Data Science For Testers with Laveena Ramchandani

Welcome to the Test Guild Automation Podcast where we all get together to learn more about automation and software testing with your host, Joe Calantonio.

Hey, it's Joe and welcome to another episode of the Test Guild Automation Podcast and today we'll be talking with Levine all about data science for testers and automation, probably in general. She's an expert on all forms of automation. If you don't know. Lavina is an experienced test consultant with a comprehensive understanding of tools available for software testing and analysis. She also aims to provide valuable insights that have high technical aptitude and really hopes to inspire others in the world through her work, which I've seen throughout the web lately. She's been doing a lot of different conferences, a lot of speaking, so that's how she caught my attention. She actually holds a degree in Business Computing from queensmary University of London and Wrigley speaks like I said, at different events on data science, models and other topics as well as blogs and trains future testing. She actually has been helping the Test Guild behind the scenes doing things, helping write blog content and other things. So really excited to introduce her to the Test Guild family. You don't want to miss this episode. Check it out.

The Test Guild Automation Podcast is sponsored by the fantastic folks at Source Labs, the cloud based test platform that helps ensure you can develop with confidence at every step from code to deployment. For every framework, browser, OS, mobile device and API, get a free trial. Visit testguild.comsourcelabs and click on the exclusive Sponsors section to try it for free today. Check it out. Hey Lavina, welcome to the Guild.

Thank you Joe. Thanks for having me.

Awesome to have you. So glad I met you. Like I said, you've been doing a lot for the Test Guild so I thank you for that.

No worries.

Before we get into it though, is anything I missed in your bio that you want the Guild to know more about?

Yes, I just wanted to kind of let everyone know that I fell into testing. So it's been nearly nine and a half years I've been testing so I joined a company thinking that it's a graduate intake. I will go over each and every sort of topic and then find my favorite area and start working there. But no, I fell into testing from day one so it's been an exciting journey and career altogether. I'm working currently as a Test Manager at EasyJet. Now I focus on the data science projects and the shape quality kind of mindset there and also I focus on enabling better quality and testing criteria in the digital side of EasyJet. As well. So, yeah, that's me in a nutshell.

Very cool. So you said you just fell into testing. What made you want to stick with it then?

I think, first of all, I'll be honest, I was confused what I'm doing, so I didn't know what exactly testing was all about. But then slowly, slowly, from my second role and third role, I felt like there's a great potential. How can we deliver software if we. We are not testing it? And I do know in today's date there's a lot of software going out there that's not tested. But what happens there is that users go to competitors and use their products because probably they're well tested and they can still have a nice user journey. So I think that's what kind of motivated me. And I thought, you know what, I'll stick to this area I was contemplating in my early years. I move into Scrum Master, or shall I become a product owner instead? Because it's an easier role I felt that I could do, but I felt like no, testing is the one. And because I started getting worried when I saw code and automation test and I was like, I can't automate for sure. But then I took it up as a challenge and I thought, you know what? It's been my fear throughout university not to code, and it's time to put your head down and see if you can do it or not. So it's been a great journey that way.

Very cool. So you mentioned you're into test management now. I actually got a question recently. I promised them I would ask more of my guests this question because they wanted to know how to make that transition. I'm already off topic, but we'll get into the main topic soon. How did you then? It seems like you, like, you're able to learn on the scene really quickly because you're able to learn things that maybe you're not confident in. How did you make that transition then, from like a tester to management? When I know a lot of people sometimes struggle with that, like, oh, how do I even. What are the steps I do to make that next move then?

