Machine Learning Deployments
- 1. STUDENT NAME: OMKAR GURAV ROLL NO. 30 (PIMPRI)
- 2. WHAT IS MODEL DEPLOYMENT? âą MAKING YOUR MACHINE LEARNING MODEL AVAILABLE FOR END USER. âą CREATE A PRODUCT BASED ON MODEL PREDICTION. âą PROVIDING SERVICE TO OTHER ORGANIZATIONS. âą GIVE PREDICTION/ CLASSIFICATION/ DETECTION AS A RESPONSE
- 3. WHY DEPLOYMENT IS IMPORTANT? âą USER SHOULD BE ABLE TO USE MODEL DIRECTLY. âą TO INCREASE THE ACCESSIBILITY. âą TO LET ALGORITHM WORK ON THE GO. âą DEPLOYMENT IS PORTABILITY.
- 4. HOW TO DEPLOY? âąCLOUD SERVICES âąWEB APPLICATIONS âąAPPLICATION PROGRAMING INTERFACE âąWEB SITES
- 5. CLOUD âą GOOGLE CLOUD PLATFORM âą AMAZON WEB SERVICES (EC2, SAGEMAKER, NETBEANS) âą MICROSOFT AZURE âą IBM BLUEMIX âą SALESFORCE EINSTEIN
- 6. CONTAINERS âą DOCKER âą KUBERNETS âą AWS FARGATE âą MS AZURE CONTAINER INSTANCE
- 7. SERIALIZATION/ DESERIALIZATION -: SAVING TRAINED MODEL IN FILE AND LOADING AT THE TIME OF PREDICTION :- >>> ACCURACY > PERFORMANCE âą PICKLING â PICKLE , JOBLIB âą WEIGHTS & BIAS â TENSORFLOW, KERAS, PYTORCH, MXNET
- 8. MORE TOOLS âąFLASK --- WEB API/APP âąDJANGO --- WEB SITES âąHEROKU --- FREE CLOUD SERVER âąGITHUB --- PLATFORM AS A SERVICE
- 9. WHY WE NEED CLOUD FOR DEPLOYMENT âą CLOUD SERVERS FOR HOSTING WEB APPLICATION, WEB SITES âą SIZE OF THE TRAINED ML/ DL MODELS. âą FOR VIRTUAL INFRASTRUCTURE
- 10. CAN WE DEVELOP ML MODELS ONLINE? âą GOOGLE COLAB â NOTEBOOK WITH GPU (FREE 12 HRS) âą KAGGLE KERNEL â DATASET WITH NOTEBOOK / IDE WITH GPU (FREE 30 HRS) âą AMAZON FORECAST â WITH DATASET âą MS AZURE NOTEBOOKS â IDEAL FOR ML PIPELINE
- 11. WHAT WE LEARNED ! âąUNDERSTOOD DIFFERENT ASPECTS OF PIPE LINE OF ML PROJECTS âąIMPORTANCE OF CLOUD âąBOTH DEVELOPMENT AND DEPLOYMENT ONLINE