Skip to main content
The Alteryx Oversample Field tool adjusts class imbalance in categorical fields (commonly for predictive modeling). It increases the proportion of a minority class or balances classes by randomly selecting additional records from one or more categories. Prophecy does not provide a single oversampling gem. Instead, Import reproduces the behavior using a combination of WindowFunction gems, Reformat gems, and Filter gems. This sequence computes class counts, assigns random values, ranks rows, and filters to achieve the requested oversample percentage.

​
Automated migration results

When Import encounters an Oversample Field tool, it generates a sequence of transformations equivalent to the Alteryx sampling logic:
  • A WindowFunction gem computes class-level counts (such as the number of Default = 'Yes' rows).
  • A Reformat gem assigns deterministic or random values used to order rows.
  • A second WindowFunction gem ranks rows using ROW_NUMBER() to support sampling.
  • A Filter gem keeps all minority-class rows, and a sampled subset of the remaining rows.
  • A final Reformat gem removes helper metadata columns.
This import pattern is highly consistent: most Oversample Field configurations map to the same sequence.

​
Manually replicate in Prophecy

To manually recreate oversampling logic:
  1. Add a WindowFunction gem, add a target column (such as num_oversample_values) under Window Use, and define the column with a count expression with custom code along the following lines: CASE WHEN (Default = 'Yes') THEN true END.
  2. Add a Reformat gem to assign a random value using rand() for majority-class rows and a fixed value (such as 1) for minority-class rows.
  3. Add a WindowFunction gem to compute a row number ordered by the random value.
  4. Add a Filter gem to keep all minority-class rows, and majority-class rows whose row number is below a threshold (such as “equal to the minority-class count”).
  5. Optionally, add a final Reformat gem to drop helper ranking and random columns.

​
Configuration options

In Alteryx, the Oversample Field tool is configured with three main parameters:
  • Field to oversample on (e.g., Default)
  • Field value to oversample (e.g., Yes)
  • Target percentage of records that should have the selected value (e.g., 40%)
Conceptually, Alteryx keeps all rows where the field equals the selected value, and randomly selects enough other rows so that the chosen value represents approximately the requested percentage of the output. In Prophecy, the same behavior is implemented explicitly in SQL:
  • A WindowFunction gem computes the count (or proportion) of rows where the field equals the oversampled value.
  • A Reformat gem assigns random values to non-oversampled rows and may tag oversampled rows with a fixed value.
  • Another WindowFunction gem ranks rows by the random value.
  • A Filter gem keeps:
    • all rows where the field equals the oversampled value, and
    • only as many other rows as needed to reach the target percentage.

​
Output behavior

Alteryx returns a table with adjusted class proportions according to the oversampling strategy. Prophecy returns a table filtered according to SQL conditions:
  • Minority-class rows are always retained (unless configured otherwise).
  • Majority-class rows are included only when they satisfy the sampling threshold.
  • Random functions execute at runtime, so preview results may differ slightly from run results.

​
Known caveats

None.

​
Example

​
Alteryx Oversample Field example

Goal: Increase the representation of rows where Default = 'Yes'.

​
Prophecy equivalent (WindowFunction + Reformat + Filter)

  1. Use a WindowFunction gem to compute the number of Default = 'Yes' rows.
  2. Use a Reformat gem to assign:
    • 1 for Default = 'Yes',
    • rand() for other rows.
  3. Use a WindowFunction gem to compute a row number ordered by the random value.
  4. Use a Filter gem to keep:
    • all Default = 'Yes' rows, and
    • majority-class rows with row_number <= num_yes.
  5. Remove helper columns in a final Reformat gem.
This sequence reproduces the oversampling behavior of the Alteryx tool using SQL-native operations.