Save and Restore variable in TensoFlow-Python

Saving and restoring of variable is frequently used in Machine learning applications. Once a model is trained, trained variable should be stored for future usage of model without re-training.
I am presenting a simple code of Linear regression using Gradient descent. If you will run this code first time, it will do training and save the variable value. If you are running the code again it will used saved value for testing.

import tensorflow as tf
import numpy as np
import os.path



itr=20#number of iteration for training

x=np.array([1,2,3,4,5,6]).astype(np.float32)
y=np.array([1.8,3,3.98,5,6.1,7.2]).astype(np.float32)
xt=np.array([2.5]).astype(np.float32) #for testing

m=tf.Variable(2.0)
c=tf.Variable(1.0)



y_p=m*x+c
yt=m*xt+c#output for test

loss=tf.reduce_mean(tf.square(y_p-y))

optimizer=tf.train.GradientDescentOptimizer(0.01)
train=optimizer.minimize(loss)

with tf.Session() as sess:
    init_op=tf.global_variables_initializer()
    sess.run(init_op)
    # Add ops to save and restore all the variables.
    saver = tf.train.Saver()
    flag=os.path.isfile("/tmp/reg_model.ckpt.index")#file presence
    

    if flag:                 
        saver.restore(sess, "/tmp/reg_model.ckpt")
        print("Model restored. No need of training")
    else:
            for i in range(itr):
                sess.run(train)
                print("Loss: ",sess.run(loss))
            print("m: ",sess.run(loss)," c: ",sess.run(c))
            save_path = saver.save(sess, "/tmp/reg_model.ckpt")
            print("Model saved in file: %s" % save_path)
       
   

    print("Test input: ",xt)
    print("Output by model: ", sess.run(yt))
   
Output on running the code first time:

Loss:  6.8055177
Loss:  3.1509101
Loss:  1.4594502
Loss:  0.6765916
Loss:  0.31426063
Loss:  0.1465635
Loss:  0.06894786
Loss:  0.03302495
Loss:  0.01639891
Loss:  0.008703876
Loss:  0.005142315
Loss:  0.003493938
Loss:  0.0027310376
Loss:  0.0023779308
Loss:  0.00221451
Loss:  0.0021388675
Loss:  0.00210386
Loss:  0.0020876636
Loss:  0.002080158
Loss:  0.002076687
m:  0.002076687  c:  0.78193426
Model saved in file: /tmp/reg_model.ckpt
Test input:  [2.5]
Output by model:  [3.448408]

 Output on running the code next times:

Model restored. No need of training
Test input:  [2.5]
Output by model:  [3.448408]

External Link: https://www.tensorflow.org/programmers_guide/saved_model

Comments

Post a Comment

Popular posts from this blog

Notes on pointer in C

A pointer is a memory address. Assume size of int and char in the compiler as 4 and 1 Byte respectively. & : address of * : Content of (de-referencing operator) int x; //x occupies 4 byte memory int *p; //p is a pointer and address of an integer. It means *p (content of p) will be an integer. p= &x; //now, p is address of x. p is pointer to integer x. p++; //move pointer to next element. As p is the pointer to integer, it's value will increase by 4. int **q; // q is pointer to pointer. (address of address) q=&p; //q is now pointer to pointer to (address of) p. Size of a pointer is fixed for a compiler irrespective of pointer type.  p=(char*)p; //now p is address of char, content of p (*p) will be of size 1 Byte. I recommend to go through this test for checking yourself : https://www.geeksforgeeks.org/c-language-2-gq/pointers-gq/
Read more

Physical Significance of Convolution

Image
Convolution is an operation that takes input signal, and return output signal based on knowledge about the system’s unit impulse response. For a continuous 1-dimensional system, System response (Output) is convolution of input signal and impulse response of the system. 1-d Convolution is defined like this.. where x(t) is input signal, y(t) is output signal and h(t) is impulse response. See this self elaborating example of convolution for physical significance…. The physical significance can be better understand by 2-d convolution. As we know that an image is nothing but a 2 -dimensional function and can be represented as a matrix. Here, second matrix is impulse response. Generally we don’t see such clear-cut pattern due to overlapping of values. Thanks for reading and your suggestions!  
Read more

Convolutional Neural Network

Image
Prequisite :  Basic knowledge of Neural Network Convolutional Neural Network is a supervised deep learning algorithm. It is used for image classification, recommender system and natural language processing. Image classification means categorization of image into particular class like finding cats out of cats and dogs. CNN is inspired by biological structure and process of animal’s virtual cortex. Alike Neural Network (NN), CNN is also made of neurons with trainable weights and biases. A neuron takes weighted sum input from previous layer neurons, and apply an activation function to it. The whole CNN model has a loss function, on training by input-output sets, weights and biases are trained so as to decrease the loss function. The first difference between CNN and NN is that, NN’s input is a vector and CNN’s input is a multi-channeled image (RGB color channels).   CNN consists of input-output layers and in between hidden layers. Input layer is set of pixels of input image,...
Read more