MLOps - End to End automated CI/CD pipeline for Continuous Deployment

#datascience #machinelearning #mlops
In this webinar I will be demonstrating end to end processes from model development and model deployment. CI/CD aspect will focus on continuous model deployment. Topic Agenda includes
Building an Image Classifier
Creating a Streamlit web application
Deploying the web application in K8s
Continuous Integration pipeline (GitHub)
Continuous Deployment pipeline

[@AIEngineeringLife]

Code used in this demo is available here - github.com/srivatsan88/ContinousModelDeploy

[@jaynouoliverbeck6181]

i still see this video and this is still relevant, one of the most underrated channel here, full of resources

[@ephremtadesse3195]

Excellent tutorial to go with simple automated ml model serving!

[@ameygirdhari8703]

Sir you tutorial is awesome. Clearly explained all complex things in simple manner.

[@tejeshivsn632]

Thank you for this amazing session. Learned many things.

[@pavanim6258]

Thank u so much Sri Sir for taking time and ur excellent explanation.Though the video is lengthy,the way u detailed it out is really worth watching.I learnt a lot from this video.

[@TheCuriousCurator-Hindi]

Lovely tutorial. Almost self sufficient. Thank You!!

[@rameshundralla6001]

Really good start for learning MLops and basics on Pipeline.

[@ashishgudla5411]

This is literally terrific work. Keep it up Srivastav Sir

[@hernanescu]

Your channel is amazing! Thank you so much =)

[@anujsingh2455]

Thanks for all your efforts 🙏

[@airsmokker]

best for mlops basic and intermediate

[@MoncyMohanNellisseril]

Excellent and thanks for your effort :)

[@AnuragKhareOracle]

Thanks a ton for such a wonderful session.

[@sweetypaliwal410]

Thank you for this video , great learning!

[@RajivMukul]

Thanks for the tutorial.Can you please tell me in which scenarios python scripting is used in MLOps workflow apart from building a model .Also in which cases we use bash.Basically want
to know how much scripting is needed in MLOps apart from building a model.

[@uday2kiran]

Hi Srivatsan, very good session indeed. Do you have any video on how to deploy on AWS without sagemaker?

[@SUGATORAY]

Do you have any tutorial on deploying using docker swarm vs. kubernetes? If not, and if you could make one with explaining the pros and cons of each, that will be really helpful.

[@AIEngineeringLife]

Sugato.. Nope I do not have docker swarm. Reason is not many use it these days and in cloud most of them provide K8s as managed service and this has made swarm adoption very low

[@SUGATORAY]

@@AIEngineeringLife Thank you.

[@venkatesanr9455]

Thanks for the informative content. Hi sir, Whether docker utilization is required or not and I believe the video does not contain a docker explanation. Kindly correct me if I am wrong. I like you to discuss/make video on retraining/monitoring approaches in detail.

[@AIEngineeringLife]

Venkat, with kubernetes yaml docker is just bundling and we dont manually push docker images. It is part of cloud build file. Docker is more tied to k8s here. I will try to make a retraining and modelling video

[@venkatesanr9455]

@@AIEngineeringLife Thanks for your kind inputs Sir

[@abhissingh1997]

Hello Sir, Great tutorial!!! I have one question when you run the pod container flask gives warning that it should only be used in development not in production. I understand that we need gunicorn and nginex When I deploy in ec2 or compute engine, but what about in case of kubernetes does kubernetes take care of multiple requests or I need to implement gunicorn and nginex for production here as well

[@TheCuriousCurator-Hindi]

dense layer has softmax activation so from_logits should be false in crossentropy loss

[@atlapavan5359]

good explanation, while deploying the pipeline, I am getting timeout error, help me where to change the timing.
Waiting for deployed objects to be ready with timeout of 5m0s
Deployed object with kind "Service" and name "imageclassifier" is ready after 800ms
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Still waiting on 1 object(s) to be ready: [{kind: Deployment, name: imgclass}]
Finished applying deployment.
################################################################################
> Deployed Objects
NAMESPACE KIND NAME READY
default Deployment imgclass No
default Service imageclassifier Yes 34.127.55.212

[@mohamedcamara6606]

I had the same problem , you need to check the permission for the service account used for k8s-cluster to pull the images from GCR.
I provided the correct roles and then it worked. Roles:
storage
object
viewer

[@adesojialu1051]

how do i use kubernetes on vscode extension and how wouild i know the specification for creating .yaml file?

[@bhagwanmundhe1059]

where are you sir, since long time there are no videos, Ihave watched all your videos

[@shrikanthsingh8243]

Is the directory structure shown in the video same as the structure shown in your GitHub repo ? If Yes, Then you are copying even the K8s folder inside your docker image, Am I right ? But the files of K8s folder are not required to be copied.
Proposed solution - add k8s folder to the .gitignore file. Am I right ?

[@gururajraykar9254]

How the Continuous training been done here? As MLOPS having feature of CT scope. Could you please make a video of that. Thank you in Advance

[@shombeetdasgupta]

on running the cloudbuild - i got an error saying "You may need to grant permission to access to the cluster:" Not sure how to resolve it ...is it something which we need to configure on the cluster itself

[@AIEngineeringLife]

You might have to make sure your cluster service account has access to execute cloudbuild

[@kpsk1627]

What is the purpose of pushing the image to a registry?

[@Raj-gc2rc]

To pull it back into the cluster

[@krishnaveninagireddy1252]

If i would like to take training how to reach you

[@alexraja7785]

Sir , ur Google Colab or github for this code pls

[@AIEngineeringLife]

It is already in video comments

[@alexraja7785]

@@AIEngineeringLife Thanks Sir