriteV3/Enter',
 'rnn/while/TensorArrayWrite/TensorArrayWriteV3',
 'rnn/while/add/y',
 'rnn/while/add',
 'rnn/while/NextIteration',
 'rnn/while/NextIteration_1',
 'rnn/while/NextIteration_2',
 'rnn/while/Exit',
 'rnn/while/Exit_1',
 'rnn/while/Exit_2',
 'rnn/TensorArrayStack/TensorArraySizeV3',
 'rnn/TensorArrayStack/range/start',
 'rnn/TensorArrayStack/range/delta',
 'rnn/TensorArrayStack/range',
 'rnn/TensorArrayStack/TensorArrayGatherV3',
 'rnn/Const_1',
 'rnn/Rank',
 'rnn/range/start',
 'rnn/range/delta',
 'rnn/range',
 'rnn/concat_1/values_0',
 'rnn/concat_1/axis',
 'rnn/concat_1',
 'rnn/transpose',
 'Reshape/shape',
 'Reshape',
 'dense/kernel/Initializer/random_uniform/shape',
 'dense/kernel/Initializer/random_uniform/min',
 'dense/kernel/Initializer/random_uniform/max',
 'dense/kernel/Initializer/random_uniform/RandomUniform',
 'dense/kernel/Initializer/random_uniform/sub',
 'dense/kernel/Initializer/random_uniform/mul',
 'dense/kernel/Initializer/random_uniform',
 'dense/kernel',
 'dense/kernel/Assign',
 'dense/kernel/read',
 'dense/bias/Initializer/zeros',
 'dense/bias',
 'dense/bias/Assign',
 'dense/bias/read',
 'dense/MatMul',
 'dense/BiasAdd',
 'outputs/shape',
 'outputs',
 'sub',
 'Square',
 'Const',
 'Mean',
 'gradients/Shape',
 'gradients/Const',
 'gradients/Fill',
 'gradients/f_count',
 'gradients/f_count_1',
 'gradients/Merge',
 'gradients/Switch',
 'gradients/Add/y',
 'gradients/Add',
 'gradients/NextIteration',
 'gradients/f_count_2',
 'gradients/b_count',
 'gradients/b_count_1',
 'gradients/Merge_1',
 'gradients/GreaterEqual/Enter',
 'gradients/GreaterEqual',
 'gradients/b_count_2',
 'gradients/Switch_1',
 'gradients/Sub',
 'gradients/NextIteration_1',
 'gradients/b_count_3',
 'gradients/Mean_grad/Reshape/shape',
 'gradients/Mean_grad/Reshape',
 'gradients/Mean_grad/Shape',
 'gradients/Mean_grad/Tile',
 'gradients/Mean_grad/Shape_1',
 'gradients/Mean_grad/Shape_2',
 'gradients/Mean_grad/Const',
 'gradients/Mean_grad/Prod',
 'gradients/Mean_grad/Const_1',
 'gradients/Mean_grad/Prod_1',
 'gradients/Mean_grad/Maximum/y',
 'gradients/Mean_grad/Maximum',
 'gradients/Mean_grad/floordiv',
 'gradients/Mean_grad/Cast',
 'gradients/Mean_grad/truediv',
 'gradients/Square_grad/mul/x',
 'gradients/Square_grad/mul',
 'gradients/Square_grad/mul_1',
 'gradients/sub_grad/Shape',
 'gradients/sub_grad/Shape_1',
 'gradients/sub_grad/BroadcastGradientArgs',
 'gradients/sub_grad/Sum',
 'gradients/sub_grad/Reshape',
 'gradients/sub_grad/Sum_1',
 'gradients/sub_grad/Neg',
 'gradients/sub_grad/Reshape_1',
 'gradients/sub_grad/tuple/group_deps',
 'gradients/sub_grad/tuple/control_dependency',
 'gradients/sub_grad/tuple/control_dependency_1',
 'gradients/outputs_grad/Shape',
 'gradients/outputs_grad/Reshape',
 'gradients/dense/BiasAdd_grad/BiasAddGrad',
 'gradients/dense/BiasAdd_grad/tuple/group_deps',
 'gradients/dense/BiasAdd_grad/tuple/control_dependency',
 'gradients/dense/BiasAdd_grad/tuple/control_dependency_1',
 'gradients/dense/MatMul_grad/MatMul',
 'gradients/dense/MatMul_grad/MatMul_1',
 'gradients/dense/MatMul_grad/tuple/group_deps',
 'gradients/dense/MatMul_grad/tuple/control_dependency',
 'gradients/dense/MatMul_grad/tuple/control_dependency_1',
 'gradients/Reshape_grad/Shape',
 'gradients/Reshape_grad/Reshape',
