Entries by Jose Quesada

What do you want to achieve with your Machine Learning project

What do you want to achieve with your Machine Learning project?

Join our youtube channel to see more detailed content. 

By Jose Quesada

Hi, this is Jose Quesada for Data Science Retreat.

Today I’m going to talk to you about one thing that I’ve been thinking about for a while which is, what do you want to achieve with a project in AI or Machine Learning or Data Science. People all care about having projects having something to show. But there are different levels, this is something that I really want to go deeper.

Most people only want to learn a skill set with that project, which is fine.

But there are like different levels which we will discuss here and I would personally opt for the last option which is to create a product which you can charge for because if you fail you’re going to have kind of a fall back on one of the others.

I have made a video on this topic for those of you who prefer watching a video than reading, see below.


Play Video

Learning a skillset that’s a given. No matter what project you’re going to do, you are going to learn something, not a big deal. Now, one other goal would be to look competent and the real goal is to find a job. This is also very common “oh I’m doing this just because I want to put it on my LinkedIn and my github and I’m going to get the attention of recruiters or hiring managers. You want to get a job, that’s totally fine. So, there’s a difference between these two, right? It could be that, the thing that you do for running a skillset is not the same thing when you move to look higher now.

 

Now we’re getting into deeper things. So, you want to solve a real problem with your projects, that’s not very common. I would like it to be more common, but it’s not. I see people saying “oh I want to have a social impact, I want to help people, I want to solve your problems” but it’s very unlikely that you find people doing something about it with machine learning. And it’s totally doable, there’s plenty of things that you can do in machine learning today to solve a real problem that real people have. It’s kind of a shame that we are not doing more of it. So that would be one more goal that is kind of deeper than looking competent to get a job, right?

 

There is one more level which is to create a product. A product is not only a project it’s a product because you can charge for it. This is something that is packaged in a way that these people that have their problem and maybe can pay for it. They are willing to do something about it, that is a product.  Most people, I would say 99% of the people that I deal with today never get to the point where they take a project and build it to a point where they can charge for it. This is not a criticism, there’s a real reason for this. Traditional companies, they all want it, but they don’t get there mainly because of how much money and effort real big companies spend trying to figure out what will work and if they build it, how to market it and so on.

 

It’s totally fine, if you as an individual cannot get to this point and build something that you can charge for. But it’s a big part. All these companies spend a lot of money in doing market research, focus group, trying to get data from their users and using that data to sell them back something like netflix does. Netflix produces series that are influenced by the data they collect. Sometimes you see a series and you can see “oh this was targeted to teenage women who are overweight, and they would love this series”. How do they find out that there is a market that would want to consume that series because they have data and they know things that the person is watching? So, the big difference between you and these big corporations is that, they need to solve a problem that affects millions ideally billions of people.

 

That’s cool for you, example they will never start a new project if it doesn’t affect billions of people. So that limits their options quite a bit. They are kind of the king of the jungle for machine learning, there is no other company doing better with machine learning so far than google. They are probably good at products as well and you can question that because they give a lot of products right, but they have a problem which is they need to be products for billions of people. You don’t have that problem; you can go to a tiny little narrow group of people that you know very well and help them. You may know them so well that you may meet one of them who will identify with your idea and you might identify with their problems. It will be a beautiful match within the two of you.

 

So, here there are some ways to think if you are going to go for the product part of it where you want to create a product that you want to charge for. I’m not saying that you need to do that, but if you want to go that way then there are like different scales.

You pay People >>>>>>>>>>> They pay you

Sometimes you come up with an idea, and you want people to use it, but they actually don’t and give you reason such as I don’t have time now. Then you pay people to use it, or you promise them a favour. Imagine that your product is a questionnaire. You tell them if you fill this questionnaire then I’m going to give you a chance to win a Macbook. So to pay people to do something for you, that’s kind of one extreme.  The other extreme is where they pay you and they will be sad if your product doesn’t exist anymore. You want to be moving from you paying to they are paying you. I’m not saying that this is easy. This is something that companies big and small like startups as well struggles with right now.

You are undifferentiated >>>>>>>>> You are the first solution they think about

Sometimes you come up with an idea, and you want people to use it, but they actually don’t and give you reason such as I don’t have time now. Then you pay people to use it, or you promise them a favour. Imagine that your product is a questionnaire. You tell them if you fill this questionnaire then I’m going to give you a chance to win a Macbook. So to pay people to do something for you, that’s kind of one extreme.  The other extreme is where they pay you and they will be sad if your product doesn’t exist anymore. You want to be moving from you paying to they are paying you. I’m not saying that this is easy. This is something that companies big and small like startups as well struggles with right now.

