Mantissa Data Science Webinar - 1 with Santhosh Shetty

Key Takeaways

are we lying I think we are Santo Trafficante no contention hello cannon let me just try share the link [Music] anyone joining us I'm not sure if anyone its life as well okay we have forgive your life okay I can see princess comment your hydrants just waiting for some toast to join us and we can begin I think there's some problem from his side okay can you really know I can do I can hear you I can't hear me yeah okay yeah I can hear it's just a little bit lag okay so I am writing the lag okay so [Music] okay so let me let me begin I think there will be a lag let me just begin Tim can you okay I'm okay I'm going to start I'm going to thank everyone who is watching us thank you so much for being here there is a lag unfortunately I am not sure how we can fix that this is first if you're trying god I I move so much you're able to hear us and I would let me know yeah I can hear you know okay there's no link yeah okay yeah it's great great so there's no lag okay so yeah so yeah thank you so much for joining us and thank you for doing this so I would like to request you to give us one direction of yourself people watching us yeah thank you very much for hosting this I am my name is Santosh I studied engineering electronics and communications specifically in Ramaiah in Bangalore and right now I am doing my I finished my masters and I'm working at data profit as a junior data scientist so yes that's about to do do you want me to go like in my background alright thank you so much for that introduction it's a great so I think you forgot to mention very a video master strong and let's get into that so you said you did a master's from I also did masters from Islamic though both of us did in electronics but we were not in the same section so we really couldn't so to all those watching I was working in offices we were not in the same team you know who there is none but we were good friends there so we know each other from synopsis and later we went on to do our masters I went to Germany given to the Netherlands and I think he would take over from here and I would like you to introduce yourself yeah so I did my masters in sustainable energy which is completely different from my bachelor's track that I did in at Emma Samia but then I wanted to work on something which was helpful for the world and then I thought transition to renewable energy was something important to work on and then I my last year of my master's I I was looking into how I could be more productive and helpful to society in general and then I realized that whatever technologies we have in the energy industry is already out there it's more about quality and trying to get him into the market so it's more of a quality related problem rather than there's a technical problem so being a technical person I started delving deep into coding and then I took a lot of online horses in machine loading and after that yeah after in the final layer of my Master's to be specific I did the Java specialization all all of these were online courses none of them was I didn't go to any Institute or something and then I did machine learning by Andrew Inc which is on Coursera which after that I followed up with deep learning dot AI on that okay let me you for a bit here so you said you were doing your master's in sustainable energy which is a good field in itself and it has like a lot of potential for future growth and all of that so how and why did you decide to transition into data science ah okay so yeah so as I was saying like renewable energy is techniques it's a policy problem more than a technical issue so then I wanted to apply AI in in technical fields to you know conserve energy or balance energy markets to get more renewable energy onto the grid but for example even in fields such as manufacturing agriculture I feel that AI is still under represented and it's still catching up so I wanted to use AI for social good disaster relief and yeah any sort of social problem also for the manufacturing industry which i think is in dire need of AI so that is why I decided to switch into this yeah so why do you doing your masters do you have any courses or any any like basic background that would like facilitate you into this or was it daddy was it something you develop when you want the interest I would say that so back in bachelors we had done a face recognition problem and we wrote a paper on it that some team mates but then after that I was away from anything related to machine learning and in my Master's I did not have any courses at the college that I took in machine learning but then I took all of the courses that I did work online and the background that I is important for this to get into is I feel uh some sort of coding experience maybe like one year or something that should be enough to you know learn these courses on your own so do you think your contribute in some way for your transition or was it because of these online courses that it would make this transition I would say it's more down to the online courses because in my track was completely different from anything related to a I so everything was through online courses okay yeah interesting so which brings me to my next question so how many actually do and you like you said online courses which forces du/dx diffic because I get this question a lot of you are almost every day they see a new course coming out and then they're like which should I take up most of the times I would recommend them professor and ruins its course that already then it and then there is faster AI which I know you're a very big fan of and I'm sure you will take any now but yeah so these these these courses are I would say the methodologies that they follow in the way they teach refers to the ng takes the more traditional Road and faster T is is the opposite and yeah so there are lots of courses that keep coming out how do you manage yeah first of all you can start with which courses you started with and then how did you narrow down yeah so initially I just did a coding specialization to learn coding because I felt I was out of touch so I took Java specialization from Duke University on Coursera which helpful for me to you know just implement code and get me familiarized my family dies myself that's how you code and after that I did machine learning by Andrew egg which is I think a really good course that every person who wants to get into this field needs to be doing it gives a very good background and basic concept founded idea of the basic concepts that are required to perform any sort of AI modeling but then after that I did deep learning got air which is the hot field right now in machine learning I found that course by Andrew into the specialization itself I learnt a lot but then I could not implement it in practice although like there was the math concept and some of the other important things like validation and the details of the machine learning but as it came to implementing of those on my own I felt handicapped and I couldn't do them so then