Data Requirements

The following data mining methods are well-suited for analyzing sales data. Typically, you will require transactional data that includes details such as the product purchased (or rented, or subscribed to), the date of the transaction, a customer identifier to track individual purchases over time, and the monetary amount associated with each purchase.

In an ERP (Enterprise Resource Planning) system, the Sales Order Details table generally contains all the necessary information. This table includes a unique key for the order, a key for the customer, a date or timestamp, as well as the line item code/description and the corresponding amount.

In POS (Point-of-Sale) data, you will find similar fields; however, the customer identifier may be represented by a loyalty card number and may only be available for certain transactions.

Transferring and transforming data from a source system to a data warehouse or platform, such as Snowflake, Amazon Redshift, or Google BigQuery, falls outside the purview of this documentation and, indeed, outside the scope of predictr.io.](http://predictr.io).) You can utilize various ETL (Extract, Transform, Load) tools for this task, and your selection will depend on the extent to which the data needs to be merely copied versus significantly transformed.

Data transformation and calculation is performed automatically- for example, given a transaction dataset with a customer ID and transaction date, predictr.io will automatically calculate the customer Recency (days since last transaction) where required.

Build Infrastructure

When a model is built, dedicated infrastructure is provisioned for the duration of the analysis creation or model build and is then disposed of. The infrastructure shares no data with any other build or data source and is completely removed once the model is saved.

Smaller infrastructure sizes launch more quickly so a general rule of thumb is to use the smallest size that fits your workload.

You may choose the size of the infrastructure that is launched. For many analysis types, pre-summarisation is performed by the data source, e.g. running one or more group-by style queries, to limit the data returned to the infrastructure during a model build. For this reason, performance is often more related to the performance of the data source than the model build environment.

For very large datasets, smaller types may not be capable of model building. If a model build fails without a specific error message, it may have ran out of memory and failed without even being able to report an error message, so it is always worth trying a larger one.

Available Infrastucture Types