filmov
tv
Building ML Blocks with Kubeflow Orchestration with Feature Store //Aniruddha Choudhury// Meetup #72
Показать описание
MLOps community meetup #72! Last Wednesday, we talked to Aniruddha Choudhury, Senior Data scientist at Publicis Sapient.
//Abstract
In this talk, Aniruddha is presenting the Kubeflow with feature store end-to-end training and serving the model, build the Lego blocks for online and offline data ingestion with spark and Kafka alongside Redis, monitor the metrics in Grafana and Prometheus, and the whole architecture building in GCP.
//Bio
Aniruddha is a Senior data scientist with experience over 5 years across various industry experiences for AI/ML journey. He is experienced in building the MLOps domain across various multi-cloud solutions for various AI synergies.
Aniruddha is a passionate AI researcher to transform the data into a machine learning end-to-end production journey to ease the life of Business users.
// Takeaways
How to set up Kubeflow in GCP and building Kubeflow pipeline.
Alongside how to use feast feature store and for training and serving.
// Other Links
--------------- ✌️Connect With Us ✌️ -------------
Follow us on Twitter: @mlopscommunity
Timestamps:
[00:00] Introduction to Aniruddha Choudhury
[02:56] Building blocks for productionizing ML projects with Feature Store in Kubeflow Orchestration
[03:16] Aniruddha's background in tech
[03:33] What the session will cover
[04:24] Touch base components
[05:02] Feature Store: Why?
[07:08] AI/ML Workflow Without Feature Store
[09:50] What is feast Feature Store?
[11:43] Five Components of Modern Feature Store
[13:48] FeatureLife Cycle with Feast
[15:01] Scenario: Churn Modeling
[16:17] What is Feast not?
[17:10] Feature Store Benefits
[18:29] High-Level Components
[19:20] Operational Flow of Feature Store with Feast
[21:23] Kubeflow + Kubernetes
[21:44] What and why Kubernetes?
[22:49] What is Kubeflow?
[24:27] Feature Store with Kubeflow
[25:10] Kubeflow with Feast Architecture
[26:55] Demo
[27:18] Demo Overview
[30:40] Kubeflow Environment
[32:12] Inference Layer
[32:17] High-Level Architecture
[34:08] Initialize all environment variables in the local
[35:24] Create a Staging Bucket
[37:41] Describe the entity
[39:37] Declare the identity
[40:51] Declare a schema
[42:40] Register the features of the data frame
[43:32] Check the schema chapter
[44:21] Ingest the data
[46:43] Start offline to online ingestion
[47:06] Declare the Kafka
[47:49] Declare the streaming source
[50:01] Save the data driver ID
[51:52] Train the data
[52:02] Save the artifacts
[52:53] Feature Store vs database where you store features
[55:36] Building the Kubeflow pipeline
[58:30] Initialize the environment and load the data from the artifacts
[59:22] Initialize feature store feast SDK and retrieve historical features
[59:59] Staging bucket
[1:00:35] Save the data in the file store
[1:01:50] Pass all the arguments
[1:02:25] Retrieve the data
[1:02:36] Run the model and save the logs
[1:03:00] Retrieve the CATIV artifacts
[1:04:08] Upload the model saved in local
[1:04:49] Publish the model waves to the GC buckets for prediction
[1:05:01] Go to the pipeline
[1:05:22] Max trial count
[1:07:30] Declare adjacent template
[1:08:13] Fit all trial specs into the trial template
[1:09:46] Feature Store retrieval
[1:10:27] Name the experiment
[1:10:46] Serialize
[1:11:21] Converting the results
[1:11:45] Evaluation
[1:13:14] Deployment
[1:14:10] Compile and upload the pipeline
[1:18:16] Predict the model to get the output
[1:21:10] Monitor the metrics
[1:22:19] Check the health metrics
//Abstract
In this talk, Aniruddha is presenting the Kubeflow with feature store end-to-end training and serving the model, build the Lego blocks for online and offline data ingestion with spark and Kafka alongside Redis, monitor the metrics in Grafana and Prometheus, and the whole architecture building in GCP.
//Bio
Aniruddha is a Senior data scientist with experience over 5 years across various industry experiences for AI/ML journey. He is experienced in building the MLOps domain across various multi-cloud solutions for various AI synergies.
Aniruddha is a passionate AI researcher to transform the data into a machine learning end-to-end production journey to ease the life of Business users.
// Takeaways
How to set up Kubeflow in GCP and building Kubeflow pipeline.
Alongside how to use feast feature store and for training and serving.
// Other Links
--------------- ✌️Connect With Us ✌️ -------------
Follow us on Twitter: @mlopscommunity
Timestamps:
[00:00] Introduction to Aniruddha Choudhury
[02:56] Building blocks for productionizing ML projects with Feature Store in Kubeflow Orchestration
[03:16] Aniruddha's background in tech
[03:33] What the session will cover
[04:24] Touch base components
[05:02] Feature Store: Why?