I was looking for some other resource that could help me get into this field to make it more practical that's where I found fast AI by Jeremy Howard it's an online course which is free and you could start off with intro about a intro to introduction to machine learning which is I found a really good starting point which builds upon your machine learning knowledge that we learnt from androids course so the reason why this was helpful is because you don't start off coding from scratch initially you implement what are the models in an abstract sort of manner already and you try to uncover the details layer by layer and it's a top-down approach which helps you implement the practicalities instead of the code which throws off people most of the time and for someone not coming from a coding background this is I think really helpful for getting into this field so you what you mentioned you will not from according background and and when you're coding using all these libraries and Python or whatever language you use so how did you actually build up your reporting skills yeah so like I said I did the Java specialization that was really good for me to get concepts of object-oriented programming although you don't use that much in machine learning Java specialization by Duke University it has a list of five courses so it starts off from very basic concepts and builds up to like a final recommender system project scan which is built in Java so yeah that is the coding course I took but after you do Java then you can handle any coding platform be fightin or anything else yeah so these days Python is the most used for machine learning and it's pretty straightforward once you know [Music] machine learning course was on MATLAB that is still very user friendly in terms of writing code and after that I did deep learning got AI which was completely in Python so the way you start off with poisonous stack overflow if you don't know any answer to any question you just google it and then Stack Overflow should be able to provide most of your answers this is there any resource which would tell you like for example the data structure he would use in Biden would be dictionary is Liz yeah so it's different from what you would actually use in Java you suggested you suggest we dive into the code directly or do you take up a course and then maybe try and start programming so how do you go about this okay so I would suggest that you do up if you are not from according background you do a specialization or do just like the basic two three courses of that specialization to get a feel of how coding works and after that I would not recommend you to directly go and learn the language Python itself it's better to you know it start off with the code because you already have some coding concepts and if you have any doubts in figuring out what the Python constructs are you can google it that is a better approach because directly going into the code is in learning a new language again is not feasible especially another language like Python specializations that people can focus on - I serve in Java just because I thought that was some language coding language I needed to learn but yes a different language when you compared with mine right so that a Python is more dynamic dynamic type in Java is very strict with its declarations of variables like that so so I don't like now of course I will not suggest people to go and learn Java because they want to get into machine learning and data science because that would be a waste of their time I said because a lot of people actually implement stuff in Java in production today for machine learning right now Java is still used as the top enterprise language for applications but for machine learning products I wouldn't see it's used but having said that Java is still a very comprehensive programming language I would say it has everything you would need to get get started with programming understand what constructs are there so that's what that was the first programming language I learned as well so I mean definitely if you look really more to explore programming understand the programming side of things like I think definitely you can explore Java or even go on further in triangle and see which is you going more lower-level here but I think really I think would be a good place to start more importantly understand understand the medicine itself how programming works right like what what are data structure and how do you use okay so we did we digressed from 38 so you say something yeah so deep learning that AI by entering did not help me in terms of implementing practical hurdles so I turned to fast AI which is a really good resource that everyone needs to check out a thing if you have basic machine learning knowledge this is like a really good cohort for any any person from any field to implement codes because it goes down from a top-down approach and it really makes you understand how to implement the model rather than you know build the model from scratch so you apply the model to a problem and then that in doing so you learn how the model works which is a better way to learn how it works rather than learn the model working of the model and then go ahead and versus if you see every year they keep adding on you I think the latest version was added just a month back or something so what version material so there are two main versions one is the deep learning for coders version and one is the introduction to these are the major divisions and then deep learning for coders keep getting on you're over in their video so what did you do and what would you recommend others yes so I think you have to start off with introduction to machine learning which dives deep into random forests which is a very important algorithm and it gives you like the basic idea of the state-of-the-art machine learning apart from neural networks once you are done with that you can start with deep learning for the yes there are two parts of deep learning which have versions that keep coming out every year you can start off with deep learning one which the version third up which is released one once back like you said and yeah the second part of the deep learning course is going to be out in March I think yeah so yeah I would suggest that the roof intro intro to ml first from far away and then you follow up with deep learning one from faster da the latest version all right okay so of course you did and then also which is learning by Professor Andre Angie so anything else apart from these resources specifically no just these software I just really brought me to the next level before that I was really struggling but I think just these doing these resources would place you at a good point and at a good level to start off building really good algorithms for the real world so when you say the next level I get reminded off fact that Jeremy retweeted your tweet I mean in general for those who know Jeremy Jeremy who would was the president of cattle cattle is a very popular website for data science competitions and he's very active on