 Hello everyone and welcome to another episode of Code Emporium where we're going to answer some of the most googled questions on machine learning. I'm sure a lot of you are pretty curious about what are the most asked questions in this world in machine learning. And today we're going to find out. Do note that I'm going to be answering these questions as a data scientist working in an industry setting and less so as an academic. But I still hope that my responses still help. Before we get started with the video please do give this a like because the more you like the more other people can see the video and the more they like other people can see the video and the process goes on. Also do check out our Discord server down in the description below we are having some amazing conversations there and we would love for you to be a part of the community and join us too. And with that let's get started with some questions. Does machine learning require coding? Yes, yes it does. But how much coding really depends on the type of company that you're in. There are certain data science roles that are really code heavy and they really want people who can code almost like software engineers. But on the other end there may be some companies where you'll be really heavy on analytics, analysis and probably just certain amount of code that you need to build a model. That said regardless of the company you do join something that you will be coding a lot is SQL. Although it's not a traditional programming language it is extremely useful as a language in the data science pipeline and you'll be using this in your job on the daily. Is machine learning AI? So machine learning is technically a subset of AI. All AI is not necessarily very predictive as is machine learning. A good example of an application that is AI and not necessarily machine learning could be like a cell phone automated dial system. Press one for sales, press two for a representative, press pound to repeat this message. There's really no predictive capabilities there but it is an automation and it is an AI on its own. Why machine learning is important? For simple cases like that phone prompt where we have like a fixed number of definite inputs and a fixed set of definite outputs we really don't need machine learning but there are situations where even this application can become pretty complicated. So for example if we start introducing voice as input the inputs now can be extremely different almost an infinite number of ways that we could actually make responses with a voice and to detect that machine learning can better be used and leveraged. And in general it's for situations like this where we need more complex pattern recognition to map a large set of inputs to a large set of outputs that is not necessarily a one is to one dictionary mapping. In situations like this machine learning can be very useful. Where is machine learning used? Machine learning can be used anywhere where we can use past experiences to make predictions about the future. Like for example in an e-commerce setting we could use machine learning to predict the number of orders that might happen 10 days from now since we already historically know purchase history. However you need to make sure that you just don't throw machine learning around especially in situations where a SQL query can get you 99% of the way there. How machine learning works? So machine learning typically works by understanding patterns in data. So if you kind of do look into the internals of many of these models specifically like supervised machine learning models you have a set of inputs you have a set of outputs and you're trying to basically use machine learning to create a mapping between inputs and outputs and this mapping is essentially what we call functions in mathematics. So machine learning is really doing is it's trying to learn some very complicated or could be simple function that you just don't know but you have observed data as a result of that function and it's using machine learning in this way that we can power very large and cool applications and also when you're working with machine learning in terms of code a lot of the time you don't really need to know the internals of the function or the model itself but rather just the kind of inputs you need and the kind of outputs you want. Does machine learning require math? So this is kind of an extension of the previous question where if you boil down a machine learning model it is a lot about mathematics and functions and in order to actually understand how machine learning models actually learn you will need some amount of math including that dreaded calculus but when you're working in the industry you might not need to know so many of those internals offhand but you do need a little bit of math to at least understand how to interpret these models like for example if we wanted to interpret models with chaplain values although you don't obviously need to compute chaplain values yourself by hand it is very useful to understand how they are computed so you can better explain in a very non-technical sense to certain stakeholders of what those chaplain values actually mean. Is machine learning hard? Machine learning does tend to have a very large knowledge curve especially when you are looking for an entry-level job it does seem like you need to wear many hats in order to actually perform the role of a machine learning engineer or a data scientist. This is true for several reasons first is that machine learning or data science on its own some of the core fundamental concepts are not completely taught in school or even in college whereas maybe there are other fields including like software engineering where you do have a pretty good coding foundation in school and it's easier to build on top of at least to get the initial job but honestly at a higher level machine learning software engineering some of these other fields are kind of relatively same on the difficulty level but I do understand that getting that initial job can be a little tough and that's why it it does come across as pretty hard. There definitely is an oversaturation in the market for a lot of people who know a little bit about machine learning and too few people who know a sizable amount especially given how research oriented this field is. You always need to be on that grind to learn something new and also figure out how you can apply it to your work. It's fun but it can be challenging. Why machine learning is the future? Machine learning is a mode for automation and automation helps makes the lives of everybody around you much easier that said a lot of companies are unfortunately not aware of how data can help them make their lives a lot easier and it's our duty as data scientists to communicate the benefits of data. If we're able to do this in a very seamless way then machine learning will percolate not only in academia but also in industry and that's when we know that it can affect the lives of everyday people rather than just stay in research papers. I do think we're headed in the right direction but we do have a lot of work ahead of us. Where is machine learning not used? Machine learning isn't really used in situations where past experience cannot make great inferences on the future. A good example of this is anytime where the data itself is not very clean not very great so you can't really have a great model to make predictions off of you're probably better off with a rule-based system. Another situation is certain examples where we just don't have enough information to make reliable predictions for example in the stock market we don't have as common people at least we don't have enough information to make accurate point-to-point estimates of how much a stock is going to vary a day. And the third situation which machine learning is not used is in cases where a rule-based system is just more than enough where machine learning is basically a mapping of functions between inputs and outputs. If the system is super simple and we already know what the mapping is we don't need machine learning because why would you want to learn something you already know and just increase the fuzziness of that prediction? Doesn't make sense. Why do machine learning projects fail? This is an excellent question and is a really hard topic in data science to talk about because it is the source of burnout and also frustration of being a data scientist. So let's consider a situation where everything usually starts out with a hunch. If you think that you have an idea of something that could potentially help you do some data analysis and then you realize two things one is that your idea is actually really good backed by data and the other is that it's not really good and there's no opportunity moving forward. If there's no opportunity moving forward you don't waste time you scrap the idea and move to another project but on the off chance that you say oh wow this is actually a pretty good idea you move to phase two which is the actual machine learning and modeling phase now let's say that you have this amazing model that works well you've spent a lot of time on it and you present this model to leadership now leadership can say one of two things one this model is super good let's use it or two now we'll shut it down now it's at this stage that if the model is shut down you've spent so much time effort and energy on this it can be a little frustrating when you see it actually just not be used at all or shut down and this is the primary reason why data scientists say well I spent so much effort on this but clearly it has no impact but let's also say that leadership says hey this is a great model let's go there might be some other dependencies that are from engineering or from other sectors of the company that you need to work with but they have blockers that are just unavoidable and so even though there's nothing wrong with what you did on your side your project will be canned and potentially you won't be able to move forward no matter what now clearly there are some forces that are beyond your control at play here but the very least that we can do to minimize this is to maximize communication try to talk to different stakeholders talk to leadership see where and how this model is used try to establish the dependencies for this model before you try to get way too deep into it and if you do follow these guidelines at least you'll save yourself a lot of time and a lot of headache getting into the modeling process machine learning projects do fail and we should talk about this more but for now that's all I have I hope you all enjoyed these machine learning questions from Google are these questions that you wanted to answer do you have other questions that you wanted to know about please do comment in the description down below or also on our discord server which is in the link in the description down below so please do join us give this video a like and we will see each other very soon take care