Why CI/CD differs for ML vs software in MLOps - Visual Breakdown

Process Flow - Why CI/CD differs for ML vs software

Start: Code Commit

↓

Software CI/CD

↓

Build -> Test -> Deploy

↓

Production Software

↓

ML CI/CD

↓

Data Collection -> Model Training

↓

Model Validation -> Deployment

↓

Production Model & Monitoring

Shows two parallel flows: software CI/CD is code-focused with build-test-deploy, while ML CI/CD adds data and model training steps before deployment.

Execution Sample

MLOps

# Software CI/CD steps
code_commit()
build()
test()
deploy()

# ML CI/CD steps
collect_data()
train_model()
validate_model()
deploy_model()

This code outlines the main steps in software CI/CD versus ML CI/CD pipelines.

Process Table

Step	Process	Action	Focus	Outcome
1	Software CI/CD	code_commit()	Code changes	Trigger pipeline
2	Software CI/CD	build()	Compile code	Build artifact
3	Software CI/CD	test()	Run tests	Verify code correctness
4	Software CI/CD	deploy()	Deploy app	Release software
5	ML CI/CD	collect_data()	Gather data	Prepare dataset
6	ML CI/CD	train_model()	Train model	Create model artifact
7	ML CI/CD	validate_model()	Evaluate model	Check model quality
8	ML CI/CD	deploy_model()	Deploy model	Release model to production
9	ML CI/CD	monitor_model()	Monitor performance	Detect model drift
10	End	Condition	All steps done	Pipeline complete

💡 All pipeline steps completed for both software and ML CI/CD

Status Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	After Step 4	After Step 5	After Step 6	After Step 7	After Step 8	After Step 9	Final
Code	Initial code	Committed	Built	Tested	Deployed	Deployed	Deployed	Deployed	Deployed	Deployed	Deployed
Data	Raw data	Raw data	Raw data	Raw data	Raw data	Collected	Collected	Collected	Collected	Collected	Collected
Model	None	None	None	None	None	None	Trained	Validated	Deployed	Monitored	Monitored

Key Moments - 3 Insights

Why does ML CI/CD include data collection and model training steps unlike software CI/CD?

ML CI/CD pipelines must handle data and model lifecycle, not just code. Execution table rows 5 and 6 show data collection and model training steps unique to ML.

Why is model validation a separate step in ML CI/CD?

Model validation ensures the model works well before deployment, unlike software tests which run earlier. See execution table row 7 for this distinct step.

Why is monitoring important after ML model deployment but not always shown in software CI/CD?

ML models can degrade over time due to data changes, so monitoring (row 9) detects this drift, which is less common in traditional software.

Visual Quiz - 3 Questions

Look at the execution table, at which step does the ML pipeline first produce a model artifact?

A Step 5

B Step 7

C Step 6

D Step 8

💡 Hint

Check the 'Action' and 'Outcome' columns for model creation in ML CI/CD steps.

According to the variable tracker, what is the state of 'Model' after step 7?

A None

B Validated

C Trained

D Deployed

💡 Hint

Look at the 'Model' row under 'After Step 7' in the variable tracker.

If monitoring was removed from the ML pipeline, which step would be missing from the execution table?

A Step 9

B Step 7

C Step 8

D Step 6

💡 Hint

Look for the step related to monitoring in the execution table.

Concept Snapshot

CI/CD for software focuses on code build, test, and deploy.
ML CI/CD adds data collection, model training, validation, and monitoring.
ML pipelines handle data and model lifecycle, not just code.
Monitoring is critical for ML to detect model drift.
Each step ensures quality before production release.

Full Transcript

This visual execution compares CI/CD pipelines for software and machine learning. Software CI/CD runs code commit, build, test, and deploy steps focused on code changes. ML CI/CD includes extra steps: collecting data, training a model, validating it, deploying, and monitoring performance. Variables like code, data, and model change state at different steps. Key differences are the handling of data and models in ML pipelines and the need for monitoring after deployment to catch model drift. The execution table and variable tracker show these steps clearly, helping beginners see why ML CI/CD is more complex than traditional software CI/CD.