MLOps
1.MLOps - Introduction
The requirements surrounding ML infrastructure
2.MLOps - Deployment
Key challenges in "Deploy in production"Softwware Engineering Realtime or BatchCloud dvs Edge/BrowserCompute resources (CPU/GPU/memory)Latency, throug
3.MLOps - Modeling
AI system = Code(algorithem/model) + DataModel-centric AI deploymentData-centric AI deploymentKey Challenges - why low average error isn't good enough
4.MLOps - Label and Organise data
Obtaining dataQuestion: How much data can we obtain in k days?Unless you know how many examples you need.Brainstorm list of data sources.Labeling data
5.MLOps - Data iteration
Data-centric AI developmentA useful picture of data augmentationGetting more data with Cafe noise and improve the model can result in improvement on o
6.MLOps - Define Data and Establish Baseline
More label ambiguity examplesIs it a bot or spam account?Is it a fradulent transaction?Is he/she looking for job?Are the two users the same person? (n
7.MLOps - Scoping
What is scoping?Example: Ecommerce retailer looking to increase salesBetter recommender systemBetter searchImprove catalog dataInventory managementPri
8.[MLflow] Intro
MLflowTracking: 실험에서 사용된 코드, 파라미터, 메트릭 등 실험에서 생성되는 모든 결과를 기록하고 조회할 수 있음. Python, REST, R, JAVA API를 제공함.Project: 어떤 플랫폼에서도 실험이 재현 가능하도록 패키징하는 방법을 제공함.
9.[MLflow] Experiments
10.[MLflow] Runs
Ways to run MLflow You can explicitly specify startrun() and endrun() Or you can use with to start and end the run Run ID: 4e24e3f695574503bd4
11.[MLflow] Logging params, metrics, artifacts, and etc
Logging hyper-parametres, metrics, artifacts, images and models Logging hyper-parametres Experiment ID: 446269725962928702 Run ID: 52f1730fe
12.[MLflow] Logging models
Logging images and models Logging image files Name: MyFirstMLflow C:\Users\dof07\AppData\Local\Temp\ipykernel_8984\2191112145.py:48: UserWarning: No artists with labels found to put in lege...
13.[MLflow] Nested Runs
Nested MLflow Runs For any child run, you have to specify nested=True Experiment ID: 838942086944409807 RUN ID parent: 339f8840eb8a40a68648e5214702cb50 RUN ID child1: 877588a82e0f4af2b63...
14.[MLflow] Custom Models
Custom Models When you are building your own custom model and it does not fit into any categories of modoling framework that MLflow provides, you can
15.[MLflow] Model Signiture(Schema)
Model Signiture with MLflow Model signiture is an object that allows us to specify the data type and shape that the model can work on. You can build a model signiture using mlflow.types.schema ...
16.[MLflow] Serving Multiple Models and Making Online Inference
mlflow.pyfunc as you are using a custom modelmodel_3 doubles the input and return as prediction
17.[MLflow] Model Registry
Centralized Model StoreModel LineageModel VersioningStage TransitionMOdel Annotations※ NOTE: As of 2025, stage has been removed from MLflow.The first
18.[MLflow] Model Registry Operations 1
Creating a new version of the model(rft_modelx) with a new name called 'registered_model_1'It is basically registering a model if it is the first vers
19.[MLflow] Model Registry Operations 2
Creating a new version of the model(rft_modelx) with a new name called 'registered_model_2'It is basically registering a model if it is the first vers