5 questions most people have before starting out their journey in machine learning

Along the journey, I found the answer or at least it seems to be as in the world of learning no one has the perfect answer.


Akash Desarda

3 years ago | 4 min read

When I begin my journey in the field of Data Science/Machine Learning I felt quite uncertain about my path or more specifically about my methodology and found keep asking this question.

Along the journey, I found the answer or at least it seems to be as in the world of learning no one has the perfect answer.

1. The whole world is talking about it, why do I have to care?

The world keeps evolving itself. All those things which are rigid, unfit, becomes obsolete over time and eventually perish. The same concept counts everywhere even in areas like business, computer science or its combination ‘IT Sector’.

As we are generating an unprecedented amount of data, it has changed our mindset. Earlier people used to make decisions based on their experience. This used to work on a small use case. But today in the world of globalization, this simply fails.

You have to back your decision by something concrete, i.e. data (which we have in abundance). Because ‘Data never lie’. It is not like this field is new. But it is correct that its importance has grown exponentially due to again availability of immense data. Also, there are lots of opportunities.

India has the most number of data analytics jobs after the US

Study: Analytics And Data Science Jobs In India 2018 — By Edvancer & AIM2

2. Ok, even It seems that I am interested so what are prerequisites needed?

Like humans have three basic needs of food, clothing, shelter it also has three basics need of coding, maths/stats, instinct.

Let’s check them out…

i) Coding:- Coding is just a tool, but a very essential tool. Python is generally the goto language due to its simplicity, OOP, platform independence, libraries(probably the most important point). Coding can be found difficult at the beginning but anyone can master it with time and practise.

Here’s why you should prefer python

ii) Maths/Stats:- Make no mistake these two things are like the soul of each and every ML algorithm. Every single ML algorithm is based on some Maths function and every business decision is based on stats. (We’ll discuss maths/stats more ahead).

Here’s why you should understand maths.

iii) Instinct:- This can be more called as a domain knowledge sort of things. Don’t worry if you don’t have it at the beginning (actually no one has it at the start). It comes with time and experience.

But why it is really so important? Here’s because even once you have a good working model you should able to make the most out of it by taking a profitable decision.

Insights, graphs, suggestion, prediction, etc are still machine generated vector it might be accurate but not right. It still needs human touch or human instinct.

3. Online material looks so overwhelming, so confusing, don’t know what to do?

That’s true. We live in the 21st century, ‘The Internet century’ (also ‘the MOOC Decade). We have everything at our fingertip but still finds it really difficult to select something. Here are some tips which I found useful

i) Before starting anything first make sure they are covering the basics if you are starting from scratch.

ii) Somehow if you manage to select one then please try to stick to it. Do not judge it too soon. (This was one particular big mistake used to commit regularly)

iii) Always prefer a book along any with any MOOC (I will share list some good book)

iv) Most important at any given time focus on only one algorithm or one concept, master it thoroughly, get to know it in depth, practise it & then move to next. Do not leave it on a cliffhanger if it seems hard and start with other. In the end, I will share some useful resource.

4. Why do I have to understand maths, stats in depth when Scikit learn do all the heavy lifting?

As I told earlier that every single ML algorithm is based on some Maths function and every business decision is based on stats. It’s true that the python library like Scikit learn has made the process really easy and pretty much automated.

Almost every algorithm follows the same blueprint in terms of coding which even school kid can perform if he is able to memorise it.

But to make the model more accurate according to our needs we have to make changes in our hyper-parameters or as the people like to call it ‘Hyper-Parameter Tuning’ you must understand the math function behind it as you cannot make changes randomly and think that you’ll hit Bull’s eye in the dark.

Let us understand this with an example. Let us assume that you know how to drive a car and know nothing about it parts (not engineering aspect, just general like an avg person/mechanic).

If your car ran into a problem (however small it may) you have to call an expert each and every time. You will still able to drive but you will never be called an expert Driver. Similarly, without maths knowledge, you may survive coding, but you can never become an expert.

Pro tip: Try not to learn every maths concept same time. Only learn that maths concept which is related to your current algorithm. This will keep your concept more fresh and crisp, also it won’t burnout you.

5. It looks like I am ready to begin the journey but couldn’t find the right track?

The important thing is to keep your self is motivated enough throughout the journey. Consider it as your fuel and learning as your vehicle. Find something which will keep you motivated throughout the journey. Following is the list of useful resources


Created by

Akash Desarda







Related Articles