Machine Learning (ML) is the word for the moment. Just like SOA has already been, just like DevOps has already been. Now everybody talk about ML. Everybody want to get its miraculous benefits but just few of them all really know what it is, its possibilities, and the shortest path to start getting its results and leveraging with it.
What is ML?
Wikipedia says “Machine learning is a field of computer science that gives computer systems the ability to “learn””.
Ok, got that. But what is Machine Learning? What does it can achieve? Let’s get to some examples:
Netflix told, during Cassandra summit on 2016, that everything it shows are recommendations. The slides are available here (there’s also a Youtube video at the end). And people like it! 80% of what we watch are recommended. We don’t actively look for it. It’s one click far.
They got 5.2 million new users from may to june in 2017. If you were on of them, you didn’t fill a long question set giving them information about which genders you like the most, not even what you don’t. Also, they don’t have the largest number of interns in the world to read 5.2 new user’s data and humanly suggest movies for them to watch. Those recommendations were made and keep always being made and updated within few minutes after you show any kind of sign you like something. Watched something? Flagged as “watch later”? Flagged as “like” or “dislike”? It all counts. It all updates your profile and new suggestions may come.
Now take a look at this second example. This Google’s Street View image shows a parking lot in Primeiro de Março Street, at São Leopoldo (south Brazil small city):
I’m 100% sure the parking lot named “estacionamento beto” didn’t send their data to Google Maps team. They didn’t fill a form telling they are at São Leopoldo. Their owner probably have a mobile phone with internet access. But probably they didn’t even know it is possible, and maybe they didn’t even know what is Google Maps. But they are there:
The Machine Learning here is something already touchable using a Google API to analyze images. They analyzed the Street View image, and the machine knew it was something with commercial purposes. Then the algorithm automatically “registered” it as something to be shown in the map. If you look at it at Street View, you will see that this street block is almost entirely residential. Google’s algorithms knew what’s the different between residential buildings and commercial buildings.
- Facebook suggests you to tag friends at their faces when you upload a new photo. It’s Machine Learning. Facebook doesn’t have one picture of each people in world to compare when they receive new pictures;
- Almost all e-commerces we know nowadays do that, but Amazon started: products suggestion based on the last purchases you done;
- Getting information from your bank account straight on Facebook’s chat. It can recognize many different questions for “whats my balance?”
How to benefit?
It’s very easy to think on new scenarios, more or less complex. Automatizing predictions or anything that today requires a human mind, is, or will be, possible.
Example! We know our stock market has patterns on its behavior and it’s affected by many things. Today many people spend their whole day (or life) analyzing data and giving predictions to others about what to do with their stocks (buy/sell orders).
The red dots (mine) on data above (from tradingeconomis.com) show moments when the market started cycles of loss over the last 10 years. During all of these examples we already had a set of technical analysis that helped us to identify when it’s more or less likely to happen again. But it (still) takes a lot of people distributed across many companies to MAYBE find something significant out. Having a good Machine Learning model can change this entire market. It can change into something unfair to put money, since parts of the analysis MAY become predictable.
But how to start?
A good new Machine Learning model invariably goes through a huge data quantity. The data is required to train the model and then achieve the real business goal. A good start is to set up a structured base of data to be used. You will also need a data scientist. Don’t worry, this scientist is not an alien. They are potentially very easy to find. But it’s subject for a new post.