 'gradients/rnn/transpose_grad/InvertPermutation',
 'gradients/rnn/transpose_grad/transpose',
 'gradients/rnn/TensorArrayStack/TensorArrayGatherV3_grad/TensorArrayGrad/TensorArrayGradV3',
 'gradients/rnn/TensorArrayStack/TensorArrayGatherV3_grad/TensorArrayGrad/gradient_flow',
 'gradients/rnn/TensorArrayStack/TensorArrayGatherV3_grad/TensorArrayScatter/TensorArrayScatterV3',
 'gradients/zeros_like',
 'gradients/rnn/while/Exit_1_grad/b_exit',
 'gradients/rnn/while/Exit_2_grad/b_exit',
 'gradients/rnn/while/Switch_1_grad/b_switch',
 'gradients/rnn/while/Switch_2_grad/b_switch',
 'gradients/rnn/while/Merge_1_grad/Switch',
 'gradients/rnn/while/Merge_1_grad/tuple/group_deps',
 'gradients/rnn/while/Merge_1_grad/tuple/control_dependency',
 'gradients/rnn/while/Merge_1_grad/tuple/control_dependency_1',
 'gradients/rnn/while/Merge_2_grad/Switch',
 'gradients/rnn/while/Merge_2_grad/tuple/group_deps',
 'gradients/rnn/while/Merge_2_grad/tuple/control_dependency',
 'gradients/rnn/while/Merge_2_grad/tuple/control_dependency_1',
 'gradients/rnn/while/Enter_1_grad/Exit',
 'gradients/rnn/while/Enter_2_grad/Exit',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayGrad/TensorArrayGradV3/Enter',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayGrad/TensorArrayGradV3',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayGrad/gradient_flow',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayReadV3/f_acc',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayReadV3/RefEnter',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayReadV3/StackPush',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayReadV3/StackPop/RefEnter',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayReadV3/StackPop',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayReadV3/b_sync',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/TensorArrayReadV3',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/tuple/group_deps',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/tuple/control_dependency',
 'gradients/rnn/while/TensorArrayWrite/TensorArrayWriteV3_grad/tuple/control_dependency_1',
 'gradients/AddN',
 'gradients/rnn/while/rnn/basic_rnn_cell/Relu_grad/ReluGrad/f_acc',
 'gradients/rnn/while/rnn/basic_rnn_cell/Relu_grad/ReluGrad/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/Relu_grad/ReluGrad/StackPush',
 'gradients/rnn/while/rnn/basic_rnn_cell/Relu_grad/ReluGrad/StackPop/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/Relu_grad/ReluGrad/StackPop',
 'gradients/rnn/while/rnn/basic_rnn_cell/Relu_grad/ReluGrad',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd_grad/BiasAddGrad',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd_grad/tuple/group_deps',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd_grad/tuple/control_dependency',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd_grad/tuple/control_dependency_1',
