How to become a machine learning engineer?

ai artificial intelligence ml machine learning vector by kohb gettyimages 1146634284 100817775 large Custom

 

The first thing you’ll need to learn is the roles in the real world.

Actually, the first thing you need to do is ignore most of the advice on this thread. Not only is most of it shit, I doubt a single poster has ever worked in the real world as a machine learning engineer.

I’ll say it a few times so it sinks it. We machine learning engineers are programmers, not mathematicians.

main qimg 5bfd20585f3d352c408b25cfb380e99a

More regulated math lies.

main qimg c4d236d26d3bb38f0b08d7c73c96d521

Ok, moving on from the shitty advice.

There’s NO SUCH THING as an AI Engineer. It’s a fabricated title.

Next up… the top job in all of AI is the machine learning engineer. It’s been at the top for three years now.

A machine learning engineer is a programmer with incredible data skills.

Here’s what one of the world’s greatest researchers has to say.

Ng observes that 80% of the AI developer’s time is spent on data preparation.

What that means is, most of the job of a machine learning engineer is data cleansing and sourcing, NOT MODELING.

More from Ng.

“The model and the code for many applications are basically a solved problem,” says Ng. “Now that the models have advanced to a certain point, we got to make the data work as well.”

Ok, now you know you want to be a machine learning engineer. Your next question is… what’s involved.

Here you go:

What do you need to learn and why?

  1. SQL. Why? Because all machine learning models are fed data. The cleaner your data the more accurate your model. Most machine learning models right now are based on data from relational data stores. All relational data stores speak SQL. You’ll need to write the Query to extract that data for modeling. Real-world data is dirty and working with it sucks. If you don’t have SQL skills you’re doomed.
  2. Python. Why? Python has become the gold standard for machine learning in the real-world. It’s the language you’ll use for just about everything once your data is amalgamated in a data store or CSV file.
  3. Libraries. Why and what are they? A library is code that’s pre-bundled. You simply have to import the library to use all the functionality. Libraries are as important as Python. If you don’t know the libraries you won’t get a job. There are libraries for just about every facet to machine learning. You don’t have to know all the libraries, but there are a core group you’ll use every day. These you must know. Pandas, Scikit-Learn, XGBoost, matplotlib… a few more.
  4. Data Cleansing. Why and what is it? Data cleansing are all the tasks you’ll need to learn to prepare your data for modeling. Remember when I said data was dirty? Extracting your data is the first part when working with data. The fun part is data cleansing. It’s what you’ll spend more of your time doing. I promise.
  5. Applied Statistics. Why? The process of working with a dataset and developing a predictive model is also a task in statistics. Although statistics is a large field with many esoteric theories and findings, the nuts and bolts tools and notations taken from the field are required for machine learning practitioners. You’ll use those libraries we talked about above, mostly Scikit-Learn, to apply statistical techniques to your data. That part of data cleansing ,knowing what statistical techniques to apply when and why. Notice the world applied? Applied means just that, applying the statistical techniques that are already written. You don’t do any statistical calculation yourself. They are written inside of the libraries.
  6. Modeling. Why? A model is simply a computer algorithm or group of them that look for patterns in your data. When they find a pattern they return and output. We call them models. There are different types of models for different types of problems. For example, the most common type of task in the real-world is classification. A group of ensemble models excel at this task. They are called gradient boosters. When you’re working with structured data and your problem is a classification task, you’ll want to use a gradient boosters. There are two core types of models. They are called traditional models and deep learning models and each has their own use cases.
  7. Tuning. What is it? Tuning a model means making tweaks to parameters called hyperparameters. The default model might not provide you with the best result. So, you’ll need to know what hyperparameters to tune for a given model. Yes, different models will have different hyperparameters. However, since you really won’t be using that many models in the real-world, learning them isn’t that difficult and the default model is pretty well tuned.
  8. Production. Why? The only reason you’re going through all this work is so that you can produce a model that will make a prediction on fresh data. Fresh data is data the model has never seen. This is the end goal of all models. Now, how a model is put in production will depend on the type of model. If the model is real-time, a front end will need to be created so it can be consumed. If the model is batch, then you’ll need to create that process to output the file to a location where the results can be consumed.
  9. Niches. Why? There are many different subfield you can move into after you have a foundation in machine learning. For example, there are people who only study NLP or Natural Language Process. There are others that only study computer vision. Yet others study reinforcement learning. I focus solely on gradient boosters and ensemble models because they are the gold standard for regression and classification, the most seen problems in the real-worlds.

So, where do you learn it? Here’s the only platform out there created by real-world machine learning engineers.

Now, some real-world advice. Machine learning engineers are some of the most technical people in all of IT. You don’t get there overnight. You’ll need to set some realistic expectations or you’ll be working with the rest of your data science buddies at Starbucks.

Companies aren’t handing 250K salaries for easy. If it were that easy, there wouldn’t be hundreds of thousands of jobs open worldwide that aren’t being filled.

All these super smart college kids would be taking them. Pssst… that’s not happening and it won’t.

In conclusion. The jobs are there and they aren’t going away. Actually, they are increasing rapidly. Most won’t make it. If you think you have what it takes I’d suggest you start reading this thread to understand what real-world machine learning looks like.

Leave a Reply

Your email address will not be published.