www.ijcrt.

org © 2018 IJCRT | Volume 6, Issue 2 May 2018 | ISSN: 2320-2882

A Review on Deep Learning Using Python Libraries

and Packages
1
Dr. Dushyantsinh B. Rathod, 2Mr. Vijaykumar Gadhavi,
1
Associate Professor & HOD,Computer Engineering, D.A.Degree Engineering &
Technology,Mahemdabad,Gujarat,India
2
HOD,Computer Engineering Department,SSIT,BHAT,Gujarat,India

Abstract- Python is a universal scripting language used programming which encapsulates code within
for wide range of applications. It is ease of use and the objects
compactable across different operating systems. Python e. Python is Easy-to-maintain – The source code is
is well known for its concise, readable code, and is
fairly easy-to-maintain
almost peerless when it comes to ease of use and
f. A Broad Standard Library – Python's greatest
simplicity, particularly provides numerous advantages
for applications in the deep learning. strengths is the huge volume of the library is
In this paper, the survey of various papers that used very portable and cross-platform compatible
python modules and libraries for deep learning are g. Python is Portable – Python can run on a wide
taken and analyzed with metrics like performance, high range of hardware.
precision stability and accuracy obtained of using h. Databases – Python provides interfaces to all
python packages and libraries in deep learning. major commercial databases
i. Python is Scalable - Python provides a good
Index Terms- python, deep learning, CNN, face
structure and supports large programs
recognition, scripting, machine learning, keras, tensor
flow, theano.
III. SIGNIFICANCE OF DEEP LEARNING
I. INTRODUCTION

The significance of deep learning is described as

This survey paper illustrates the various python
follows,
packages that are been used in deep learning which
a. Deep Learning is the best in performance on
significantly increases the performance of complex
problems that particularly outperforms other
real world problems.
solutions in different domains. This includes
speech recognition, image classification, natural
II. SIGNIFICANCE OF PYTHON
language processing etc.
b. It is one of the most time-consuming parts of
The significance of python is described as follows,
machine learning practice. Utilization of
a. Python is Beginner's Language - Python is a
common resources
great language for the beginner programmers and
c. It can be adapted to solve new and complex
supports the development of a wide range of
problems.
applications
d. It has the potential to solve the real world
b. Python is Interpreted- It converts the source code
applications.
into intermediate language which is again
e. Deep Learning unlocks the repository of
translated to machine language.
unstructured big data.
c. Python is Interactive- Python prompt helps to
f. It provides massive amount of data, to solve
interact with the interpreter directly to write
problems end to end.
programs.
d. Python is Object-Oriented- Python supports g. It takes much less time to run.
Object-Oriented style or technique of

IJCRT1134596 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org 83

www.ijcrt.org © 2018 IJCRT | Volume 6, Issue 2 May 2018 | ISSN: 2320-2882
IV. PYTHON LIBRARIES AND PACKAGES FOR possible to write PyLearn2 plugins making use of
DEEP LEARNING mathematical expressions.