 'gradients/rnn/while/Switch_1_grad_1/NextIteration',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/MatMul/Enter',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/MatMul',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/MatMul_1/f_acc',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/MatMul_1/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/MatMul_1/StackPush',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/MatMul_1/StackPop/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/MatMul_1/StackPop',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/MatMul_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/tuple/group_deps',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/tuple/control_dependency',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul_grad/tuple/control_dependency_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd/Enter_grad/b_acc',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd/Enter_grad/b_acc_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd/Enter_grad/b_acc_2',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd/Enter_grad/Switch',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd/Enter_grad/Add',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd/Enter_grad/NextIteration',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/BiasAdd/Enter_grad/b_acc_3',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/Rank',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/mod/f_acc',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/mod/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/mod/StackPush',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/mod/StackPop/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/mod/StackPop',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/mod',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/Shape',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/f_acc',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/StackPush',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/StackPop/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/StackPop',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/f_acc_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/RefEnter_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/StackPush_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/StackPop_1/RefEnter',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN/StackPop_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ShapeN',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/ConcatOffset',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/Slice',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/Slice_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/tuple/group_deps',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/tuple/control_dependency',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/concat_grad/tuple/control_dependency_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul/Enter_grad/b_acc',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul/Enter_grad/b_acc_1',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul/Enter_grad/b_acc_2',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul/Enter_grad/Switch',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul/Enter_grad/Add',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul/Enter_grad/NextIteration',
 'gradients/rnn/while/rnn/basic_rnn_cell/basic_rnn_cell/MatMul/Enter_grad/b_acc_3',
 'gradients/rnn/while/Switch_2_grad_1/NextIteration',
 'beta1_power/initial_value',
 'beta1_power',
 'beta1_power/Assign',
 'beta1_power/read',
 'beta2_power/initial_value',
 'beta2_power',
 'beta2_power/Assign',
 'beta2_power/read',
