Using Sample Data for Software Testing

I've collected my 7 favorite test data generation techniques that save hours of manual work: - 𝗕𝗼𝘂𝗻𝗱𝗮𝗿𝘆 𝗩𝗮𝗹𝘂𝗲 𝗔𝗻𝗮𝗹𝘆𝘀𝗶𝘀: Testing at the edges (min, min+1, max-1, max) reveals more bugs with fewer test cases. I once found a critical payment processing bug by testing exactly at the $10,000 threshold—something that wouldn't have surfaced with random values. - 𝗗𝗲𝗰𝗶𝘀𝗶𝗼𝗻 𝗧𝗮𝗯𝗹𝗲𝘀: Perfect for complex business logic with multiple conditions. I use these for tax calculation engines, discount rules, and permission systems to ensure comprehensive coverage. - 𝗘𝗾𝘂𝗶𝘃𝗮𝗹𝗲𝗻𝗰𝗲 𝗣𝗮𝗿𝘁𝗶𝘁𝗶𝗼𝗻𝗶𝗻𝗴: Group similar inputs to reduce redundant test cases while maintaining coverage. This trimmed a 100-case test suite down to 35 cases without sacrificing quality for a previous client. - 𝗦𝘆𝗻𝘁𝗵𝗲𝘁𝗶𝗰 𝗗𝗮𝘁𝗮 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝗼𝗻: Tools like Mockaroo or https://fakerjs.dev/ create realistic test data sets in seconds. I generated 5,000 realistic user profiles with international addresses and phone numbers in less than 5 minutes. - 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗗𝗮𝘁𝗮 𝗔𝗻𝗼𝗻𝘆𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻: Real patterns without privacy risks using data masking tools. This preserves the complexity of real-world data while eliminating PII concerns. → Tools: open-source Gretel for smaller teams - 𝗣𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿𝗶𝘇𝗮𝘁𝗶𝗼𝗻: One test script + data file = hundreds of test scenarios. This approach cut our regression suite development time by 70% on my last project. → JUnit Parameterized Tests (https://lnkd.in/e4yG-iCp) offers similar functionality ,RestAssured (https://rest-assured.io/) with CSV Data for API testing - 𝗣𝗿𝗼𝗽𝗲𝗿𝘁𝘆-𝗕𝗮𝘀𝗲𝗱 𝗧𝗲𝘀𝘁𝗶𝗻𝗴: Let the computer generate thousands of inputs to find edge cases you'd never imagine. This caught a date parsing bug that only occurred on leap years on February 29th—something we'd never have explicitly tested for. → Tools: jqwik (https://jqwik.net/) for Java, fast-check (https://lnkd.in/e5P68rFj) for JavaScript, Hypothesis (https://hypothesis.works/) for Python 𝗕𝗢𝗡𝗨𝗦: Check out Pairwise Testing (https://lnkd.in/ecqyQ8_u) to dramatically reduce test combinations while maintaining coverage of interactions between parameters. These techniques have helped me design more effective test cases in less time. Which technique do you use most often? Or is there another tool I should add to my list? #TestersLife #SoftwareTesting #TestDataGeneration

Data integrity + realistic testing = Project success. 💯 My current mission: Developing a data simulation app to stress-test a complex ArcGIS Experience Builder application for Gotham city. The trick? The real dataset has huge image files. To check website behavior and performance properly, I'm using the Unsplash API to dynamically pull size-matched image samples that act like the originals. Proving the new Experience Builder can handle the load before launch. No private data touched, all performance insights gained. Win-win.   Leveling Up: High-Fidelity Data Simulation for Natural Resources 🌳 This approach of creating a high-fidelity synthetic dataset is vital in the Natural Resources sector, particularly for monitoring applications. Imagine a Forest Inventory application built on ArcGIS Experience Builder. Foresters need to view records of trees, each linked to large, recent aerial photos or hyperspectral imagery for health assessment. To stress-test this application's performance on a wide scale without using sensitive proprietary data, we could replicate the scenario. Instead of Unsplash, we'd use the NASA Earthdata API or the Sentinel Hub API 🛰️ to programmatically pull and integrate publicly available, size-and-resolution-matched satellite imagery (like Sentinel-2 or Landsat) to simulate the immense data load. This ensures the system runs smoothly when field teams are accessing critical asset information in remote areas. #ArcGISExperienceBuilder #APIIntegration #SoftwareTesting #Geospatial #NaturalResources #RemoteSensing #DataScience #EarthObservation

A good practice for working with data transformations in a development environment is to use a sample of your data. Instead of running your modifications against the entire table, you can save time and resources by running them against a random or deterministic subset of your data. Random Sample: A randomly selected subset of your data. Each record has an equal chance of being included, which helps ensure that the sample is representative of the whole dataset. This method is useful for getting a general idea of how your transformation will perform on diverse data. Some data warehouses have their sample functions. - Snowflake: SELECT * FROM example_table SAMPLE (10) - BigQuery: SELECT * FROM dataset.my_table TABLESAMPLE SYSTEM (10 PERCENT) - Databricks: SELECT * FROM test TABLESAMPLE (30 PERCENT) Deterministic Sample: A subset of your data selected based on specific criteria or rules. For example, you might choose records from a particular date range or those that meet certain conditions. This method is useful when you want to consistently reproduce the same sample for testing or focus on a specific data segment. - SELECT * FROM customers WHERE purchase_date >='2024-06-01' #dbt has an open issue for the v1.9 milestone to add sampling to dbt as a built-in feature. https://lnkd.in/dckM2N-K In the meantime, it is possible to use it with dbt with a custom materialization or add an if/else Jinja block in your model Follow me for daily dbt content 🔶 #sql #dataengineering #analyticsengineering