Integration of Machine Learning with DevOps

Here we will be Integrating Machine Learning with the DevOps .. (GitHub , Jenkins , Docker )

Lets get started with the implementation ...

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

FROM python:3


WORKDIR /usr/src/mlops


COPY req.txt ./


RUN pip install --upgrade pip 


RUN pip install --no-cache-dir -r req.txt


COPY . .


CMD [ "python", "./main.py" ]

• This will create a Docker Image for Machine Learning models .
• req.txt file contains the following libraries ..

Keras
numpy
tensorflow
pandas

Note : pip will configure all the dependencies required for the libraries .

• I have given a extra line in the Dockerfile "COPY . . " . It can be used to either copy your python code directly to the image while building or we can also use mount while launching the image .

Job1 : Pull the GitHub repository automatically when developers push repository to GitHub.

• For this we will be using GitHub Web-hooks integrated with Jenkins.

• Now creating Jenkins job to automatically trigger and download the files from GitHub repository .

• Now the Job is created ..when the developer push code to the GitHub ..it will automatically trigger and download the code in the workspace folder .
• You can get the code "main.py" from my GitHub repository .

Note : I have manually copied my dataset to "/root/ws/" for the code to automaticaly load it . (Don't waste internet by pushing in GitHub and Pulling it on workspace..)

Job2 : Launch the image based on the ML code.

• Here I have used the same image for ML Models (CNN , ANN ) ....as CNN only requires additional Keras and Tenserflow rest libraries are same for both .

sudo docker run -v /root/ws:/usr/src/mlops --name model act3:v1

Note : Job2 is Downstream it will automatically start when the job1 is finished i.e, Job Chaining .

• When the image is launched it will start training the model and predict accuracy or metrics.
• The Accuracy of the model is stored in "acc.txt" file .

Note : Log of the Model training can be accessibleby Console Oputput in Jenkins .

Job3 : if metrics accuracy is less than 80% , then tweak the machine learning model architecture (Tweaking Model).

filename="acc.txt"


while IFS= read -r line
do 

if [[ $line >= 0.8* ]] 
then
    printline="yes"       
    echo "-----------------------"
    echo "MODEL HAS TRAINED SUCESSFULLY"
    echo "-----------------------"   
else
	echo"MODEL ACCURACY IS LOW"
fi


if [[ $printline == "yes" ]] ;
	then
		echo "-----------------------"
		echo "MODEL ACCURACY IS $line"
		echo "-----------------------"         
else
		echo"MODEL ACCURACY IS LOW"
fi  


if [[ $line < 0.8* ]] 
then       
    printline="no"
    if sudo docker ps -a | grep model
    then
        sudo docker rm -f model 
        echo "MODEL REMOVED"
    else
        echo "TWEAK ERROR"
    fi
else
fi


if [[ $printline == "no" ]] 
then
    echo "-----------------------"
	echo "MODEL ACCURACY IS LESS $line"
	echo "-----------------------" 
	echo "TWEAKING MODEL" 


	sed -i '64i model.add(Dense(units=10,activation="relu",kernel_initializer="he_normal" )) ' main.py


	sudo docker run -v /root/ws/:/usr/src/mlops --name tuning act3:v1       
else
echo " Couldn't add Dense Layer"
fi




done < "$filename"

Note : Here you can add as per your requirement to tweak the model . I have done as per my understanding .. !! can add

• This is also a Downstream project it will be automatically triggered by job2.
• When the Tuning is done it will relaunch the job3 to check the accuracy .

Job4: Retrain the model or notify that the best model is being created

• If the Accuracy is Less than 80% then the tweak will launch the tuning model and the Developer will be notified by mail .

filename="acc.txt"


while IFS= read -r line
do 

if [[ $line < 0.8* ]] 
then
    exit 1 
else
	exit 0
fi

• If Accuracy is less then it will send the job failed mail to Developer & If it's a Success it will send the job success mail .
• Configure the E-mail Notification Plugin in the Jenkins .

Job5: Monitoring

• It will monitor the ML images if job2 or the tuning of the model job3 failed due to any error it will automatically start the model .

if ( sudo docker ps -a | grep model )
then 
echo "working "
else
sudo docker run -v /root/ws/:/usr/src/mlops --name model act3:v1
fi


if (sudo docker ps -a | grep tuning )
then 
    if sudo docker ps | grep model
    then
       sudo docker rm -f model
    else
       echo "tuned "       
    fi

elif (sudo docker ps -a | grep tuning)
then
    echo"working"
else
    sudo docker run -v /root/ws/:/usr/src/mlops --name tuning act3:v1
fi

• This will launch the images if it isn't running and relaunch the job3 and job4 for the model accuracy predictions and tweaking ( by using downstream or upstream based on your requirements ).

Outputs :

Implementation isn't time consuming ..requirements gathering is and everything depend on trial and error ..so keep trying ..!!

Well thanks to this activity I got to learn many new things by doing it practically & looking forward to learning by implementing various upcoming activities ... !!!

Hope you liked it .. Integration of Machine Learning Model with DevOps (Docker , Jenkins , GitHub ) .... \\(*^*)//