avanish474 / mlops3

Repo made for mlops task 3

MLOPS Task 3

Tired of tweaking and waiting for your classification models to be perfect, here is how you automate them using JENKINS + DOCKER + GITHUB

SETUP :

• Copy the get_model.py , make_model.py and modify_model.py from this repo
• Create a github repo for your project and add make_model.py , modify_model.py and get_model.py to it
• Now go to your home directory and clone the project using git clone
• We will be using tensorflow with docker, follow the official documentation to set up your docker to run tensorflow https://www.tensorflow.org/install/docker

How to use this project :

• Put your data for ML model structured as follows :

  • faceData
    • class1
      • class1_1.jpg
      • class1_2.jpg
      • . . .
      • . . .
      • class1_n.jpg
    • class2
      • class2_1.jpg
      • class2_2.jpg
      • . . .
      • . . .
      • class2_n.jpg
    • . . .
    • . . .
    • . . .
    • classm
      • classm_1.jpg
      • classm_2.jpg
      • . . .
      • . . .
      • classm_n.jpg

• Put the faceData in the same folder where you have cloned the git project !

• Run the get_model.py script

• VIOLA its done, now sit back and wait for the mail, that the best model is trained

The setup we wil do below, will automatically train different variations of your model and outputs the best possible model !

Pre - requisites

• Basics of Docker
• Basics of Linux Operating System
• Basics of Jenkins
• I am assuming you have already set up Docker and Jenkins in your OS and has given root privileges to jenkins user, if not, then do the following

echo "jenkins        ALL=(ALL)       NOPASSWD: ALL" >> etc/sudoers

Now let's create our jobs and jenkins

TASK :

• 1. Create container image that has Python3 and Keras or numpy installed using dockerfile

• 2. Create a job chain of job1, job2, job3, job4 and job5 using build pipeline plugin in Jenkins

• 3. Job1 : Pull the Github repo automatically when some developers push repo to Github.

• 4. Job2 : By looking at the code or program file, Jenkins should automatically start the respective machine learning software installed interpreter install image container to deploy code and start training( eg. If code uses CNN, then Jenkins should start the container that has already installed all the softwares required for the cnn processing).

• 5. Job3 : Train your model and predict accuracy or metrics.

• 6. Job4 : if metrics accuracy is less than 80% , then tweak the machine learning model architecture and retrain it.

• 7. Job5: Notify that the best model is being created

• 8. Create One extra job job6 for monitor : If container where app is running. fails due to any reason then this job should automatically start the container again from where the last trained model left

TASK 1 :

• 1. Create container image that has Python3 and Keras or numpy installed using dockerfile

• Copy the Dockerfile from tihs repo and paste it into a new dir

mkdir /paste_docker_file_in_this_new_repo
cd paste_docker_file_in_this_new_repo/
docker build -t tf:v5 . 

• Now we have an Image named tf:v5

( The image might take a little time to pull from docker hub )

TASK 2 : Create a job chain of job1, job2, job3, job4 and job5 using build pipeline plugin in Jenkins

• JOB 1 : Pull the Github repo automatically when some developers push repo to Github.

• This is easy, make a new job named pull_from_github

• JOB 2 : By looking at the code or program file, Jenkins should automatically start the respective machine learning software installed interpreter install image container to deploy code and start training( eg. If code uses CNN, then Jenkins should start the container that has already installed all the softwares required for the cnn processing).

( we are making classification model using Transfer Learning for now, so all we need is to start the tf:v5 we built )

• Make a new job named run_the_container and add the following

• JOB 3 : Train your model and predict accuracy or metrics.

Here we will train the initial model made by user

• JOB 4 : if metrics accuracy is less than 80% , then tweak the machine learning model architecture and retrain it

I am using a python script to manage all this so the complex part is already handled by it

• JOB 5 : Notify that the best model is being created

*We will mail the user, notifying that the best model is trained, which can be found in jenkinsDownload folder along with history file

Image here

• JOB 6 : Create One extra job job6 for monitor : If container where app is running. fails due to any reason then this job should automatically start the container again from where the last trained model left

• We will check is the container is working or not, if not re-run the JOB 2

Build Pipeline

• Insatll the build delievery pipeline from Jenkins Plugins Manager

• In the Dash Board Click on ' + ' to Add a new View

Then Follow the images Below

• Select Initial job as pull_from_github

FINAL BUILD PIPELINE

Sit back and let automation do its work

• 1. Create container image that has Python3 and Keras or numpy installed using dockerfile

• 2. Create a job chain of job1, job2, job3, job4 and job5 using build pipeline plugin in Jenkins

• 3. Job1 : Pull the Github repo automatically when some developers push repo to Github.

• 4. Job2 : By looking at the code or program file, Jenkins should automatically start the respective machine learning software installed interpreter install image container to deploy code and start training( eg. If code uses CNN, then Jenkins should start the container that has already installed all the softwares required for the cnn processing).

• 5. Job3 : Train your model and predict accuracy or metrics.

• 6. Job4 : if metrics accuracy is less than 80% , then tweak the machine learning model architecture and retrain it.

• 7. Job5: Notify that the best model is being created

• 8. Create One extra job job6 for monitor : If container where app is running. fails due to any reason then this job should automatically start the container again from where the last trained model left