Yeah, I think the funny bit is I come from an Indian background and I get a push a lot from my parents, I always have. And I think that's the kind of mindset I always had, that I need to keep pushing myself. So I started as an analyst and I think that's what everyone starts as, as a junior test analyst or an analyst. And that's where I started going from. So I started as a BI test analyst, I became a senior tester and then I became a senior test consultant at Deloitte and then I became a lead at EasyJet. And within three months of at EasyJet, I got promoted to being a test manager. So it's been a nice kind of, how shall I call it, a nice journey because I didn't jump from an analyst straight to a manager. It was very gradual and it felt right as well that, you know, sometimes internally you like, feel that I'm doing all the tasks as a lead. So I think I've ticked that off. Why don't I bring another challenge onto myself and see if I can manage teams and members and do line managing or understanding the budget side of things, not just doing testing. So I think it was a nice kind of career journey I had and I think I would suggest to everyone that start slowly and then once you feel like you are there, you can definitely pick up more challenges. Like you can ask your managers to maybe shadow them. Sometimes that helps because they see the eagerness that you want to also be there one day, so why not? So it's nice to kind of have your managers nurture your ambitions.

Absolutely. So once again, it's curious. So you get into management but before that you're a tester as well. So it's not like you're not a tester anymore. So I love how you have all this experience. So how did you get then into data science? Did you have a background from business, from college?

From that, Nothing at all. That was a complete surprise. It was at Deloitte, so nearly three and a half years ago, that's when I landed into a job at Deloitte. It was really exciting times because it was. It's amazing to be part of huge organization like that. They never told me in my interview that the project's going to be data science based. So on my first day I met the technology manager and I met the leads and they all were talking a lot of data science language and I was really shocked. I was like, is this the right kind of project I'm meant to be in? And they were like, yes, you are meant to be here. And yep, that's where my journey started. So I've never had a degree in data science, I've never done Python scripting before. But with the help of great data scientists and the technology managers, I think I started understanding Jupiter notebooks and how to kind of create my own models and understand the data. And not just the data, it's the story behind the data. Why are we trying to optimize our models to kind of have benefits for our users out there, our clients? How is a model a simple, like not a simple, I shouldn't say simple. Loads of algorithms in there. But why should all of that help a consumer, a company? How can it help? It is such a big thing. And so that's what got me interested into it. And that's where I thought this is the first time I'm hearing a tester myself testing a data science model. I've never heard any test to do it. And that's where I got the idea that you know what, you should be sharing knowledge out there because this is a niche area that many more testers could start getting involved. And till date I feel data scientists, I sometimes get confused as to what we do and I need to tell them I basically test what you are building. So it's been great. I think I landed into this niche and I really thank for that opportunity that I got that kind of a stage to do things and bring in quality in that area.

I love it. So let's break that down. So if someone's listening to this, like I guess let's start the basics, like what is data science? I think statistics is just statistics. What is it?

So data science is a study of data and data is gold today as we know. So it's not just studying the data, it's not just anonymizing your client's data, it's a lot more. It's trying to understand the story behind it. It's how can the data that has been provided to us from our clients help them become better at what they do, make them become probably faster. So I'll take an example of a company, let's say it's a clothing brand. How can I help this clothing brand create clothes quicker, efficiently and in a quality manner? So I shouldn't make them deliver quick but bad quality products. I should help them understand from the start what am I going to create, how long are they going to sit in the depot and when are they going to be in the shops, for example. So it's an entire end to end process and that's where the models actually suggest to you and the things you should be doing from the pricing side, from the material side and from the creation side till the end phase where the customers can actually go and buy your product. So that's a clothing example. I can give you a product creation as well. So let's take a random example. Let's take hand sanitizers for example, it's the COVID time. So let me use that. We've seen that initially we had hand sanitizers. Yes. And I've got some here as well. We saw that they came in the usual transparent kind of solution. And then after a few months we saw that there were solutions with strawberry flavor, with cucumber, with aloe vera, for those people who like to have softer hands. So we saw that more aromas were added. So now think about it in this perspective. So our model basically decides what is the best way to optimize the manufacturing of all of this, the supply cycle of all of this. So what you do is you create a solution which is your transparent solution. And then you'd add probably cocoa flavor. And then how quick can I make this? Is what my model will tell me. My model will then tell me. Before you change ingredients, we suggest you to change potentially not from cocoa to maybe strawberry, but change from cocoa to maybe a cucumber flavor, for example. And because that's quicker to clean up the machines and the vessels where this is being created and then continue the process and then the stickering, like the stickers you place on these little pots and all of that, whether you need to do it in English, how quick can we do it in English, how quick can we do it in another language? Or how quick can we actually transport them to another country as well? So it's all end to end process and then the obsoletence cost. If we are keeping these products in a depot, what is the kind of date, what is the date they will live till? So product can be maybe 12 months, maybe 18 months. But we need to think about this ahead of time because we don't want to waste product. Right. So you need to think from all of that and then your transport costs and then basically they go onto the shelf in a shop or a pharmacy. So it's an entire end to end. And it's really interesting. That's what actually intrigued me and made me curious that what if the model went wrong? The whole supply cycle would just mess up. And it is potentially to happen that way.

