reading time
Published at 11 / marzo / 2020

Machine learning with TensorFlow in Python | BETWEEN Technology

The furore over machine learning has put TensorFlow, Google's open source software for machine learning projects, on the crest of the wave, especially when combined with the Python programming language. The last report The State of the Octoverse from GitHub highlights that TensorFlow has been one of the projects that have recorded the most movement on the coding platform in 2019. 9,900 users contributed to its growth by asking questions and issues and, to date, it has opened 46,000 repositories using TensorFlow, like those in the NumPy or PyTest libraries.

What is TensorFlow and what is it used for?

TensorFlow is an open source platform for graph-based numerical computing. It has the Google stamp and allows the development of machine learning applications with relative ease. These applications will provide us with predictions, ratings or recommendations based on the analysis of a data set. As the system collects more and more information, it learns automatically and the reliability of its results increases. TensorFlow can run on single or multiple CPUs and GPUs as well as mobile devices.

TensorFlow has multiple uses. One of the most important is the development of artificial neural networks used in:

  • Automatic image recognition and classification, identifying the people and objects in the images.
  • Understanding audio signals and voice searches (very much in vogue with the rise of smart speakers).
  • Time series analysis to evaluate behaviours and forecast what will happen next. Very useful in marketing to offer product recommendations to potential customers or to anticipate service cancellations; in health to prevent pathologies; or in insurance, to calculate the risk of the policyholder experiencing an incident.

Machine Learning Projects: Why TensorFlow in Python?

Combining TensorFlow and Python is one of the most effective and simple formulas for developing machine learning applications. Why?

Advantages of Python in machine learning

Python will be one of the most widely used programming languages in the year 2020. It is the second most popular in the latest Developer Survey from Stack Overflow and third in the TIOBE index. Using Python in machine learning offers great advantages because:

  • It is user-friendly, easy to use and learn. Ideal for beginners in the development sector.
  • It is a full stack language, suitable for programming from both the server and the client side.
  • The Python Package Manager, PyPi, already has more than 10,000 packages for scientific use available to users.

codigo-lenguaje-programacion-python

Advantages of TensorFlow for machine learning projects

For TensorFlow, it is worth highlighting:

  • Its interoperability. The same model of machine learning written in one language can be easily extrapolated to another.
  • The large number of models already developed by the TensorFlow community which means that you don't have to start from scratch to build your own.
  • The high volume of resources offered by Google, in the form of tutorials or ready-made neural networks, which help make learning easy.
  • Its versatility. It is a generic computer library, not only focused on deep learning , as is the case with Keras or PyTorch. It is suitable for small projects as well as for complex and major challenges.

Key concepts in machine learning projects with TensorFlow

When developing machine learning projects with TensorFlow in Python, you need to become familiar with some key concepts such as:

  • Tensors: algebraic structures that contain different numerical values and through which data will flow in our system.
  • Models: functions that aim to obtain a certain result from some input data.
  • Periods: number of times the model receives the same input data during its training process
  • Loss function: calculates the reliability of the result obtained in each iteration.

In short, we could say that TensorFlow aims to develop a model of machine learning that, with an input and training data set, it is possible to set generalizations. Before the data set is added, it must be pre-processed so that the system can get the most out of it. Then, for the training, enough periods must be forecast, so that the model learns; yet not too many, as this could jeopardise the results.

In machine learning projects with TensorFlow in Python it is possible to use additional libraries that will make programming easier. For example, NumPy and SciPy for mathematical functions; Pandas and Seaborn, for statistical use and data visualization; or Matplotlib to generate graphics.

 


 

Tags: Informática

Related Posts

What is a sandbox and what is it for? | BETWEEN

If it weren't for sandboxing, doing something as simple as reading a pdf, downloading an attachment, or surfing the Internet would be as dangerous as driving through a minefield. ...

( reading time )

Topics: Informática

Being a WordPress developer in 2020 | BETWEEN

Enter your browser and visit any Internet page. What CMS (Content Management System) would you bet it is made with? In BETWEEN we give you a trick: answer WordPress and you will ...

( reading time )

Topics: Informática

Serverless: what is it and what are the advantages | BETWEEN

Serverless architecture, based on the execution of applications and processes in the cloud, is here to stay. Alcide's latest report The State of Securing Workloads has found that ...

( reading time )

Topics: Informática