You are guessing >>>>>>>>> you have validated the problem and solution

The last one is, you are consuming, you read a lot, you see blog posts about companies being built, about people making money with products maybe somebody just came up with something as simple as a newsletter.They’re making money with it, you read the newsletter you consume. You must aspire to go to the point where you produce, when you write. People should follow you because of what you write.

You consume, read >>>>>>>>> You produce, write

You can go from guessing the problem to be in a spot where you know the problem really well. You could be the person who knows about the problem the most in the world, and you have the solution. That is a wonderful place to be. So, you want to go in that direction.

What is that one hack to get all these scales much faster to move from one extreme to other?


It is

Let’s just look at the scales now, assume that you are in a niche. It’s easier to get people to pay you if you are targeting the right people and they have a problem that is very narrow to themselves. It’s a problem that’s worth paying for.

You’re probably the first solution to think about if you are the one person talking about this. Nobody else cares about it but you do, and you care for it very much. You know the problem very well, and you’re building a solution for that niche. You can be producing and writing about that niche and then it’s easy for people to follow you to find you. So basically, all these things become easier if you have a niche.

This is actually the big advantage that you have over google. You don’t need to create a product for a billionpeople, but a product that’s probably for few thousands. You’re going to be looking competent. So, like I mentioned earlier with this product, you’re going to learn a lot, solve your problem and you may be able to charge for it.

Therefore, the most important message I wanted to convey is that, it’s much easier for you in every possible way to aim to create a product that you could charge for and then degrade into solving a problem, finding job and lastly learning a skill. Companies are going to be impressed if you show up and show a product that you build with your hands that solves a problem in the market which you could charge for.

This is something that I would hire if I ever see in my doorstop, two data scientists with similar profiles but one of them have built a bankable project then he/she is way out of the league of anybody else and I would hire them.

To conclude “Aim to work on building a product rather than a mere project”

Is it too late for me to move into Data Science?

Is it too late for me to move into Data Science?

Join our youtube channel to see more detailed content. 

By Jose Quesada

Hi, this is Jose Quesada for Data Science Retreat. I want to cover a topic today that I find very interesting. Many people come to DSR, thinking that:

They are too late

They could have started earlier, and now they are late in their careers…

How are they going to make this happen…

They really want to be data scientists, but they think that they are maybe late at their late 30s or maybe even 40s..

So, I’m here to tell you that no, I don’t think it’s too late for you. I’m going to give you some reasons and maybe examples too. I have made a video on this topic for those of you who prefer watching a video than reading, see below.


Play Video

 

So, the first thing that you need to understand is that data science is totally a lateral move for everybody. It’s not like there is a career path that you go to university and get some kind of certificate (certificates in general of even master programs, I don’t see them working in the market).

 

I’ve interviewed many people with master’s in data science no, they didn’t pass the interview, and looking at the curriculums in their programs, they’re not going to do any better on their own, so that’s good news for you. That’s a level playing field.

 

If you’re coming from a different industry, so what? Everybody is doing the same. So that’s an advantage, and that comes with the territory when you are experienced. You have already produced something in your life that is valuable to somebody.


"Remember to produce something that is



to somebody"

So good news for you! You’re an experienced person you may have a leg up just because of your experience, so you may say yeah that is true, I agree but other people who are younger than me have advantages over me.

Let me tell you one more thing that I think could benefit you, which is your network. What happens when you are new in any task, you don’t know people in that particular domain then everything becomes much harder. But when you’ve been around in one industry, it doesn’t have to be data science you magically have other people that you can ask things, and every person in your network is

a potential source of problems, which is very important. Everything in the life of a data scientist is to find a good problem, where you can provide a solution with this new technology, which is data science.

How do you do that? It’s much harder with your experience when you haven’t seen much in the industry.

I’m sure that I cannot produce problems to say the golf industry because I’ve never played golf. But if you have 30 years of golf experience, then I’m sure you exactly know what a good solution for a problem that exists in golf is.

Photo by Courtney Cook on Unsplash

Let’s say that you look at style transfer or keypoint detection like the skeletons for your pose like dance pose, and you say wow, this would be fantastic to learn and doing something with golf. You come up with that idea, and it may be a great idea, I cannot because Idon’t have any experience there and I don’t have a network either. 

So the same moment you have the idea you can call three people and tell them the idea and people will be like ‘Oh my god I will definitely use that if you create a system so I can learn the x,x,y golf moves they do because I don’t even know the names.’

 

They are going to love you, and you can test that because you have a network. More so, you can go and sell your things to that network before you build them, which is the best validation possible. If this idea resonates with people and they tell you Yes, I would definitely use it and you tell them okay how about 30

bucks per month, and they say No. Then maybe the idea is not that valuable, maybe it’s not so powerful for them to solve this problem.