a. TensorFlow Python V. RELATED RESEARCH WORKS

TensorFlow is a python library for numerical
computation uses data flow graphs. The Google A. Modern Face Recognition with Deep Learning Jothi
Brain Team researchers developed tensorflow with Thilaga.P et al., [1] has proposed that most predictable
the Machine Intelligence research organization. way to measure a face is by using deep learning
TensorFlow is an open source and available to the techniques. A python application
public and works great for distributed computing. is developed to recognize the faces.
b. Keras Python A robust face recognition system developed using
Keras is written in python and an open source python for security and verification causes which could
framework. It is efficient of running on top of recognize faces independent of the prevailing
TensorFlow, Microsoft Cognitive Toolkit, and conditions.
Theano. It is designed to enable fast The python application is used for identifying the
experimentation with deep neural networks and faces of the persons that pass by the system or the
focuses on being user-friendly, modular, and image is feed into the system by the admin. The user
extensible. interface is designed in such a way that the admin can
c. Apache MXNet decide whether the user can pass through the system or
Apache MXNet is an open-source deep learning not, which can beautomated.
software framework which is used to train, and Histogram of Oriented Gradients (HOGs) is applied for
deploy deep neural networks. It is scalable, which face recognition. The result obtained was the original
supports a flexible programming model and image turned into a very simple representation that
multiple programming languages likepython. features the basic structure of a face.
d. Theano Python A deep convolutional neural network is used to train
Theano is a Python library and enhancing compiler the images and store the measurements using
for manipulating and evaluating mathematical OpenFace. OpenFace is the python and torch
expressions, especially matrix- valued ones. In this implementation for facial recognition.
computations are expressed using a NumPy-esque Python language is used to code the software since
syntax and compiled to run efficiently python is scalable and portable. It uses network mapper
e. PyTorch package as a plugin.
PyTorch is an open-source machine learning library Hence, the result of person identification using HOG
for Python, based on Torch, used for applications techniques performs promising results.
such as natural language processing. It is primarily
developed by Facebook's artificial- intelligence B. A Highly Accurate Deep Learning Based Approach for
research group, and Uber's "Pyro" Probabilistic Developing Wireless
programming language software is built on it. Sensor Network Middleware
f. Lasange Remah A. Alshinina et al [2]., introduced a Secure
Lasagne is a lightweight Python library that helps Wireless Sensor Network Middleware (SWSNM)
us build and train neural networks in Theano. This which is based on an unsupervised learning technique
is a HTTP client library that supports HTTP called generative adversarial network algorithm.
libraries.
g. PyLearn2 SWSNM consists of two networks:
PyLearn2 is a machine learning library with most a. generator (G) network and
functionality built on top of Theano. It is b. discriminator (D) network

IJCRT1134596 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org 84

www.ijcrt.org © 2018 IJCRT | Volume 6, Issue 2 May 2018 | ISSN: 2320-2882
The framework is implemented in Python with in data analysis studies. The functions used for
experiments performed using Keras library. It is implementations are Rectified Linear Unit
a high level neural network API. Keras uses a (ReLu), Hyperbolic Tangent (tanH), Exponential
NSL-KDD dataset by using 40 features. The Linear Unit (eLu), sigmoid, softplus and softsign
analytical model was developed using In this approach, Convolutional Neural Network
MATLAB. (CNN) and SoftMax classifier are used as deep
Middlewares bridges the gap between the learning artificial neural network. The "Modified
application and Wireless sensor networks National Institute of Standards and Technology"
(WSNs) which is an essential medium for the (MNIST) is a huge dataset which has hand
transmission of data. written numbers used for training of image
The paper proposes unsupervised learning for the processing. This dataset was used to measure the
development of WSNs middleware to provide performance of the Tensorflow library.
end-to-end secure system. Hence, in this study, a classification task was
Hence, the SWSNM provides stronger security carried out on the MNIST data set which is
mechanism by recognizing and replacing widely using TensorFlow in deep learning
malicious nodes which leads to lesser energy applications. The accuracy values acquired
consumption and higher throughput. according to the iteration numbers of the ReLu
activation function obtained as the best result.
c. Implementation of Deep-Learning based Image
Classification on Single Board Computer e. E. A Convolutional Neural Network based on
Hasbi Ash Shiddieqy et al.,[3], represented a TensorFlow for Face Recognition
algorithm based on convolutional neural-network Liping Yuan et al.,[5], researched and found that
which is performed using raspberry pi 3 platform in traditional hand-crafted features, there are
in deep learning. uncontrolled environments such as pose, facial
This approach is implemented using python and expression, illumination and occlusion
tflearn for image classification. influencing the accuracy of recognition and it
TFlearn is a high level API and a transparent has poor performance. Hence the deep learning
deep learning library built on top of Tensorflow. method is adopted.
The raspberry pi 3 is efficient to run the CNN in In this paper a Convolutional Neural Network
2D. (CNN) based on TensorFlow, It is an open
The images of cats and dogs were used in the source framework which is used in the effective
classification. The train folder consists of 25,000 face recognitions.
images of dogs and cats. The output from these Here an experiment is conducted on the Linux
images will be a numpy array 50x50 for every system, training CNN model based on
image. TensorFlow.
Thus the result shows as the technique Thus, the result shows that when compared with
implemented in system has the ability to classify traditional hand-crafted features, CNN learning
two category cat and dog which have many features have better robustness to face
similarity. By increasing the size of network the recognition in complex environments.
accuracy can be improved. TensorFlow is the latest second-generation of
Google artificial intelligence, which has been
d. Data Classification with Deep Learning using improved in all aspects, better performance, fully
Tensorflow open source and can be run on more devices. It
Fatih Ertam et al.,[4], illustrates the Tensorflow, obtained the better results for face recognition.
one of the most popular deep learning libraries to
classify MNIST dataset, which is frequently used

