Devops Vs MLOPS- Understand The Differences And Why IT is Important

Key Takeaways

hello all my name is krishnak and welcome to my youtube channel so guys today in this particular video we are going to discuss the basic difference between devops versus mlaps if you don't know guys uh this week we had conducted four days community session every day was around two hours of session on ml ops and if you have not seen that particular playlist the link will be given in the description but still many people had a confusion regarding devops and mlaps what are the basic difference between devops versus mlops and in this particular video i'm going to completely clear it out for you what is the exact difference so one one more thing that you need to understand guys devops and ml ops are also very good job requirements in most of the companies specifically where software the softwares are getting developed probably a web application that is getting developed in short whenever you are working in some agile process which involves software development life cycle devops is pretty much compulsory before going ahead with respect to this particular videos guys i want to announce i neuron is coming up with an advanced dlcv nlp course and this is basically starting from 17th april 2021 the class timing is between 12 30 pm to 2 30 pm on saturdays and sundays and apart from this the doubt clearing session will be on 8 p.m to 10 p.m on thursdays uh the prerequisite of this particular course is that you need to know python programming language and the basic of machine learning then only you will be able to understand the deep learning things uh you will also be getting all the pre-recorded videos as soon as you enroll for this particular course apart from this you also have an email support and skype support all seven days in a week where the live skype support team if you have any kind of queries you can approach them and you can ask them any queries apart from this all the information is given in this particular link the link will be given in the description of this particular video so let's try to understand what is the basic difference and we'll try to get to know a lot of things regarding this so to begin with guys if you have some basic experience if i talk about something like an agile process now in an agile process what happens is that whenever we are implementing some stories the stories are divided into sprints like it may be sprint one sprint two and in every sprint we try to create some number of stories after suppose if i take the duration of sprint one it is usually two weeks right two weeks of duration where we need to probably implement some of the stories within a team after that particular stories is implemented and this i'm considering right now i'm discussing regarding devops after the stories has been completed in sprint one what happens is that it is basically sent to the cicd pipeline that is called as continuous integration this is my ci cd and then this is basically my continuous deployment so these two things are pretty much important continuous integration and continuous deployment first of all let's discuss about continuous integration now in continuous integration what happens is that as soon as a set of stories has been developed we definitely have to give this particular set of stories to the qa team in a team guys you will be having a development team you'll be having a qa team the development team will be responsible in developing the stories the qa team will basically be testing those stories right uh and this i'm basically discussing when you're developing a software or whether you're developing a web application right at that time simultaneously and paralleling an agile process as the sprint is getting completed qa usually tests in this set of environments so what happens is that initially after sprint one everything is deployed in the development environment so they are also different different environments in this development environment what happens is that the development team does a thorough testing completely end to end with respect to the application they have developed and whenever the qa wants to test it you know from this development environment it is pushed the entire thing is pushed to the queue environment and this actually happens in the form of continuous integration and continuous development right so that is the reason we say ci cd pipelines now what does the ci cd pipeline basically means and once it is pushed once it is pushed to the qa environment the qa team will basically be doing the testing and once they give a sign up this is basically moved later to the staging environment right and finally it is then moved into the production environment right in the broad environment so this is how the process usually goes off as the qa gives a sign off again it is moved to staging environment in staging environment also there will be a separate round of testing by the qa team uh with respect to all the bugs and all whether it is solved or not and once they give the staging it is then pushed to the production environment after the staging there is also an environment called as you at environment but i'll just limit it till the staging environment itself now before pushing into this different different environments there will also be a separate pipeline where they will be putting uh in the form of ci cd pipelines so this is nothing but ci cd pipeline suppose qa raises some issues with respect to sprint one then what happens is that those issues are fixed uh by doing continuous development and then it is probably integrated with the ci cd pipeline and then again it is done the deployment is basically done in the dev environment and later on it goes like this so whenever you are developing a software uh or probably a web application usually we follow this kind of strategies to do the deployments and to plan the strategies itself for every sprint we take up stories we develop it and then we continuously develop it and with the help of ci cd pipelines we continuously push from one environment till that next environment till it reaches the broad environment now the question rises what is then ml ops when i talk about devops guys the first basic difference is that whenever we are talking about devastation we are basically talking about uh you know a software application probably or when we are talking about probably a web application okay now whenever i talk about ml ops at that time i'm specifically talking about a data science project it may be a machine learning project it may be a deep learning project okay now why ml ops is pretty much important with respect to a data science project because understand one thing if i consider a machine learning project probably there are various steps the life cycle completely changes guys okay even though we will be having sprints and all but as stories are getting planned usually a common pipeline is basically created what are the basic pipelines one is probably the requirement gathering requirement gathering will be the part of will be the work of a data business analyst then we basically go with a next step which is called as exploratory data analysis eda right we do a lot of exploratory