 'rnn/basic_rnn_cell/kernel/Adam/Initializer/zeros',
 'rnn/basic_rnn_cell/kernel/Adam',
 'rnn/basic_rnn_cell/kernel/Adam/Assign',
 'rnn/basic_rnn_cell/kernel/Adam/read',
 'rnn/basic_rnn_cell/kernel/Adam_1/Initializer/zeros',
 'rnn/basic_rnn_cell/kernel/Adam_1',
 'rnn/basic_rnn_cell/kernel/Adam_1/Assign',
 'rnn/basic_rnn_cell/kernel/Adam_1/read',
 'rnn/basic_rnn_cell/bias/Adam/Initializer/zeros',
 'rnn/basic_rnn_cell/bias/Adam',
 'rnn/basic_rnn_cell/bias/Adam/Assign',
 'rnn/basic_rnn_cell/bias/Adam/read',
 'rnn/basic_rnn_cell/bias/Adam_1/Initializer/zeros',
 'rnn/basic_rnn_cell/bias/Adam_1',
 'rnn/basic_rnn_cell/bias/Adam_1/Assign',
 'rnn/basic_rnn_cell/bias/Adam_1/read',
 'dense/kernel/Adam/Initializer/zeros',
 'dense/kernel/Adam',
 'dense/kernel/Adam/Assign',
 'dense/kernel/Adam/read',
 'dense/kernel/Adam_1/Initializer/zeros',
 'dense/kernel/Adam_1',
 'dense/kernel/Adam_1/Assign',
 'dense/kernel/Adam_1/read',
 'dense/bias/Adam/Initializer/zeros',
 'dense/bias/Adam',
 'dense/bias/Adam/Assign',
 'dense/bias/Adam/read',
 'dense/bias/Adam_1/Initializer/zeros',
 'dense/bias/Adam_1',
 'dense/bias/Adam_1/Assign',
 'dense/bias/Adam_1/read',
 'Adam/learning_rate',
 'Adam/beta1',
 'Adam/beta2',
 'Adam/epsilon',
 'Adam/update_rnn/basic_rnn_cell/kernel/ApplyAdam',
 'Adam/update_rnn/basic_rnn_cell/bias/ApplyAdam',
 'Adam/update_dense/kernel/ApplyAdam',
 'Adam/update_dense/bias/ApplyAdam',
 'Adam/mul',
 'Adam/Assign',
 'Adam/mul_1',
 'Adam/Assign_1',
 'Adam',
 'init',
 'save/Const',
 'save/SaveV2/tensor_names',
 'save/SaveV2/shape_and_slices',
 'save/SaveV2',
 'save/control_dependency',
 'save/RestoreV2/tensor_names',
 'save/RestoreV2/shape_and_slices',
 'save/RestoreV2',
 'save/Assign',
 'save/RestoreV2_1/tensor_names',
 'save/RestoreV2_1/shape_and_slices',
 'save/RestoreV2_1',
 'save/Assign_1',
 'save/RestoreV2_2/tensor_names',
 'save/RestoreV2_2/shape_and_slices',
 'save/RestoreV2_2',
 'save/Assign_2',
 'save/RestoreV2_3/tensor_names',
 'save/RestoreV2_3/shape_and_slices',
 'save/RestoreV2_3',
 'save/Assign_3',
 'save/RestoreV2_4/tensor_names',
 'save/RestoreV2_4/shape_and_slices',
 'save/RestoreV2_4',
 'save/Assign_4',
 'save/RestoreV2_5/tensor_names',
 'save/RestoreV2_5/shape_and_slices',
 'save/RestoreV2_5',
 'save/Assign_5',
 'save/RestoreV2_6/tensor_names',
 'save/RestoreV2_6/shape_and_slices',
 'save/RestoreV2_6',
 'save/Assign_6',
 'save/RestoreV2_7/tensor_names',
 'save/RestoreV2_7/shape_and_slices',
 'save/RestoreV2_7',
 'save/Assign_7',
 'save/RestoreV2_8/tensor_names',
 'save/RestoreV2_8/shape_and_slices',
 'save/RestoreV2_8',
 'save/Assign_8',
 'save/RestoreV2_9/tensor_names',
 'save/RestoreV2_9/shape_and_slices',
 'save/RestoreV2_9',
 'save/Assign_9',
 'save/RestoreV2_10/tensor_names',
 'save/RestoreV2_10/shape_and_slices',
 'save/RestoreV2_10',
 'save/Assign_10',
 'save/RestoreV2_11/tensor_names',
 'save/RestoreV2_11/shape_and_slices',
 'save/RestoreV2_11',
 'save/Assign_11',
 'save/RestoreV2_12/tensor_names',
 'save/RestoreV2_12/shape_and_slices',
 'save/RestoreV2_12',
 'save/Assign_12',
 'save/RestoreV2_13/tensor_names',
 'save/RestoreV2_13/shape_and_slices',
 'save/RestoreV2_13',
 'save/Assign_13',
 'save/restore_all']
6.2關於不同版本的checkpoint檔案理解
  • 對於tensorflow1.2版本及以上，直接書寫完整的ckpt檔案的路徑中的model_name(比如，my_model.meta書寫my_model即可)即可