IJCRT1134596 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org 85

www.ijcrt.org © 2018 IJCRT | Volume 6, Issue 2 May 2018 | ISSN: 2320-2882

Summary of the Research Related Works – Table 1

IJCRT1134596 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org 86

www.ijcrt.org © 2018 IJCRT | Volume 6, Issue 2 May 2018 | ISSN: 2320-2882

VI. CONCLUSION problems like neural networks. One of the benefit of keras
is it can be executed on the top of python libraries like
This survey paper comprises of various python libraries TensorFlow and CNTK (Cognitive Toolkit)
and modules which are been used in deep learning. The Python supports various libraries and modules that can be
python standard libraries reduce the length of code to be used in deep learning wherein extreme complex problems
written significantly. and multi-stage flow graphs are been solved. Python
Python is capable of interacting with most of the other libraries help in reducing the cognitive overhead on
languages and platforms which are used to solve developers and make them to concentrate on problem-
complex algorithms like, Convolutional Networks, Image solving and achieving project goals.
classification and Neural Networks that are used in the Thus, this survey provides the comparative study of
deep learning. various papers implemented using python libraries and
TensorFlow python library is a simplified approach which shows the benefits of the python packages used in
is used for solving numerical computation using data flow solving the complex algorithms in deep learning.
graph. Python enables reusability of code with the
packages and libraries offered. This library is used in
solving complex applications like face recognitions [5].
Keras library in python is used for solving high complex

REFERENCES

[1] Thilaga, P. J., Khan, B. A., Jones, A. A., &

Kumar, N. K. (2018, April). Modern Face
Recognition with Deep Learning. In 2018
Second International Conference on Inventive
Communication and
Computational Technologies (ICICCT) (pp.
1947-1951). IEEE.
[2] Alshinina, R. A., & Elleithy, K. M. (2018). A
Highly Accurate Deep Learning Based
Approach for Developing Wireless Sensor
Network Middleware. IEEE Access, 6, 29885-
29898.
[3] Shiddieqy, H. A., Hariadi, F. I., & Adiono, T.
(2017, October). Implementation of deep-
learning based image classification on single
board computer. In 2017 International
Symposium on Electronics and Smart
Devices(ISESD) (pp. 133-137). IEEE .
[4] Ertam, F., & Aydın, G. (2017, October). Data
classification with deep learning using
Tensorflow. In 2017 International Conference on
Computer Science and Engineering (UBMK)(pp.
755-758). IEEE
[5] Yuan, L., Qu, Z., Zhao, Y., Zhang, H., & Nian,
Q. (2017, March). A convolutional neural
network based on TensorFlow for face
recognition. In 2017 IEEE 2nd Advanced
Information Technology, Electronic and
Automation Control Conference (IAEAC) (pp.
525-529).IEEE.

IJCRT1134596 International Journal of Creative Research Thoughts (IJCRT) www.ijcrt.org 87