Machine learning for grandmothers

Computers have completely transformed the world, but this transformation would not have been possible without the development of the algorithm. At the heart of every task or function a computer performs is an algorithm organizing or processing information to make this something useful.

However, the algorithms used so far had to be fully specified and usually did not vary in their execution, to adjust them they had to be rewritten again.

This is why a new type of algorithms emerged that make up what we know today as machine learning, which do not require a total knowledge of all the parameters and variables involved to have a good performance, instead require certain conditions of the data that is they enter him, for this reason they are divided into two categories mainly: supervised and unsupervised.

A supervised algorithm is is one in which the input data is required to be labeled with the expected outputs of these, so that through mathematical operations you can approximate and adjust the parameters or weights within the algorithm and finally obtain the expected outputs having only the input data.

However, these types of algorithms can also be divided into two types depending on the type of tags that the data has, such as classification and regression.

The classification algorithms are those that are designed to fit labels in categories, such as an algorithm to identify the color of an object between: white, green, blue, etc. having only the image pixel data.

On the other hand, regression algorithms are those that are designed to adjust data that have continuous or discrete numerical values as labels, in which the outputs depend on one or several variables such as: an algorithm to predict the stock’s price.

An unsupervised algorithm is one in which the input data is not labeled, so it is up to the algorithm to adjust the commonalities that it considers to classify the input data into some category. An example of this type of algorithm is market segmentation to classify the types of customers a business has.

An advantage of this type of algorithm is that, of course, the data is not usually labeled, therefore, it is usually ideal for a first analysis, since the labeling process can be expensive in some situations.

They are a type of algorithms where you learn through a behavioral efoque, in which the algorithm tries to accumulate the greatest amount of reward.

These are some examples of machine learning algorithms:

Linear Regression:

This most simple machine learning algorithm that tries to adjust a linear relationship by minimizing a cost function, the most used method being least squares. Although the initial concept seeks to approximate a linear relationship between the input data and the output data, it can also be adjusted to non-linear data by including the cross terms.

k-nearest neighbor:

This algorithm works like a classic Colombian saying: “Dime con quién andas y te diré quién eres” or its equivalent in english “ hunt with cats and you catch only rats” which consists in judging an element by the closest elements, in which the element to be judged is finally defined according to the majority feature among its closest elements. This algorithm receives as a parameter the number of closest elements for the classification of an element, for example: As in the image above if the parameter k = 3, we could define the green circle (undefined) as a red triangle, while if the parameter is k = 5, we would define the element as a blue square.

K-Means Clustering:

It is one of the most common unsupervised algorithms and groups the input data according to some criteria in k groups, where k is a parameter that must be specified. For example, in the image above, the input data did not have a classification, but after specifying k = 3 the algorithm classified them into the three groups that can be seen in different colors.

Deep Learning:

Deep learning is a specific part of machine learning where you try to emulate the functioning of the human brain, through the implementation of neural networks.

Neural networks are multiple layers where, according to the input data, each of the variables in each layer accumulates a greater weight of importance, so that it fits better to all the output data.

Related posts:

ada sebuah gelas yang tidak pecah, kau panaskan di terik matahari pun tidak retak atau memuai kacanya, kau jatuhkan ia dari atap apartemen Puncak Kertajaya pun tidak akan terbelah partikelnya…

What if instead of fretting over the goal, you might never reach, you think about creating a process that no only will help you to achieve the goal, but also in case you lose, make it worth your…

And now we need to convince ourselves that this code does what it claims to do. Happily, Clojure comes equipped with a simple and capable library for writing traditional unit tests: clojure.test. In…