For smaller datasets, the Transformer page displays the entire dataset. For larger ones, the source data is sampled for use in the Transformer page.
At the top of the Transformer page, the type of the current sample is displayed next to the dataset name. To open the Samples panel, click the link. In the example below, the Full Data link indicates that the current sample in the Transformer page is the entire dataset:
Figure: Click the Samples link.
The Samples panel is displayed on the right side of the screen:
Figure: Samples Panel
At the top of the panel, you can review the currently loaded sample. Each user has his own active sample on a dataset.
Initial: By default, the application loads the first N rows of the dataset as the initial sample when the Transformer page is opened. The number of rows depends on column count, data density, and other factors. If the dataset is small enough, the full dataset is used.
NOTE: By default, samples may be up to 10 MB in size. For datasets smaller than this limit, the entire dataset is loaded.
Click the link in the current sample card to see the list of all available samples.
Below the current sample, you can review the available options for creating new samples. Each type of sample reflects a different method of collection.
To collect a new sample, click the appropriate sample card. See below.
To cancel a sample collection, click the X next to the progress bar. The interrupted sample is listed as unavailable. You can download the logs from the unfinished sample collection.
To view the sampling job on Cloud Dataflow, click the icon next to the progress bar.
- After a sample is created, you can load it at any time, as long as it is still valid. Next to a collected sample, click Load sample.
- For more information on sampling methods, see Overview of Sampling.
At the bottom of the Transformer page, you can review the number of rows and columns and count of data types in the currently displayed sample.
NOTE: As you add transform steps to your recipe, the values in the status bar change to reflect the current state of the loaded sample.
NOTE: Some operations, such as
union, may change the row counts without invalidating the sample. If the operation increases the size of the dataset beyond the sample size limit enforced by the application, then a subset of those rows is displayed. This is a known issue.
Collect new sample
When a new sample is collected, it is gathered based on the current location in the recipe when the sample is gathered. So, if the recipe contains steps that join in other datasets, those joins are performed to bring together the data from which the sample is executed.
NOTE: Except for the initial sample, all samples are generated based on the steps leading up to the location of the cursor in the recipe. If earlier steps are deleted or modified, the collected sample can be invalidated.
NOTE: When sampling from compressed data, the source is uncompressed, and a new sample of it is loaded into the data grid. As a result, the sample size you see in the grid corresponds to the uncompressed data.
- In the Samples panel, select the type of sample to create. For more information on sample types, see Overview of Sampling.
In the Collect new sample panel, specify the following parameters, some of which may not be required for your sampling method:
Choose a sampling method: Select or enter the type of sample. If you already selected a sampling method, this value is pre-populated for you.
Scan Type: (Does not apply to all sampling methods) Types of scans:
Quick- performs a random scan of the dataset to extract the appropriate number of rows for the sample
Full- gathers the sample from the entire dataset. Depending on the size of the dataset, this method can take a while.
- Column or columns: (Stratified, Cluster-based) Name of the column from which to gather values to evaluate (Anomaly-based) Specify the name or names of one or more columns containing the anomalies to include in your sample. Multiple columns can be specified by comma-separated values. A column range can be specified using the tilde (
Condition: (Filter-based, Stratified, Cluster-based, Anomaly-based) Filter the sample based on a specified condition. For example:
invoiceDate > 90
- Anomaly type: (Anomaly-based) Select the type of anomalous values to include in your sample: invalid, missing, or both types.
- To begin collecting the sample, click Collect.
- You can continue working while the sample is collected. When the sample is available, a status message is displayed in the Transformer page.
- You can click Load Sample in the Samples panel to begin using it.
In the Collected samples panel, you can review the available and unavailable samples.
To use one of the available samples, select its card. The sample is loaded in the data grid.
NOTE: If you add recipe steps that change the number of rows in your dataset (or a few other edge case steps), some of your existing samples may no longer be valid. When you execute a join, union, or delete action or edit steps before this action, you may be prompted with the Change Recipe dialog, which includes the following message:
Your change will invalidate some of the currently available samples for this source. The invalid samples will be deactivated.
For more information on the types of transformations that can invalidate samples, see Reshaping Steps.