參考：
(1) tensorflow的checkpoint檔案的版本
(2) TensorFlow檢視ckpt中變數的幾種方法

二、學習其他簡單的

1. 儲存變數

# Create some variables.
v1 = tf.get_variable("v1", shape=[3], initializer = tf.zeros_initializer)
v2 = tf.get_variable("v2", shape=[5], initializer = tf.zeros_initializer)


inc_v1 = v1.assign(v1+1)
dec_v2 = v2.assign(v2-1)

# Add an op to initialize the variables.
init_op = tf.global_variables_initializer()

# Add ops to save and restore all the variables.
saver = tf.train.Saver()

# Later, launch the model, initialize the variables, do some work, and save the
# variables to disk.
with tf 
 

            
  

           

              
              

            

            
相關推薦

			   
            
            
            
 

    


    
    
Tensorflow的變數和模型儲存以及模型應用

     
  
 
  
  
 Table of Contents 
 
  
   一、模型部分(成功)
    
     1.儲存的模型
     2.載入模型並用於預測
      
       1.載入圖結構和引數
       2.獲取圖
       3.獲取tensor
       4.新的inpu 



  
 

    


    
    
《TensorFlow：實戰Google深度學習框架》——5.4 模型持久化（模型儲存、模型載入）

     
  
 
 目錄 
 1、持久化程式碼實現 
 2、載入儲存的TensorFlow模型 
 3、載入部分變數 
 4、載入變數時重新命名 
 
 1、持久化程式碼實現 
 TensorFlow提供了一個非常簡單的API來儲存和還原一個神經網路模型。這個API就是tf.train.Saver類。一下程式碼給出了儲 



  
 

    


    
    
caffe 網路模型檔案中的引數含義(top bottom lr_mult decay_mult)與模型編寫以及模型自定義

     
  
 
  
  
 
 
  文章目錄
  
   
    基本概念
    資料層
    Convolution Layer
    Deconvolution 反捲積層
    Batch Normalization 層
    Polling Layer
    crop層　裁剪
    全連線層 



  
 

    


    
    
變數和作用域以及記憶體問題

     
 在javascript中有五種基本型別：string/boolean/null/undefind/number,一種引用型別：object 
  
記憶體中的未知： 1/因為基本型別資料是簡單型別資料執行時建立在棧記憶體中即可，2/object是複雜型別資料佔記憶體多所以儲存在堆記憶體中以便於其他變 



  
 

    


    
    
java基礎-靜態變數和靜態常量以及靜態方法的理解

     
  
 
 
 
 
  靜態變數是用static修飾符修飾的變數，靜態常量是static  final修飾的變數；
 
 
  宣告和訪問許可權：
 
 靜態變數：必須是成員變數，static不能修飾區域性變數。靜態成員變數可以通過類名訪問，也可以通過物件名進行訪問。
 靜態方法：用static修飾 



  
 

    


    
    
java併發中的原子變數和原子操作以及CAS介紹

     
 
      很多情況下我們只是需要一個簡單的、高效的、執行緒安全的遞增遞減方案。注意，這裡有三個條件：簡單，意味著程式設計師儘可能少的操作底層或者實現起來要比較容易；高效意味著耗用資源要少，程式處理速度要快；執行緒安全也非常重要，這個在多執行緒下能保證資料的正確性。這三個條件看起來比較簡單，但是實現起來卻難 



  
 

    


    
    
Android內部儲存和外部儲存以及快取清理和記憶體清理！

     
 
                android內部的儲存結構和路徑獲取方法：清除本地資料和快取：計算快取和記憶體資料的大小：計算快取大小以及刪除的方法：這文章裡面的檔案刪除方法有一些問題：沒有迴圈是資料夾的刪除方法；另外：File.list()； 獲取的是該資料夾下的所有子檔案的相對路徑；File.list 



  
 

    


    
    
tensorflow學習（2.網路模型的儲存以及提取）

     
 
                第一篇學習了CNN網路的構建以及程式碼的基礎結構，第二篇則是實際專案過程中需要的網路模型的儲存

先放上儲存的程式碼：

#tf可以認為是全域性變數，從該變數為類，從中取input_data變數
import tensorflow.examples.tutorials.mni 



  
 

    


    
    
AI-031: Google機器學習教程（ML Crash Course with TensorFlow APIs）筆記8 - 練習使用校驗集和測試集以及評估模型

     
  
 
 本文是Google機器學習教程（ML Crash Course with TensorFlow APIs）的學習筆記。教程地址： 
 https://developers.google.com/machine-learning/crash-course/ml-intro 
 8. 使用校驗集和測試集 



  
 

    


    
    
tensorflow 檢查點和模型,儲存與恢復使用,官方教程(一)

     
 
                檢查點：這種格式依賴於建立模型的程式碼。SavedModel：這種格式與建立模型的程式碼無關。示例程式碼本文件依賴於 TensorFlow 使用入門中詳細介紹的同一個鳶尾花分類示例。要下載和訪問該示例，請執行下列兩個命令：git clone https://github.co 



