The real issue of Machine Learning
Studying Machine Learning and Deep Learning and trying to understand all the underlying aspects of it, such as the mathematics behind it and the coding part, I came very quickly across a wall, once I started doing Kaggle on my own.
I also found a really good video regarding the fundamentals of Deep Learning, which I highly recommend as it's so easy to follow and understand, on Alexander Amini's YouTube channel. Alexander is a professor at MIT and he teaches Introduction to Deep Learning, a course that lasts only one week. But you should definitely check this one out:
The challenge is quite simple: you're given a data set that contains details from passengers, such as the port they boarded the Titanic, ticket cost, ticket number, class they were travelling, their gender, age and most important, if they survived or not. As this is the question that our model needs to answer. And that's the tricky part.
Once you get a good grasp of the mathematics and the coding behind ML/DL models, you'll have to work with data. Real data, preferably, solving real issues. And these to requires a few questions to be answered, such as what do I want/need to find out? How do we get the data? How do we make sure it's relevant data? Which features do we select? Is this a classification or regression issue?
Features it's another name (in ML/DL terminology) for the data input you'll be working with. For instance, in Titanic Dataset's case, important features might be their gender, class and whether the person survived or not. Not useful would be details such as the port where they boarded or how much the ticket cost. The latter details won't help you at all deciding if someone did survive.
As a Data Scientist, a ML/DL enthusiast, model creator, let's say for instance, you have to develop a model for a cooling system for servers that we'll have to inform you when one of the components might fail. You'll have to work close to the engineers who designed the system, understand how it works, which are the most important parts to be added from the data collected (the features), and, again what is the question your model will answer.
Then you'll have to start collecting the data, develop the collecting system, implement the data in your model, choose the right algorithm, bias, etc. and measure its performance.
This is what I found the real issue behind this, asking the right questions regarding the issue at hand. The mathematics and coding can be taught, understood and learned how to apply in different case scenarios, but you'll need to develop an intuition around Machine Learning models, which algorithm to choose, but, most importantly, is Machine Learning a solution to the problem?
I also found a really good video regarding the fundamentals of Deep Learning, which I highly recommend as it's so easy to follow and understand, on Alexander Amini's YouTube channel. Alexander is a professor at MIT and he teaches Introduction to Deep Learning, a course that lasts only one week. But you should definitely check this one out:
Now, after I got passed a few tutorials, read a few books on this topic, I decided to get my feet wet, but in a real way into this matter. So I chose Kaggle's Titanic Challenge -> https://www.kaggle.com/c/titanic
Once you get a good grasp of the mathematics and the coding behind ML/DL models, you'll have to work with data. Real data, preferably, solving real issues. And these to requires a few questions to be answered, such as what do I want/need to find out? How do we get the data? How do we make sure it's relevant data? Which features do we select? Is this a classification or regression issue?
Features it's another name (in ML/DL terminology) for the data input you'll be working with. For instance, in Titanic Dataset's case, important features might be their gender, class and whether the person survived or not. Not useful would be details such as the port where they boarded or how much the ticket cost. The latter details won't help you at all deciding if someone did survive.
Then you'll have to start collecting the data, develop the collecting system, implement the data in your model, choose the right algorithm, bias, etc. and measure its performance.
This is what I found the real issue behind this, asking the right questions regarding the issue at hand. The mathematics and coding can be taught, understood and learned how to apply in different case scenarios, but you'll need to develop an intuition around Machine Learning models, which algorithm to choose, but, most importantly, is Machine Learning a solution to the problem?
Comments
Post a Comment