It's actually to me, mind boggling. So all those different pieces that you just brought up, those will go into a model and then you test the model to make sure it actually performs to your expectations or something. And then you say, okay, this model performed better than this other one.

Yeah, there's tons of configurations that go in. So if you make a mistake in one config, your parameter is different, you will see a major difference. And that's where it became alarming for me sometimes that this graph doesn't look like what I was used to seeing. So then I checked my parameters and something was wrong. So the best way of kind of testing this as well is you should have a baseline run where you trust that run, and it's been accepted by the data scientists. But also when you're testing after that, maybe say in your following regression or release testing, you should run and compare against your baseline to understand what went different this time, because there's new features going in, things might impact the model. So you need to understand what's different this time versus before. And if my graphs make sense, if my data makes sense, and if the whole process from end to end is actually the right kind of details we should be seeing. Now, with data science, there's one other kind of area. You can have a data model which is off the shelf, or you could have a inbuilt model. So that's the tricky one. That's the one where you need a lot of kind of algorithms going in, different types of best practices going in. So our model was stochastic. What I mean by that is every time you run the model, you will see a different number. Now, that was so confusing. As a tester, why can I not see the numbers I'm expecting? Because that's what we usually do when we are verifying things. So that's where I started speaking to data scientists that this is really confusing to test. We need some sort of standard approach for us to kind of understand that what we're testing is the right way of testing. If not, this is really confusing. So that's where I came up with something called thresholds. So if the values I saw in certain areas were above the 5%, then I know I need to rerun my model or something's not gone well. Some sort of glitch has happened in the model, and I would even confirm that with the data scientists when we verify all the details. But otherwise, if it's within the 5% threshold and the number's a bit different, we would still go ahead with it. So there was a lot of Excel validation going on. I used to be on Excel, like day in, day out with so many tabs open, and it's been an exciting journey, I must say.

So how did this work then? Normally, testers work with the developer. So you would work with a data scientist. They create a model, they throw it over to you and say, hey, is this correct? You would test it, validate, and say, yes, let's move on with this model. How did that process work?

So it was one model which we were working with just one, and we would enhance that model. So it started with a POC first, how do you call it? Inception phase, like an idea. So if it's an idea that will definitely provide revenue, you go ahead with it. If not, it's a fail fast product. So it's been something that has been there for many years now. So only things that we would do is enhance it or remove things that we do not need. But this was one model that the data scientists were focusing on. And whenever they finish their new features, for example, that need to be added to the model or something that needs to be removed, or a new functionality that the customer has asked us to enable, they would do the coding side. So the Python heavy side, I would actually sometimes even code review with them to understand that what have you actually coded and your unit tests. And then I would tell them, this is what I'm going to test, hope this is accurate. And then we would kind of work collaboratively at all times. And then when I finish my testing, I would demo them my results and then we would say that, yes, it's done now sometimes demoing about 15 to 16 Excel slides. Not Excel slides, no, that's PowerPoint, Excel sheets. Sometimes it's really, really complicated to keep demoing those to the data scientists because they are busy. I think another approach that one of the data scientists came up was having an automated process to validate your previous baseline run versus the current run, validate them and run this automated check to see what's different from the previous baseline to now. And we would just click the button, run it and we would see if I've entered a wrong parameter. I would see and then I would need to rerun the whole process, but otherwise I would only see maybe sometimes one column that had a bit of a change of the 5% or not, or sometimes no change at all. That would give us the confidence that, you know what, all the new features that have gone in, they've not impacted the model in a negative way. We can see that the front end is also looking fine, it's running well. And then we would complete the testing and sign it off from our side as well. But I've been involved with the data scientists from the very scratch. It's more of the shift left process.

