Google Cloud Data Practitioner

Dirty data ruins good analysis. You choose between ETL and ELT, assess data quality, and clean missing values, duplicates and inconsistent formats with SQL and simple tools.

• ETL versus ELT and when each fits
• Assessing data quality and completeness
• Handling nulls, duplicates and outliers
• Standardizing formats and types
• Cleaning with SQL in BigQuery
• Documenting what you changed and why

You build A cleaned dataset with a short log of every change and the reason for it.

Where data lands affects cost and speed later. You pick Cloud Storage or BigQuery for each dataset, choose a file format, and use the right transfer tool for files, databases and software services.

• Cloud Storage versus BigQuery for different data
• File formats: CSV, JSON, Avro and Parquet
• Load jobs into BigQuery
• Storage Transfer Service and BigQuery Data Transfer Service
• Loading from databases and other clouds
• Checking that a load worked

You build Three datasets loaded from different sources with validation checks.

SQL is the working language of analysis. You write queries that join, aggregate and rank data, understand how BigQuery charges for queries, and use partitions and views to keep them cheap.

• Selecting, filtering and joining tables
• Aggregation and grouping
• Window functions for ranking and running totals
• Views and saved queries
• Partitioned and clustered tables
• Estimating and controlling query cost

You build A set of queries that answer five business questions, with their costs recorded.

Some questions need more than SQL. You use notebooks in Colab Enterprise to pull BigQuery data into Python, explore it, chart it and share the analysis with colleagues.

• Starting a Colab Enterprise notebook
• Reading BigQuery data into Python
• Exploring data with summary statistics
• Charts for exploration
• Mixing SQL and Python
• Sharing and reproducing notebooks

You build A notebook that explores one dataset and states three findings.

Insight has to reach people who do not write queries. You choose the right chart, build dashboards in Looker with filters and clear measures, and share them with the right audience.

• Choosing charts for the question
• Measures, dimensions and filters
• Building dashboards in Looker
• Explores and reusable definitions
• Scheduling and sharing
• Making dashboards readable and honest

You build A dashboard for one team with filters, clear measures and a short guide for readers.

Basic predictions are within reach of a practitioner. You frame a prediction task, train and evaluate a model with BigQuery ML, try AutoML, and keep models organized in the Model Registry.

• Framing a prediction task
• Creating and training a BigQuery ML model
• Evaluating model results
• Making predictions with a trained model
• AutoML for tabular data
• Organizing models in the Model Registry

You build A trained model that predicts one outcome, with an evaluation summary and registry entry.

Data work should repeat without you. You design a basic pipeline, schedule queries and workflows, trigger processing when new files arrive, and monitor runs with Dataflow and Cloud Composer.

• Designing a simple pipeline
• Scheduled queries in BigQuery
• Cloud Composer workflows at a basic level
• Dataflow for simple processing
• Event-driven ingestion with Eventarc
• Monitoring runs and handling failures

You build A scheduled pipeline that loads new files and refreshes a table, with failure alerts.

Data is only useful if the right people can see it. You grant least-privilege access, restrict sensitive columns and rows, and share datasets inside and outside the organization with Analytics Hub.

• IAM roles for data users
• Dataset, table and view-level access
• Column and row-level restrictions
• Sharing data with Analytics Hub
• Documenting and cataloging datasets
• Reviewing who has access

You build An access plan for a dataset with restricted columns and a shared view.

Data has to be kept, protected and eventually removed. You use storage classes and lifecycle rules, plan backup and recovery, and choose encryption options that match the sensitivity of the data.

• Storage classes and lifecycle rules
• Retention and deletion policies
• BigQuery time travel and table snapshots
• Backup and recovery planning
• Default and customer-managed encryption
• Privacy measures for personal data

You build A lifecycle and protection plan for three datasets with different sensitivity.

Take data from raw files to a shared dashboard. You load and clean it, analyze it, add a simple model, automate the refresh, and set access, then explain the choices you made.

• Defining the question and the data
• Loading and cleaning
• Analyzing and visualizing
• Automating refresh and setting access
• Presenting the result

You build A governed, automated dataset and dashboard that answers one business question.