[07:08] AI/ML Workflow Without Feature Store
[09:50] What is feast Feature Store?
[11:43] Five Components of Modern Feature Store
[13:48] FeatureLife Cycle with Feast
[15:01] Scenario: Churn Modeling
[16:17] What is Feast not?
[17:10] Feature Store Benefits
[18:29] High-Level Components
[19:20] Operational Flow of Feature Store with Feast
[21:23] Kubeflow + Kubernetes
[21:44] What and why Kubernetes?
[22:49] What is Kubeflow?
[24:27] Feature Store with Kubeflow
[25:10] Kubeflow with Feast Architecture
[26:55] Demo
[27:18] Demo Overview
[30:40] Kubeflow Environment
[32:12] Inference Layer
[32:17] High-Level Architecture
[34:08] Initialize all environment variables in the local
[35:24] Create a Staging Bucket
[37:41] Describe the entity
[39:37] Declare the identity
[40:51] Declare a schema
[42:40] Register the features of the data frame
[43:32] Check the schema chapter
[44:21] Ingest the data
[46:43] Start offline to online ingestion
[47:06] Declare the Kafka
[47:49] Declare the streaming source
[50:01] Save the data driver ID
[51:52] Train the data
[52:02] Save the artifacts
[52:53] Feature Store vs database where you store features
[55:36] Building the Kubeflow pipeline
[58:30] Initialize the environment and load the data from the artifacts
[59:22] Initialize feature store feast SDK and retrieve historical features
[59:59] Staging bucket
[1:00:35] Save the data in the file store
[1:01:50] Pass all the arguments
[1:02:25] Retrieve the data
[1:02:36] Run the model and save the logs
[1:03:00] Retrieve the CATIV artifacts
[1:04:08] Upload the model saved in local
[1:04:49] Publish the model waves to the GC buckets for prediction
[1:05:01] Go to the pipeline
[1:05:22] Max trial count
[1:07:30] Declare adjacent template
[1:08:13] Fit all trial specs into the trial template
[1:09:46] Feature Store retrieval
[1:10:27] Name the experiment
[1:10:46] Serialize
[1:11:21] Converting the results
[1:11:45] Evaluation
[1:13:14] Deployment
[1:14:10] Compile and upload the pipeline
[1:18:16] Predict the model to get the output
[1:21:10] Monitor the metrics
[1:22:19] Check the health metrics
1:26:03
Building ML Blocks with Kubeflow Orchestration with Feature Store //Aniruddha Choudhury// Meetup #72
1:06:02
Building a Machine Learning Pipeline with Kubeflow | Full Walk-through
0:01:43
What is Kubeflow? // Aniruddha Choudhury // Meetup #72 short clip
0:25:41
ML Pipeline Using Kubeflow
0:33:31
Managing Machine Learning in Production with Kubeflow and DevOps - David Aronchick, Microsoft
0:03:45
What is Kubeflow?
0:29:53
Building ML Products With Kubeflow - Jeremy Lewi, Google & Stephan Fabel, Canonical
0:19:52
ML Ops design patterns with Kubeflow Pipelines - Amy Unruh, Google
0:34:22
DevOps for AI: running LLMs in production with Kubernetes and KubeFlow
0:10:27
Keynote: Platform Building Blocks: How to Build ML Infrastructure with CNCF Projects
0:26:13
Accelerate ML Model Development for Autonomous Vehicles in Aurora (Kubeflow Summit 2022)
0:03:21
How to Build Flexible, Portable ML Stacks with Kubeflow and Elastifile (Next Rewind '18)
0:19:52
Document processing with ML on Kubeflow
0:03:18
Training an ML model with Kubeflow
0:42:39
Kubeflow vs MLFlow
1:04:05
Yaron Haviv Simplify ML Pipeline Automation and Tracking using Kubeflow and Serverless Functions
0:36:08
Machine learning workflows using KubeFlow for the Hybrid Cloud
0:32:59
Chanchal Chatterjee Quickly Deploy ML Workloads on Multi Cloud Using Kubeflow Pipelines
0:13:20
Streamline ML Model Deployment with KubeFlow Pipelines!
0:24:23
How to Create and Deploy ML Pipeline using Kubeflow SDK | kubeflow | What is Mlops ?
0:52:36
Building Foundational ML Platforms with Kubernetes and Kubeflow with Ali Rodell - #595
0:38:37
Kubeflow: Machine Learning on Kubernetes | Machine Learning with Kubeflow | Intellipaat
0:02:39
Live on 16th May: KubeFlow for ML Ops [ Part-3 ]
0:59:45
Webinar: Planet-scale ML Workflows with Advanced Data Management on Kubeflow
Комментарии