  
 

    


    
    
Keras 多GPU下模型和引數儲存Modelcheckpoint callback報錯問題以及在單GPU/CPU下載入

     
 
							
							
							儲存



多GPU下無法直接儲存多GPU模型結構和引數，而應該使用template model進行儲存


  keras文件：To save the multi-gpu model, use .save(fname) or .save_weights(fna 



  
 

    


    
    
tensorflow儲存模型、載入模型和提取模型引數和特徵圖

     
 
                1.tf.train.latest_checkpoint('./model_data/')這一句最終返回的是一個字串，比如'./model_data/model-99991'這個方法本身還會做相應的檢查，比如checkpoint中最新的模型model_checkpoint_p 



  
 

    


    
    
【tensorflow】模型儲存和恢復

     
 
							
							
							  tensorflow中，模型的儲存和恢復使用tf.train.Saver類，模型儲存使用該類的 save 方法。模型恢復使用restore 方法。



模型儲存

  模型儲存使用tf.train.Saver.save()方法。以saver.save(se 



  
 

    


    
    
tensorflow-模型儲存和載入（一）

     
 
								
								            
						
                模型儲存和載入（一）TensorFlow的模型格式有很多種，針對不同場景可以使用不同的格式。格式簡介Checkpoint用於儲存模型的權重，主要用於模型訓練過程中引數的備份和模型訓練熱啟動。GraphD 



  
 

    


    
    
[TensorFlow]入門學習筆記(6)-Tensorboard簡易教程和模型儲存

     
 
							
							
							模型儲存

tf.train.Saver()

The Saver class adds ops to save and restore variables to and from checkpoints. It also provides convenienc 



  
 

    


    
    
TensorFlow學習系列（三）：儲存/恢復和混合多個模型

     
 
							
							
							
  這篇教程是翻譯Morgan寫的TensorFlow教程，作者已經授權翻譯，這是原文。






目錄

















在學習這篇部落格之前，我希望你已經掌握了Tensorflow基本的操作。如果沒有，你可以閱讀這篇入門文章。

為什麼要 



  
 

    


    
    
tensorflow的一些程式碼分析（五） tensorflow模型儲存和視覺化

     
 
                

儲存與讀取模型

在使用tf來訓練模型的時候，難免會出現中斷的情況。這時候自然就希望能夠將辛辛苦苦得到的中間引數保留下來，不然下次又要重新開始。好在tf官方提供了儲存和讀取模型的方法。

儲存模型的方法：
# 之前是各種構建模型graph的操作(矩陣相乘，sigmoid等 



  
 

    


    
    
一份快速完整的Tensorflow模型儲存和恢復教程（譯）

     
 
							
							
							

在本教程中，我將介紹： 
 - tensorflow模型是什麼樣子的? 
 - 如何儲存一個Tensorflow模型？ 
 - 如何恢復一個Tensorflow模型用於預測/遷移學習？ 
 - 如何匯入預訓練的模型進行微調和修改？





什麼是Tenso 



  
 

    


    
    
Tensorflow模型儲存和過載

     
 
							
							
							最近因為專案要求，需要把模型的訓練和測試過程分開，這裡主要涉及兩個過程：訓練圖的存取和引數的存取。 
以下所有/home/yy/xiajbxie/model是我的模型的儲存路徑，將其換成你自己的即可。



tf.train.Saver()

Saver的作用中 



  
 

    


    
    
TensorFlow：儲存/恢復和混合多個模型

     
 這篇教程是翻譯Morgan寫的TensorFlow教程，作者已經授權翻譯，這是原文
目錄





在學習這篇部落格之前，我希望你已經掌握了Tensorflow基本的操作。如果沒有，你可以閱讀這篇入門文章。
為什麼要學習模型的儲存和恢復呢？因為這對於避免資料的混亂無序是至關重要的，特別是在你程式碼中的不同圖。