Skip to main content

MNIST Handwritten Digits Recognition

Hello guys, in this post we will learn about MNITS(Modified National Institute of Standards and Technology) Hand written digits recognition. Here we will be creating a Deep Neural Network model that recognizes the hand written digits. MNIST is nothing but a dataset that contains 60000 training images and 10000 testing images of hand written digits and these are actually gray scale images. You can find full code here 

The below are the sample images of MNIST dataset.
See the source image

Now lets dive into code!!!

Setup

First we will import some necessary modules like numpy and keras. Numpy is used to perform algebric or numerical operations and Keras is a deep learning API written in Python, running on top of the machine learning platform Tensorflow(Tensorflow is an end-to-end, open-source machine learning platform.)

import numpy as np
from tensorflow import keras
from tensorflow.keras import layers


Load the dataset

Now lets load the MNIST dataset.

(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()

Scaling and Normalizing the images

We scale images so that  the range of each pixel will be in the range of [0,1].To do that we want to divide each pixel with 255.

x_train = x_train.astype("float32") / 255

x_test = x_test.astype("float32") / 255

Prepare the data

Now the shape of the training and testing images should be in the form of(num_images, img_height, img_width, num_channels). For MNIST dataset img_height = img_width=28 and  num_channels=1(gray scale images).


x_train = x_train.reshape(x_train.shape[0],x_train.shape[1],x_train.shape[2],1)

x_test = x_test.reshape(x_test.shape[0],x_test.shape[1],x_test.shape[2],1)


Similarly training and testing labels should be in the form of (num_images,num_classes).Here num_classes=10(0-9).


y_train = keras.utils.to_categorical(y_train, 10)

y_test = keras.utils.to_categorical(y_test, 10)

Build the Model

Now lets build our CNN(Convolution Neural Network) Model.

model = keras.Sequential([

    keras.Input(shape=(28,28,1)),
    layers.Conv2D(32, kernel_size=(3, 3), activation="relu"),
    layers.MaxPooling2D(pool_size=(2, 2)),
    layers.Conv2D(64, kernel_size=(3, 3), activation="relu"),
    layers.MaxPooling2D(pool_size=(2, 2)),
    layers.Flatten(),
    layers.Dropout(0.5),
    layers.Dense(10, activation="softmax"),
    
])

model.summary()

This gives us output as follows which consists of nearly 34k parameters


Train the model

Now, we will compile our model and fit our model. Here we use Categorical CrossEntropy as loss function and Adam opimizer

model.compile(loss="categorical_crossentropy", 
                        optimizer="adam", 
                        metrics=  ["accuracy"])

model.fit(x_train, y_train, batch_size=128, epochs=15, validation_split=0.2)


Evaluate our model

Now lets see how well our model performs we will use model.evaluate to evaluate our model

score = model.evaluate(x_test, y_test, verbose=0)

print("Test loss:", score[0])

print("Test accuracy:", score[1])

This gives us 0.99 accuracy and 0.01 loss.

Finally we have build our very first model that gives 99% accuracy. You can find full code on my github account which I mentioned below. You can find my Introduction to Deep Learning here.
Thank you!

Contacts:

ph.No: +91 9182530027
gmail: hunnurjirao2000@gmail.com
github: github.com/hunnurjirao


 

Comments

Popular posts from this blog

Introduction to Deep Learning

Hello guys, today we will learn what is deep learning and difference between Machine Learning and Deep Learning and also a small introduction about Neural Networks. What is Deep Learning? Deep learning is a type of  machine learning  (ML) and  artificial intelligence  (AI) that imitates the way humans gain certain types of knowledge. Deep learning is an important element of data science, which includes statistics and predictive modeling.   Difference between ML and DL In machine learning, first we should extract features from input and then fed to the model. But when comes to Deep learning there will be no need of feature extraction. There is one of the most important algorithms called Neural Networks which we will discuss in this post. Lets discuss some more concept about the difference between ML and DL. Before that lets know about different types of data. There are two types of data  Structured Data - Price of house, user ID etc.   Unstructured Data - Images, audio, text, etc. Now,

Introduction to Machine Learning

Hello all, welcome to the practical machine learning hub. Here you will be learning some of the projects on machine learning and  deep learning. But in this post let us know what exactly is machine learning and types of machine learning.

Generating Fake Faces

Hello guys, welcome to this new project of Generating Fake Faces. There is a wonderful concept called GANs (Generative Adversarial Network) which generates fake images. You can generate any thing you want. If you have person's faces dataset, you can generate fake faces of persons. If you have Pokémon's dataset, you can generate new Pokémon that never exists on the Earth. GANs were invented by Ian Goodfellow in 2014 and first described in paper Generative Adversarial Nets . If you have zero knowledge about GANs don't worry about it, you will learn how it works here and by the end of the post you will be able to generate whatever you want (only if you have a good and huge dataset!!!). One last thing I want to inform you is that this project is done using PyTorch . You can find the full code here . You can find more fake persons that doesn't exist on the earth  here , once go through it.  OK guys lets start our project. Generative Adversarial Networks Generative Adversari

Hand Signs Using KNN

Hello guys, welcome to this new project named "Hand Signs using KNN". There are different types of Hand Signs, but here we use only hand signs of  Zero, One, Two, Three, Four & Five. But wait.! we are not using any Deep Learning or Neural Networks in this project. Instead we are using KNN a.k.a K-Nearest Neighbor which is one the Machine Learning Algorithms. We will talk about this later in this post. But how can we give directly images as input to the model, if we use a Machine Learning algorithm? Don't worry about it there are two files (train.h5, test.h5) that consists of features of the images (here features is nothing but pixel values of the image). You can find these files and also full code  here . Hand Signs Now lets talk about K Nearest Neighbor. K Nearest Neighbor(A Supervised Learning Algorithm) Introduction KNN algorithm is the laziest algorithm , we can also say KNN as instance based learning. It is called as the laziest algorithm because   it does nothin

A Convolution Neural Network Application

Hello guys, welcome to this new project. In this project you will learn about Convolutional Neural Networks and the project is to Detect You whenever you are in-front of your laptop. One thing I want to conform before starting our project, this project is totally different from face detection. I will discuss about face detection in another session. And guys don't worry about code. 'Why worry, when we have Google's Tensorflow!'.You can find the code here , have a look. OK guys lets start our project.  Convolution Neural Network Convolution Neural Network(CNN) Why CNNs? Let us start the session from this question - Why CNNs? We have DNN(Deep Neural Networks) which performs well on images but why CNNs? To know the reason, let us assume we have an RGB image of shape (64,64,3). This has 12288 (64*64*3) number of pixels (It is also called as 12288 Dimensional feature vector).It is not too bad! But as technology increased we can find images of (1000,1000,3) that has nearly 3 M