- MLOps Consists of ML system development (Dev) and ML system operation(Ops).
- ML ops does automation and monitoring for machine learning models.
- It involves construction, integration, testing, releasing, deployment and infra management of ML system.
- Apart from unit and integration testing , data validation, model quality evaluation and model validation is required.
- Even after deployment model needs to be trained continuously.
- Model performance in production needs to be monitored because data profile can change overtime. e.g fashion trends data.
- Data extraction : extract data from various sources
- Data Analysis: explore data analysis, understand data characterstics, prepare data , feature engineering
- Data preparation: clean, split data into train,test and validation set.
- Model training: tuning model
- Model evaluation: evaluate on test set
- Model validation: check predictive performance
- Model serving : deploy model to target env
- Model monitoring: monitor the model performance to replace model
-
Model is retrained and validated whenever chnges are pushed or merged to master branch. And latest model is deployed to heroku.
-
Model is served as an API at
https://mlops-api.herokuapp.com/ -
Models are saved in
modelsfolder file name contains timestamp. -
Latest model version is in file
models/version -
Models matrics are saved in
models/matrics -
File paths are defined in
config.py -
data files are stored in
datafolder. -
To run validation/prediction on new dataset update
VALIDATION_DATA_FILEinconfig.py. -
Put the data file in data folder.
-
Then run
python predict.py.