Twitter as well so he's the one who runs faster day and the other day Jeremy had retweeted santosha screen and he was over the moon so can you tell us how and how did that happen yeah yeah so what I was trying to implement some of these models for the hackathons that I was attending so the one of the hackathons I attended was organized way The Hague the public department of The Hague they wanted to classify damages of the damage levels of buildings after a disaster so they can provide relief for based on that information so we built a model using fast idea is in built library and we achieved a really good score which was which beats the correct which beat the old state of the art and yeah so it was is really easy to implement we could build a neural network in just 20 plying of code because libraries are very modular and you can pass in your data in if you're parsing the data in a specific manner to the code it can build these things so Jeremy looked at this since it has a new state of the art it was it we could be tweeted this saying this was so the aim of fast idea is to enable other people from backgrounds which are not coding or machine learning specific to get into this field and he tweeted the saying this was an example of something which could be done by anyone which is really good I mean that's great I mean amazing acknowledging your work there so really cool so ok coming so we've already discussed the resources you use you've already told us what do what inhale you coming - okay after you study what do you do often anymore to start looking in the history and really get a job so that you can further your knowledge so how did that face go like when did that start how long did it last can you display more of that okay so when I finish the course all the courses until deep learning one by fast out AI that was June of last year and then I was interested in applying this to the real world so I try to use kaggle datasets and try to enter competitions unsuccessfully I did not win I haven't won any competition it but it slowly started improving because other people post kernels and comments and discussion each happening on whichever competition you attend and the playground competitions and cable are really good for you to get a good hold on what is happening and how you actually have to build a model once you have the skills so that's a really good place to start apart from that I was I got lucky in sense that I one of my friends wanted to enter a competition in Switzerland in trying to use these machine learning concepts for a business that he was building so there I learned like the business side of things which you know helps you to understand how to bring this to them as a product and also to make you understand what sort of issues you are going to face in the real world so I would recommend that you try to get datasets that are present on cable or public datasets or there are many competitions that are happening online hackathons buy analytics Vidya or yeah that all these places can be really good resources for you to start on with learning with implementing real models that could be used for the actually environment okay that makes sense but how do you actually make yourself differentiate out of everyone else or sometimes because so I also went through this job search faith and it takes a lot of effort even though you know nobody wants to look at you and you're not being considerate you're sending in applications you're getting rejected so how do we do this how long did it take for you to actually end up a position so I started interviewing in June of last year that's when I finished my masters and I was looking for jobs I had made my mind that I'd want to transition to machine learning and trying to find a job in that so the first two interviews were unsuccessful so I got data sets which were pretty easy to handle but I was not I was not so confident of my skills which is why I did not analyze the data set properly but then when you give a menu the way to get people more attracted to your CV or your resume or your portfolio say it is to put up like the courses that you have done and also like the competitions that you have entered if you have got a really good score that's a big plus and also if you have done any side projects that by yourself you don't have to enter any competition as such you could have like a really good model that you have built on any data set that you found was interesting and like to do this one right medium post or a blog post and use Twitter to share your information post your girl codes on github not just post your codes on github make them readable write a good read B file so that like a mr. anyone can wants to access and know wants to know more about you means I can go to your github and see that if you have commented them properly and if you if it's readable if it's understandable because building models is one part of the problem but you also want to make it accessible and readable for any person who wants to use it so attract traction by trying to make write blogs and have a really good github portfolio that's also for every project you did you we're making sure that you write blog post on that and making sure they're clean and all that yes yes my blog posts my github all these were very important for my job search I think at least because I did not have any background and these helped me differentiate myself because I despite having not having done any courses any courses in college universities I was able to put up models and an explainable article that people have liked and asked for suggestions one and people recruiters actually valued this sort of information even though if you haven't done like an annuity level course because they know that and you're trying to make it understandable or the yeah I would say that those the projects that you put up on your blogs or github needs to be unique do not do m-miss and those are practice courses which practice models that you build upon yourself take a new data set for example in my case I was very interested in rapes that were happening in India so I I took the rape dataset from the Indian Republic it's hosted on their website and then I made an analysis on what which stage has been improving which state has been reducing it was not any there was no model built or this just like excels not even Excel a tableau analysis of graphs and seeing the pattern of change over time about how these rapes have changed over time in India so that was my first blog actually and I share that to the public community after that I started I took data set to predict housing prices so you could I would suggest that find a dataset that interest you and not like a common data set which like for example like you specified MLS something which is unique or something which is interesting for you so that you can keep going forward there delve deeper into it and then make an analysis and put it up online if you're using the data sets that you practice on and that have introduced you to the field [Music] so to almost the end so