Cool. So the automation then, was it just an Excel macro or did you actually end up picking up Python and creating a script automated for you?

This was the Data scientist, he created a script with the 5% threshold embedded into that script with Python. Sorry. So we used to use pycharm and it was a quite straightforward product. I really used to like it and when I have to compare it now with Visual Studio, I get a bit confused because I'm so used to pycharm. But yeah, it's a good kind of experience for testers to be in because I feel like there's so much we can do and if we are not part of data science teams, if we know that there is some sort of data science happening and you are getting the end result to test, I think I would definitely suggest get involved. The sooner the possible pair with them, learn with them. Because I say pairing is caring and sharing is also caring. So it's good to have that knowledge around you. And as testers we are super curious, so why not being involved earlier rather than later?

So far we talked about being a tester, how to test a data model. Can you also then you create a data model for your testing efforts? I don't know if that makes sense but since say you're on a project and like I have all these test runs, all these permutations, all this risk, like can I create a data model of my application to know I should at Focus Manager on testing feature Y and Z or something? I don't know.

Yeah, yeah, yeah, I'm sure that's really, really possible. However, we didn't do that. It was very manual. Aim was to have the product stabilized and clients we were piloting with to make sure that they become our clients, not just pilot clients.

So.

So it was important to for us at that point that we need our client base to be firm and then continue from then then onwards. But this wasn't one of the priorities to have a model to basically test what we are testing and just let us know. But why not? This could be an option that a company could use actually. So yeah, definitely I would like to research into that.

Cool. I know we talked about models a lot. I don't know if we ever define what is a model. Okay, I gave an example but like at a high level, like what is a model?

Okay, so I'll give you an analogy because I like giving analogies. So think about a model as you're driving into work, so you get into your car, you put on the aircon, it's a summer day, put the radio on, that's your experience. Then you start driving and you're driving into work and you find some traffic. So that's your data, you're seeing traffic, you're trying to understand route you should get. Should you take a U turn quickly or should you maybe just park somewhere and take the metro or something? That's your data side. So you're thinking and then the model side is the experience. So when you drove into your car, if you reached work on time, that was your model's output. So it's good, it's an optimized way to use your model. However, if you reached late to your work so you have a delay, that means your model needs more optimization. So think of it in that way. It's more of a. How shall I call it, It's a genetic kind of algorithm that we use within the models and try and understand what is really the data trying to do and what are we trying to make this model do with the data that we are injecting through it. So that's the model, the thinking, the main kind of the brain to all of this.

Awesome. So you said this may be a niche topic, but actually I'm curious to know, are you doing this now? And if not, what skills did you learn then from testing data science that you could still apply to your test management gig you have currently?