But you have a network, and you can use it to validate and fine-tune problems. This network is also going to be valuable to the company that hires you as a data scientist. You may not expect that, but imagine that you’re a medical doctor, and you are actually learning machine learning on the side, and you are

becoming good at this. So, you apply for a job at a company that is doing something with x-rays, and you are 40. Do you think that they’re going to look at you and say well this guy is 40, We don’t want this type of profiles? No, they’re going to look at the medical doctors, and they are in the domain of x-rays, so you are going to see things that they don’t see because they don’t have domainknowledge, but you do.

This is fantastic; you have domain knowledge, it’s going to open doors for you. So, make sure that both your network and your domain knowledge is valuable to the company that you’re joining, and you may say but Jose all companies that I’ve worked for in the past they were boring, and I didn’t learn anything there.

 

That’s not true you probably know more things than anybody else in a room that is trying to hire and this is a really good match because vertical AI companies are really really valuable when they have domain expertise domain knowledge that nobody else has, and it’s the mixture is really hard to find then so much more valuable the companies

 

So remember network domain expertise, and you must have some management experience because if you’ve been around for a while, you may go well, I actually have only three people in my team, but  I was the most senior of them so they looked up to me when we had to make decisions or maybe I could interface better with the decision-makers because I was kind of in the same age range or I knew them for a long time. This is valuable definitely play that up in your LinkedIn in your applications and describe how you solve problems because you had management experience. When you are interviewed, remember not all the interviews are super deeply technical some are really like oh this is not going to fit with your culture.

 

Also, you must have some experience period, not just management experience, but you must have done something that the left trace does.

 

So let’s say that you are an athlete, you were an athlete and you were swimming competitively for many years or even something more radical you were a martial artist and you may say well that experience is completely worthless to anybody trying to hire me as a data scientist. Well, No, you have experience of how it is to really try hard at getting something and failing multiple times. You have the experience of training and improving and measuring your performance. You have experience of trying different things and finding one thing that works better than the others and doubling down that. This is totally transferable to the things that you want to do as a data scientist. You may go in a direction that is something related to AB testing or bandits. That’s exactly the same use case that I have described.

 

 

But there are disadvantages to being experienced or being older. What are those?

You are more expensive than others. There’s no way around. You’re not going to get paid the same as a fresh out of university person. You may not be able to want to pull long week hours like 60 or 80 or

typical startup business hours. That is personal, but maybe you can do as much with fewer hours. That’s the issue of culture, so if anybody in the company is really of a particular mindset, being the elder outsider may not help you, and maybe this is the biggest showstopper.

You cannot really do much about that if you have been trained if you’ve been working for years in big corporations, maybe that you don’t fit straight in at a sloppy culture where things happen very fast, and people make jokes about things that you may not even understand. That is a problem, and there is no way around it unless you spend months really trying to enter that culture. I particularly love that to find a new culture and see how you can understand it better that is not only countries or big cultures like that, but it could be just the culture of people who love a board game that you’ve never played. I don’t play both games; I don’t know anything about this. They have a culture, right?

Can I understand their culture can I understand why they like what they like? So, this is a wonderful thing for me to do and to learn;it’s valuable anywhere you go. All this is fine, but there’s one more thing that makes a big difference, and that is “Business Value.” This is super important, and there is no way that you can tell me, ‘oh I’m more serious, I’m older now, and I don’t know how to evaluate business value better than when I was younger’ I don’t believe you. If anything, now that you’ve been around, you’ve seen projects go up and down. You’ve seen companies going to market and failing and coming back. You’ve seen that this is very hard to communicate.

We try to teach this at Data Science Retreat. It’s not easy to get people to produce projects that have business value or value for any definition of value. This is a huge added advantage that you have; you can see it when it pops in front of you, you say ‘yep this is business value.’ This happens only when you have the experience; when you have been around the block. I think this is a very important skill that it’s totally underestimated. There are too many people doing things in data science that don’t necessarily translate to value, and the market is taking notice. So, we don’t want to be known with Data scientists as the people who never care about value and produce technically sophisticated projects but don’t add business value.

That’s it for this session, see you in the next topic where we talk about creating or finding value as a Data Scientist.

Getting your first job as a Deep Learning engineer: The current state

This post should give you an insider view of how it feels to be in the market for a deep learning engineer job. I have interviewed thousands of people in machine learning in the last five years; for deep learning, only a few dozens in the last year; I’ve been paying attention to the market, who goes where, salaries etc. It’s enough for me to form an impression.