Broadly speaking, data quality identifies the degree to which data is usable and responsive to your use case. When you assess data quality, you are designing tests to assess its suitability for generic usage and for your specific uses.
Data Quality Characteristics
Data quality covers the following characteristics:
- Completeness: values are present where they are needed and expected
- Accuracy: data is substantively free of errors
- Consistency: a dataset can be matched across different data sources of the enterprise
- Timeliness: data values are up-to-date
- Uniqueness: aggregate data are free from any duplication via filters or other transformations of source data
- Validity: data are structured based on an adequate and rigorous classification system
- Availability / Accessibility: data are made available to the relevant stakeholders
- Traceability: the history, processing and location of the data under consideration can be easily traced
When data is imported, the Trifacta attempts to infer the data types in the source and to type columns in the dataset accordingly. Type inference uses the first 20-25 rows of the initial sample to assess the appropriate data type to apply to the column. For more information, see Type Conversions.
Some imported data, such as relational tables, may include schema information to identify the data type of each column. In some cases you can disable type inferencing on imported data:
- Connections: As part of the definition of a connection, you can optionally choose to disable type inference. For more information, see Create Connection Window.
- Per-dataset: When you import a dataset, you can modify the import settings for the selected source to disable type inference. See Import Data Page.
To assist in your transformation efforts, you can assign a target schema for each recipe. This target schema is super-imposed on the columns of your data. Using visual tools to review differences and select changes, you can rapidly convert the structure of your dataset in development to meet the expected target schema. For more information, see Overview of RapidTarget.
In the Transformer page, you can use the available visual tools to review the data quality characteristics of the columns in your data. These data visualizations and type-based quality bars can assist in identifying potential anomalies and quality problems.
Data quality bar
At the top of each column, you can see a data quality bar, which uses the following color coding to validate the column values against the selected column type.
|green||Values that are valid for the current data type of the column|
|red||Values that are mismatched for the column data type|
|gray||Missing or null values|
Tip: Click any of the color bars to receive suggestions for transformations to add to your recipe. See Overview of Predictive Transformation.
Tip: You can change a column's data type in the column header. See Column Menus.
For more information, see Data Quality Bars.
In the column header, you can review the count and distribution of values in the column. A column's histogram can be useful for identifying anomalies or for selecting specific sets of values in the column for further exploration.
Tip: Click and drag over any set of values to receive suggestions for transformations to add to your recipe. See Overview of Predictive Transformation.
See Column Histograms.
Through the Column Details panel, you can explore the quality and distribution of the values in the column. The contents of the panel vary depending on the data type. For example, if the column is typed for Datetime values, then the Column Details panel includes information on the distribution of values across the days of the week and days of the month.
For all data types, you can review useful statistics on statistical quartiles, the uniqueness of values, mismatches, and outliers.
Tip: The Column Details panel is very useful for acquiring statistical information on column values in a visual format. Click any data quality bar to be prompted for suggestions of transformation steps. See Overview of Predictive Transformation.
For more information, see Column Details Panel.
Data Quality Rules
You can build rules that test aspects of data quality that are specific to your organization. These rules persist throughout your recipe development and are used to evaluate the results that you generate. For more information, see Overview of Data Quality Rules.
You can use the Standardization tool to standardized clustered sets of column values to values that are common and consistent throughout your enterprise's data. For more information, see Overview of Standardization.
Data Quality Functions
The following functions are available for assessing data quality.
The following functions measure counts of values within a column, optionally counted by group.
Generates the count of rows in the dataset. Generated value is of Integer type.
Generates the count of non-null rows in a specified column, optionally counted by group. Generated value is of Integer type.
Generates the count of distinct values in a specified column, optionally counted by group. Generated value is of Integer type.
Extracts the set of unique values from a column into an array stored in a new column. This function is typically part of an aggregation.
Computes the average (mean) from all row values in a column or group. Input column can be of Integer or Decimal.
Computes the sum of all values found in all row values in a column. Input column can be of Integer or Decimal.
Computes the minimum value found in all row values in a column. Input column can be of Integer, Decimal or Datetime.
Computes the maximum value found in all row values in a column. Inputs can be Integer, Decimal, or Datetime.
Computes the mode (most frequent value) from all row values in a column, according to their grouping. Input column can be of Integer, Decimal, or Datetime type.
Computes the minimum value found in all row values in a Datetime column.
Computes the maximum value found in all row values in a Datetime column.
Computes the most frequent (mode) value found in all row values in a Datetime column.
Statistical functions - single column
Variations in these functions:
- Some of these functions have variations that use the sample population method of computation.
- IF conditional functions can be used to compute statistical computations based on a condition.
Computes the variance among all values in a column. Input column can be of Integer or Decimal. If no numeric values are detected in the input column, the function returns
Computes the standard deviation across all column values of Integer or Decimal type.
Computes the median from all row values in a column or group. Input column can be of Integer or Decimal.
Computes a specified quartile across all row values in a column or group. Input column can be of Integer or Decimal.
Computes a specified percentile across all row values in a column or group. Input column can be of Integer or Decimal.
Computes an approximation for a specified quartile across all row values in a column or group. Input column can be of Integer or Decimal.
Computes an approximation for a specified percentile across all row values in a column or group. Input column can be of Integer or Decimal.
Statistical functions - multi-column
Computes the covariance between two columns using the population method. Source values can be of Integer or Decimal type.
Computes the correlation coefficient between two columns. Source values can be of Integer or Decimal type.
Data Quality in Job Details
When you run a job and generate results, you can review the the quality of the data of the generated output.
In parallel with executing the job, you can generate a visual profile of the generated results. This visual profile provides graphical representations of the valid and mismatched values against each column's data type, as well as indications about missing values in the output.
Tip: Visual profiles can be downloaded in PDF or JSON format for offline analysis.
Visual profiling is selected as part of the job definition process. See Run Job Page.
For more information, see Overview of Visual Profiling.
When visual profiling is enabled for your job, the Rules tab in the Job Details page contains the results of the data quality rules for the job's recipes applied across the entire dataset.
Tip: Data quality rules are available for download in JSON and PDF format. For more information, see Job Details Page.
For more information, see Job Details Page.
This page has no comments.