data analysis the third part is basically we do a lot of things with respect to feature engineering okay the fourth step that we probably do is something called as feature selection and this whatever i am discussing guys these are called as the life cycle of a data science project then the fifth step that we basically follow is basically called as model creation the sixth step is basically the model hyper parameter tuning i'm sorry for the bad handwriting i've never been a good handwriter i i from for my young days itself right my uh teachers used to scroll with me with respect to the handwriting that i used to write model hyperparameter tuning and then model deployment and the final step is about the retraining approaches retraining approaches now these are all the steps with respect to my life cycle offer my this is are nothing but this is the life cycle of my ds project or i can say ml project so this is my complete life cycle of my ml project and we usually follow this particular strategy now remember one thing guys when this strategy is again formed in ml ops we basically say this as uh you know just understand devops software you're developing software you're developing web application and we're following some ci cd pipeline strategies all this life cycle that you see that is from requirement gathering to eda to feature engineering feature selection model creation model hyper parameter tuning model deployment retraining approach everything right apart from this requirement gathering what we do is that we set up all these things in a form of pipelines now when i say pipelines the best pipelines that i come to mind is ci cd pipelines obviously here also we use ci cd pipeline this ci cd pipeline will probably do the steps of eda feature engineering feature selection model creation model hyper parameter tuning this is very very much important guys in the ci cd pipeline because usually model right uh whenever we develop any kind of models we do lot of uh you know hyper parameters we play with the parameters unless and until the model gives us a very good result so we need to continuously do this and based on this the ci cd pipelines is basically created then we also have the model deployment now after this there is something called as retraining approach for this retraining approach along with the ci cd we also write retraining approach separately because the retraining approach of a model is completely different when compared to the devops in the devops we don't have anything like retraining why we don't have because in devops we develop a software and software do not degrade whereas a ml model do usually degrade okay let me just write it down over here okay just let me make it a little bit over here okay what happens is that with respect to devops whenever we are discussing about a software this does not degrade what does the what what do the meaning of d grid is i'll tell you whereas in the case of ml models as we get new and new data new and new test data this may degrade what is the meaning of degrade over here in this particular scenario see guys when we are actually talking about degrading that basically means the ml model as more and more data comes more and more new data comes usually the performance becomes performance is lower right performance degrades in short okay it degrades it will not be having the same performance that we had initially probably because the data may have changed probably now we are having more users information probably more user behavior we are able to get from this whereas in this case of software you know whenever we develop one software a software can only be improved okay but it will never degrade suppose if i'm developing a software to do some kind of work it will do that particular work right and after some point of time you may get some new bugs and that bugs can be definitely integrated and that is the reason why we use ci cd right continuous integration continuous development continuous development basically means if there is a bug i'll try to improve it i'll try to fix that and i'll try to include that or integrate in my software the software never degrades once it is developed whereas an ml model degrades so this because of this what i have written over here ci cd plus retraining approach this retraining approach is also a pretty important task in my ml ops that basically means whenever i am solving a machine learning or deep learning projects this retraining approach along with ci plus cd pipelines we have to develop and the ci cd pipelines that are actually created over here is completely different because the life cycle of a data science of our ml of our dl project is completely different when compared to a devops project right this is the most important thing that you need to understand so let me just give you the basics and the main differences between devops and ml ops in devops you know that it is the combination of just ci cd plus deployment deployment is also here guys when i say ci cd when i say continuous integration i'm also talking about development uh deployment but in but in the case of ml ops you have one more additional thing that is your retraining approach that is your retraining approach so you have this new step called as retraining approach the second most important thing is that usually your ml models degrade whereas your software application or your web application do not degrade you can just find out more bugs you can still add more features on top of that that is the reason why we include this important pipelines called as retraining approaches in our ml ops activity now these are the most important uh differences when i compare devops and ml ops but remember one thing guys with respect to mlobs the complete life cycle of the project is different we start with eda we do feature engineering we do feature selection model creation model hyper parameter tuning where we play with lot of parameters then we do the model deployment and then finally we go with the retraining approach that is the reason and these are the some basic difference between mlofs and devops yes both are involved here we are also getting involved in deploying from one one environment to the other environment like from dev to qa qa to uh you add to staging staging to prod in devops also we do that but again the entire pipeline is completely different this pipeline needs to be always there whenever we are training our model in case of mlofs we can also train our model in offline mode we don't have to follow this entire pipeline also sometimes you can develop you can probably train your model separately but it is always good that we develop this particular pipeline and then we train our model yes for serving the model there may be a separate pipeline based on the application flow and all so i hope you understood this particular difference between devops and mlaps please do give a comment if you have not understood anything i'll be happy to help but uh and sorry for my bad handwriting but i tried my level best to write it because when i'm using this kind of pen it usually becomes difficult uh when i'm having connected monitors at all so yes i'll see you in the next video have a great day ahead thank you bye