you recently started your job India yes I started in February and how's that going so far it's been really good my learning curve is very steep for I have to know what I have learnt through my my own projects are quite different to what happens in production because you need to take care of all and ensure the data has been like you know passed properly and the datasets that you work on during projects are very small actually but then when you come to production there's extremely big so you need to figure out a smart way of going about how to collect this data and put this data together to make sense of it and I wanted to apply AI in manufacturing so I was looking something in this field or supply chain or anything related to industry so I got lucky I applied on LinkedIn and then there was this company in South Africa a startup called data profit they're working on optimizing the manufacturing industry using AI and predictive maintenance and so to give a few examples they do predictive maintenance and object detection optical character recognition and on industrial products part time series analysis so these different products and find out water Annamma anomalous instances of apart makes the qualities of the products better so they cut down energy consumption this is the kind of project they work upon and this is very interesting for me all right so we've covered you journey right from the time you started your bachelor's you do your masters and then you took these online courses and then you had this job interview phase did a profit certificate I know you are working for them yeah all the best for your future to do greater greater things that you've achieved so far last last bit we'll take some few questions from the audience there are a few questions on the live chat so I will read them out for you and maybe you can try and answer so before that I would like to just put 3/4 editions and three will be helpful so one is when you're coding always use a top-down approach in coding which I found really helpful what this means is you already assume what you have for example if you are building the code then you want some kind of an output you already assume the sort of inputs are already filled in and then you can write the code for that piece of information and then later on you can decide ok these inputs are not present how do I obtain these in from these inputs write a function for that so you approach from the top to the bottom it I have found it really helpful when I'm coding and I think it's helpful for you that's one thing then try to formulate your problems in an ascetic mathematical manner and decide what if you have certain data decide what I can do with the existing data to achieve your goals and also make a note of what extra data you might need to achieve those that's important another thing I think is really helpful is use Twitter to be in touch with the latest trends in AI the there are people three things about these latest information in AI and what sort of things you need to get into the most exciting things that are happening so I follow I use Twitter as my source for information of the latest happening in AI and I think that's about it here alright so I think we have a few questions should can I proceed with the questions deep charity his question was in day-to-day functioning of ML engineer or a scientist a lot of data is unstructured or there is no ready-made trader destructive for a new person in this field especially from a non CS background how could I learn those techniques like how to build data pipelines or data Natalie okay so I think for it is something so data wrangling at least I would go through online resources and try to find something this is helpful for that particular specific problem for example if I'm handling the time series information where my information is highly unstructured for my current projects right now so I need to go through a lot of set of files and it's not I cannot make a simple data frame but I need to pass through multiple files and make them on the fly so it was a big challenge for me to understand how to go about those things so what I did was I just not noted what are the important points from each file so that I can use that for later purposes and it's it's hard to specify for a specific problem for a general problem but for a specific problem you need to find out ways which are I would say not intensive on the system so you can use those informations which are represent in each file and try to try to try to make things modular I think that's the most important thing if you if you make things modular then you can fix that piece of code and go on with the next part of the course that is nothing one of the basics of best practices or program try and keep as much as possible okay so next question is from Vijay who asks which language is most relevant to get started with immense vital person that's Python is the best language to start off with and head on it not only helps with your normal machine learning algorithms but then when you get deeper into deep learning then yeah then you have also libraries that are coming in which are tensorflow white arch and all those things I personally prefer white art because it's very easy to implement but yeah I think my name is not yeah yeah and I think Prince wants to know more about the machine learning goes by and you can you please give links for the same I think since you can adjust I mean it's the first book that comes out if you search for machine learning Coursera and to inching right I think this is the most you to the standard yes yes yes that's on based on MATLAB if I'm not mistaken octamers I'm not sure there was one course in octave is could use either of those yeah but yeah so prints maybe be just search for Sarah and I think that's the course and some toasted and everyone else does all right so we've come to the end if you have feel free to contact me on LinkedIn I'll but please put a note if you want to be asked a specific question and find out any information I'm always there to help because I have been through this journey and I understand how it's hard and to get up to speed so I had yeah and if you feel free to reach out on LinkedIn alright thank you so much Santosh for doing this I understand how the journey has been for you because I think along with you and I understand how the journey is for the ones who are currently in the process and that's why I Thomas would be a good idea to actually go through we go through journeys on people who have walked along this part maybe and it and the exciting thing about this field is that there are people from past fields that are coming like different backgrounds and you never know which is your path so I think maybe someone is already walking on your path right now and they see now they see now really what their next steps might be so thank you so much for doing this and with this I would like to close the webinar for today thank you so much for joining us thank you so much goodbye have a good evening or afternoon yes all right right right