So I think testers are. Well, testing is a role that is really a great kind of role because I feel like we can interchange our qualities with any kind of role we go into because of course we are curious, we like to ask a lot of questions. So when I was doing the data science project, I asked a lot of questions and I love to ask questions and no question is a silly question for me. So I do ask a lot currently as well. I feel like digging deep into issues and trying to understand from a data perspective is also good. Normally that's something that a data engineer, a data analyst might do, but I think I was doing it from tester's perspective as well. So that's something that you can carry forward as well. As a manager, I feel like we don't get a lot of chance to test, but if I did get a chance to test, I would still use all my abilities that I've gained like a problem solving, even decision making as to yes, we should go live or not with this kind of bugs that we have or if we don't have any bugs, then yes, we go live. So this, I think it's a transferable skill from any kind of role. Specifically with data science, I think is more of the python side, the jupyter notebook side, the tools, the special tools we Use when we do data science, the special shortcuts we have as data scientists, how do they understand things? And that's slowly coming into your mindset. Okay. They think like this. We need to kind of test in the similar format. So that's slightly different. But again, because I am managing the data science side at Easyjets, I can still use that sort of mentality and make sure that we are still shifting left. We are not leaving the testers at the end and letting them test at the end. We should have them earlier so they can even review some code with the data scientists. They can pair with them, they can ask as many questions as they like and can continue from there because it helps it us create better quality test scenarios. It helps us validate our product in a better manner as well. So I think it's a quality embedded culture that is one that I like to have.

Absolutely. I know every time I look at jobs of the future, it's like data scientists is always there. So it seems like a really good skill to have. Also, as you know, AI machine learning have begun really big. I assume that's like just an extension of data science. So if you, a lot of people ask me how can I learn about more about AI machine learning, I think data science skill would be a foundation that they would have to have.

I assume so. If a tester was applying for a data science role, I would definitely suggest go through some sort of courses. I think Udemy has really, really good courses. Pluralsight has really good courses. If you have a friend who is a data scientist, maybe just learning with them. If you have a data scientist currently working within your project, just pairing with them is a good kind of step as well because you learn a lot when you're doing the project with them rather than going over a course. So I would suggest pair up as much as possible because that's the real value you're actually learning and applying. So that's what I would suggest. I never did a degree in data science, so I just learned it on the job again and did a little bit of reading and researching. So I would definitely say do that. There are certain algorithms data scientists use, like genetic algorithms and all the different types of best practices for models. So I would suggest maybe just skim over these algorithms to get an idea. What is this actually trying to do? So genetic algorithm selects the best kind of outcome when using it. So these are the things I was learning on the go, so I would suggest the same.

Great. Okay, Lavina, before we go, is there one piece of actual advice you can give to someone to help them either with their data science testing efforts or animation testing efforts in general. Since you're an expert in all kinds of testing and what's the best way to find or contact you?

Okay, yeah, the best advice I'd like to give you is hair at all times. Collaborate with your team members. Make sure everyone is on the same boat. Learn from your peers. Learn around the product and its benefits. Don't just test. Learn about everything around it. Learn about the story behind the data you're using. Why is this helping someone? How are you helping test this product as well? So boost about your testing. If you feel like there's other type of testing that you can bring in, whether functional or non functional within the data science area, please please do not hesitate to bring it. Such as if you are testing on a front end, make sure accessibility is part of it as well. Because if your future customer uses it and they have some sort of permanent or temporary disability, we cannot miss them out as well. So focus on these things as well and you can all find me on Twitter, LinkedIn and I also blog regularly, so I'm on medium as well. So hope to see you all there and if you have any questions, feel free to ask me as well.

Thanks again for your automation awesomeness. The links to everything of value we covered in this episode. Head on over to testguild.com a401 and while you're there, make sure to click on the Try for Free Today link under the Exclusive Sponsor section to learn all about Sauce Lab's awesome products and services. And if the show has helped you in any way, why not rate it and review it in itunes? Reviews really help in the rankings of the show and I read each and every one of them. So that's it for this episode of the Test Guild Automation Podcast. I'm Joe. My mission is to help you succeed with creating end to end full stack Automation awesomeness. As always, test everything and keep the good Cheers.

Thanks for listening to the Test Guild Automation Podcast. Head on over to testguild.com for full show notes, amazing blog articles and online testing conferences. Don't forget to subscribe to the Guild to continue your testing journey.