Documentation

    Sample

    Keep the first, last, random or every Nth rows, or specific row numbers, as a count or a percentage, optionally within each group. Or balance classes: Undersample reduces every class to the smallest, Oversample duplicates rows up to the largest and Target ratio grows one class to a chosen share. Set Seed to get the same result each time.

    Take a few rows

    Manual Input has five people in this order: Lin, Sam, Bea, Ada and Noor. Sample starts in First rows mode with Amount set to 3, keeping Lin, Sam and Bea.

    The live example needs a desktop screen. The result it produces is shown here, and the explanation and settings below work on any screen.

    Result
    namescore
    Lin8
    Sam12
    Bea3
    1. Change Amount to 2. The output now contains Lin and Sam, the first two people in the input.
    2. Change Mode to Last rows. The output now contains Ada and Noor, in that order. Noor's blank score stays NULL.
    3. Change Mode to Random rows. Sample now selects two rows from anywhere in the table, and the Seed setting appears. Change Seed from 42 to 7 to try another selection.

    When to use

    Use Sample to work with part of a table: the first rows, the last rows, a random selection or every Nth row. It can also balance classes for a model. To select rows by a condition, use Filter.

    Read the output

    The example keeps three of the five input rows: Lin (8), Sam (12) and Bea (3). Both columns, name and score, keep their values and types. The other two rows are left out of Sample's output. The input table doesn't change.

    Sample has one output. It doesn't give you a separate table of the rows it leaves out.

    First and last follow input order

    First rows takes rows from the beginning of the input. Last rows takes them from the end and keeps their relative order. With Amount set to 2, Last rows returns Ada followed by Noor, not Noor followed by Ada.

    The score values don't decide which rows are kept. If the input order changes, First rows and Last rows can select different people.

    Use a seed for random selection

    Random rows selects from the whole input table. Each input row can be selected at most once. Don't rely on the sample keeping the input's row order.

    The seed makes a random selection repeatable for the same input and run conditions. Changing it can select different rows, but doesn't guarantee a completely different sample. Changes to the input can also change which rows are selected.

    To choose rows by a condition instead of a count, use Filter.

    Inputs and outputs

    Inputs

    1 port
    InRequired

    Accepts: Tables, Geospatial data.

    Accepts one connection on this port.

    Outputs

    1 port
    Out

    Configure the block

    Sampling

    Row sampling keeps a subset of rows. Class balance resamples so classes are evenly represented.

    Default: Rows

    Options: Rows, Class balance.

    Mode

    First rows takes from the beginning of the input, Last rows takes from the end, Random rows selects from the whole table, Every Nth row keeps one row at a fixed interval and Row ranges keeps the row positions you list.

    Default: First rows

    Options: First rows, Last rows, Random rows, Every Nth row, Row ranges.

    Amount unit

    Interpret Amount as a row count or a percentage of rows.

    Default: Rows

    Options: Rows, Percent.

    Amount

    With Rows, enter a positive whole number; requests larger than the input return all available rows. With Percent, enter more than 0 and at most 100.

    Default: 100

    Step (N)

    Keep one of every N rows (the 1st, N+1th, 2N+1th,...).

    Default: 2

    Row ranges

    1-based row positions to keep, e.g. 1-10, 25, 50-60.

    Default: 1-10

    Class column

    Column whose values define the classes to balance.

    Method

    Undersample keeps each class down to the smallest class. Oversample duplicates rows so each class matches the largest class. Target ratio grows one class to a chosen share.

    Default: Undersample

    Options: Undersample, Oversample, Target ratio.

    Target class

    The class value to grow (matched as text against the class column).

    Target share (%)

    Desired share of the output made up by the target class (0-100, exclusive). Only grows the class; it's a no-op if the class already exceeds this share.

    Default: 50

    Seed

    Repeats a selection for unchanged input, row order and execution conditions. Changing the seed can select different rows, but samples may overlap.

    Default: 42

    Group By

    Optional columns that sample within each group (e.g. first 5 per customer, 10% per region). Leave empty to sample over all rows.

    Sample FAQs

    What if Amount is larger than the input table?
    Sample returns all available rows. It doesn't add rows or repeat them to reach the requested count.
    Can I set Amount to zero?
    No. With Rows, Amount must be a positive whole number. With Percent, it must be more than 0 and at most 100. Zero and negative values produce an error.
    Does Sample remove rows with missing values?
    No. Missing values don't affect whether a row is selected. For example, Last rows can include Noor even though Noor's score is NULL.
    Does Random rows remove duplicate records?
    No. It selects input rows without replacement, but different input rows can contain the same values. Sample doesn't deduplicate the data.

    Prepare your data. Build it in Strata.

    Strata opens in your browser, with no account.