This section describes techniques to normalize numeric values in your datasets.
Ideally, your source systems are configured to capture and deliver data using a consistent set of units in a standardized structure and format. In practice, data from multiple systems can illuminate differences in the level of precision used in numeric data or differences in text entries that reference the same thing. Within , you can use the following techniques to address some of the issues you might encounter in the standardization of units and values for numeric types.
In , mathematical computations are performed using 64-bit floating point operations to 15 decimals of precision. However, due to rounding off, truncation, and other technical factors, small discrepancies in outputs can be expected. Example:
Suppose you apply the following transformation:
The expected output in the MySum column: -3101160.56
The actual output for in the MySum column: -3101160.5599999996
NOTE: For 64-bit floating point mathematical operations, deviations like the above are intrinsic to the Decimal data type and how the platform performs computations.
Depending on your precision requirements, you can manage precision across your columns using a transformation like the following, which rounds off MySum to three digits:
For more information on floating point computations, see https://en.wikipedia.org/wiki/Numeric_precision_in_Microsoft_Excel.
If decimal values in a column are of varying levels of precision, you can standardize to a single level of precision.
.In the Column Details panel, select the Patterns tab.Among the patterns, select the following:
ROUNDfunction. Click Edit.
ROUNDfunction to match the number of digits of precision.
You can generalize this formatting across multiple columns by applying the
$col reference in the transformation's function, as in the following:
Tip: Each column that contains numeric values should have an identified unit of measurement. Ideally, this information is embedded in the name of the column data. If the unit of measurement is not included, it can be difficult to properly interpret the data.
does not impose any units on imported data. For example, a column of values in floating point format could represent centimeters, ounces, or any other unit of measurement. As long as the data conforms to the specified data type for the column, then can work with it.
However, this flexibility can present issues for users of the dataset. If data is not clearly labeled and converted to a standardized set of units, its users are forced to make assumptions about the data, which can lead to misuse of it.
Tip: The meaning of some units of measure can change over time. For example, a US Dollar in 2010 does not have the same value as a dollar in 2015. When you standardize shifting units of measure, you should account for any time-based differences, if possible.
In many cases, units can be converted to other units by applying a fixed conversion factor to a column of data. For example, your dataset has the following three columns of measured data:
The above data has the following issues:
Problem 1 - remove units
Weight_kg column contains a unit identifier. On import, these values are treated as strings, which limits their use for analysis.
The transformation should look like the following:
Decimal, depending on the values in it.
Problem 2 - convert English to metric units
To normalize to English units, the first issue is easily corrected by multiplying the Weight values by 2.2, since 1 kg = 2.2 lb:
If you want to round the value to the nearest integer, use the following:
After the above is added to the recipe, you should rename the column:
Problem 3 - convert ft/in to in
The final issue involves converting the
Height_ft values to a single value for inches, so that these values can be used consistently with the other columns in the dataset.
On import, your data for the column might actually look like the following:
Select the variant that deletes all quotes in the column.
The full command should look like the following:
Derive the value in inches:
In some cases, the conversion rate between two different units of measures is dynamic. A common example involves mismatches between currency. For example, one dataset can be using U.S. dollars while another represents values in Euros.
If you have inconsistent units within a column, it might be possible to correct these values by applying a multiple. For example, you might be able to determine that some values are in kilometers, instead of meters, based on their much smaller values. Multiplying the kilometer values by 1000 should standardize your units. The following multiplies all values in the column
Distance that are less than 1000 by 1000.
Note the implied assumption that there are no distances in kilometers that are over 1000.
NOTE: Inconsistency in units within a column indicates a problem in either the source data or how the column data was modified after import. Where possible, you should try to fix these issues in the source data first, as they can introduce problems when the data is used.
For numeric values that are used for measurement, you can adjust the level of precision within and across columns of values. For example, you have the following columns of data:
In the above, you can see the following precision mismatches:
Where precision in measurement is important, you should consider rounding to the lowest level of precision. In this case, within the Height column, that level is to two significant digits after the decimal point (e.g.
12.22). However, across all of the columns of the dataset, the level of precision is to one significant digit after the decimal point, as the Width values are all restricted to this level of precision. While you could choose to round off to four digits across all columns, the extra values of
0 do not accurately reflect measurement and are therefore misleading.
You can use the following transformations to perform rounding functions within these columns:
NOTE: The above assumes that the number of significant digits remains fixed in the source data. If this varies over times or uses in your recipe, you might need to revisit these specific transformation steps.
NOTE: The above formatting option drops the zero for values like
For data hierarchies, you can use aggregations to adjust the granularity of your data to the appropriate grouping level. For example, you want to join a dataset that is organized by individual products with a dataset that is organized by brand. In most cases, you should aggregate the product-level data in the first dataset to the brand level.
NOTE: When aggregation is applied, a new table of data is generated with the columns that you specifically select for inclusion.
For more information, see Pivot Data.