3 Wrong ways to learn machine learning

Machine Learning is a process to Iteratively learn from big data to make better predictions. It is a convergence of Computer Science, Math, and Statistics and so it can be very easy to get deep in one of the subfields while neglecting other fields, but to be good one must be balanced in all the subfields. A common reason for this lack of balance is the wrong ways beginners approach Machine learning.  In this Post I share 3 wrong ways to view developing skills in machine learning.

1.      The first is Underestimating the role of programming languages in machine learning deployments. Even though tools exist today in the ML ecosystem that have made deployments easier, programming is a core skill required by ML engineers. Beginners have some appreciation of the use of programming in the ML deployment cycle, but they tend to underestimate the depth of its requirements.  Programming languages play a key role in ML so aiming at mastery level in commonly used languages such as python will ease at one’s journey. Also, understanding web technologies will be useful to deploying solutions.

2.      Another wrong way is the “follow the script approach” in which beginners follow an ML script to implement an ML algorithm. Whereas this method helps beginners to appreciate in an intuitive way what an ML does, it is important not to lose fact that all ML training script, for example, has data preprocessing stages, training, and model performance evaluation parts.  Identifying the core repetitive steps in the ML pipelines is a more useful skill than implementing a script that for example successfully classifies a dog from a cat.  

3.      The third problem is focusing on how packages work instead of understanding how they fit into the ML processing chain. They are tons of ML packages and having an integrating view of them can be confusing. But for example, understanding that TensorFlow and Pytorch can both be used for data automation, model tracking, performance tracking and model training is more useful than knowing how it works although that is not to say knowing how they works is not important.  However, in addition to knowing how it works, know how it fits into the ML value chain.

In summary, to become good in ML development and deployment, first you will need to be good at a programming language, especially Python since most of the ML packages have implementation in that language. Secondly, understand common Statistics concepts such as but not limited to cross entropy, Root Mean Square Error (RMSE), Area under the curve (AUC), R squared, etc., which are used to evaluate a model performance. Additionally, pick a problem, recognize the nature of the ML problem and what other similar algorithms are candidate for this problem and contrast their strength and weaknesses. Iterate this for different ML problems and finally look out for knew ideas introduced into the ML algorithm the next time that you follow a script and see yourself grow. 

My name is Lovis Kwasi Armah and I am passionate about breaking complex machine learning concept in simple terms. Follow me and let's learn this exciting field together.