How to start with Machine Learning and Neural Networks

Share

Many people want to start their studies in neural networks and machine learning as a whole. So I decided to make a guide that I’m using to study these two technologies.

First, you have to choose a language. I chose Python.
Python can be downloaded through the Anaconda distribution, which in addition to packing in a functional way, still has a control panel for installations from other libraries: https://www.anaconda.com/download.

Now, you’re going to need an IDE. Do I currently use Visual Studio Code https://code.visualstudio.com.

Don’t know Python?

Follow the Basic Tutorial category at https://www.tutorialspoint.com/python/index.htm. It’s a start. There are many good courses in Coursera, Udemy, and EDX.

After doing the tutorial, some libraries are a “must  have” for machine learning:

-Numpy, library for arrays and mathematical functions: https://pypi.python.org/pypi/numpy-Matplotlib

– For plotting graphics and viewing data: https://matplotlib.org/downloads.html

-OpenCV, for viewing and editing images via Python: https://opencv.org/releases.html

For those who want to deal with classic machine learning, we have:

-Scikit-learn, Python library with all sorts of algorithms: http://scikit-learn.org/stable/install.html

-Weka, application with a graphical interface for reading data, preprocessing and machine learning algorithms: https://www.cs.waikato.ac.nz/ml/weka/downloading.html .

For those who plan to deal with neural / deep learning networks, it’s a different track.

There are four major frameworks: TensorFlow, Keras, PyTorch, and Theano. I use TensorFlow.
Do you want to understand how these initial networks work with a visual explanation?

The TensorFlow has a playground for you: http://playground.tensorflow.org .

After reading this material, it’s time to install the TensorFlow. His installation is a little nauseous, so READ. It will prevent further headaches, but in a summary it is:

-Install the CUDA Toolkit, and check that the system variables are correct (check even, there is a chance that you would not install it correctly)

-Install the CUDA Toolkit drivers-Install the cuDNN-Install TensorFlow, version CPU, or GPU (preferably have only one installation).

When you install, follow the step-a-step of the TensorFlow itself.  https://www.tensorflow.org/install

Installed? Tested?  Now you don’t know where to start?

TensorFlow himself has good tutorials to start dealing with him. Recommend two tutorials:

-A MNIST-based tutorial, a historical basis of handwritten digits: https://www.tensorflow.org/tutorials/layers

-A tutorial for CIFAR-10, a historical base of 60,000 images of 10 different types: https://www.tensorflow.org/tutorials/deep_cnn

You want more tutorials? It also has: https://www.tensorflow.org/tutorials. Learn how to use the TensorBoard, manager and visualizer of TensorFlow’s neural networks.

Until you save the current state of the network to reload then you can: https://www.tensorflow.org/progra…/summaries_and_tensorboard http://ischlag.github.io/2016/06/04/how-to-use-tensorboard.

Do you want a site with historical and classic databases?

Access the UCI Machine Learning Repository: https://archive.ics.uci.edu/ml

Do you want a site with current and complex databases?

Create an account on Kaggle: http://www.kaggle.com

 

Do you want a list of datasets with the current state of art and other applications for these bases (including the MNIST and CIFAR-10)?

You have it here: http://rodrigob.github.io/are_we_there_yet/build/

Do you find a base you want to work on?

You want to know how people are solving a particular problem? So get ready to read papers, get ready to read LOTS OF PAPERS, and they’ll probably be posted here: https://arxiv.org.

Have you got any questions about how some network works?

Probable that the Siraj Raval has already explained: https://www.youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A

If you want any more reliable sites for explanations of any network / architectural/problem solution that you have, I recommend the O’ Reilly Media (https: //www.oreilly.com/) and Medium (https: //medium.com/)

It also has the following list:

http://ai.google/education-Google course

https://see.stanford.edu/course/cs229-Classic CS229, milestone in the area, but quite extensive.

https://br.udacity.com/courses/machine-learning

-There are several courses in Udacity, beyond the nanodgrees, some paid, other free, it’s even difficult to choose which to study if you have any other questions about what or how to search, remember: Google Is Your Friend.

Thanks to Ayrton Denner for this guide.