Ter this article
APPLIES TO: SQL Server Analysis Services Azure Analysis Services
This topic describes some technical considerations to keep te mind when processing gegevens mining objects. For a general explanation of what processing is, and how it applies to gegevens mining, see Processing Gegevens Mining Objects.
Queries on the Relational Store during Processing
For gegevens mining, there are three phases to processing: querying the source gegevens, determining raw statistics, and using the monster definition and algorithm to train the mining monster.
The Analysis Services server issues queries to the database that provides the raw gegevens. This database might be an example of SQL Server 2017 or an earlier version of the SQL Server database engine. When you process a gegevens mining structure, the gegevens ter the source is transferred to the mining structure and persisted on disk ter a fresh, compressed format. Not every katern ter the gegevens source is processed: only the columns that are included ter the mining structure, spil defined by the bindings.
Using this gegevens, Analysis Services builds an index of all gegevens and discretized columns, and creates a separate index for continuous columns. One query is issued for each nested table to create the index, and an extra query vanaf nested table is generated to process relationships inbetween each pair of a nested table and case table. The reason for creating numerous queries is to process a special internal multidimensional gegevens store. You can limit the number of queries that Analysis Services sends to the relational store by setting the server property, DatabaseConnectionPoolMax. For more information, see OLAP Properties.
When you process the prototype, the proefje does not reread the gegevens from the gegevens source, but instead gets the summary of the gegevens from the mining structure. Using the cube that wasgoed created, together with the cached index and case gegevens has bot cached, the server creates independent threads to train the models.
Processing Mining Structures
A mining structure can be processed together with all dependent models, or separately. Processing a mining structure separately from models can be useful when some models are expected to take a long time to process and you want to defer that operation.
If you are worried about conserving hard disk space, note that Analysis Services retains mining structure caches locally. That is, it writes out all the training gegevens to your local hard disk. If you do not want the gegevens cached, you can switch the default by setting the MiningStructureCacheMode property on the mining structure to ClearAfterProcessing. This will ruin the cache after models are processed, however, it will also disable drillthrough on the mining structure. For more information, see Drillthrough Queries (,Gegevens Mining),.
Also, if you clear the cache, you will not be able to use the holdout test set, if you defined one, and the definition of the test set partition will be lost. For more information about holdout test sets, see Training and Testing Gegevens Sets.
Processing Mining Models
You can process a mining monster separately from its associated mining structure, or you can process all models that are based on the structure, together with the structure.
However, te SQL Server Gegevens Instruments (SSDT) and SQL Server Management Studio, you cannot multiselect mining models to process with the structure. If you need to control which models are processed, you vereiste select them individually, or use XMLA or DMX to process models serially.
When Reprocessing is Required
You vereiste process the Analysis Services models that you define before you can commence to work with them. You vereiste also reprocess the mining models whenever you switch the mining prototype structure, update the training gegevens, switch an existing mining proefje, or add a fresh mining prototype to the structure.
Mining models are also processed te thesis screenplays:
Deployment of a project: Depending on the project settings and the current state of the project, the mining models ter the project are typically processed ter total when the project is deployed.
When you initiate deployment, processing starts automatically, unless there is a previously processed version on the Analysis Services server and there have bot no structural switches. You can deploy a project by selecting Deploy solution from the drop-down list or by pressing the F5 key. You can
For more information about how to set Analysis Services deployment properties that control how mining models are deployed, see Deployment of Gegevens Mining Solutions.
Moving a mining proefje: When you budge a mining proefje by using the Uitvoer instruction, only the definition of the specimen is exported, which includes the name of the mining structure that is expected to provide gegevens to the monster.
Reprocessing requirements for the following screenplays using the Uitvoer and Invoer guidelines:
The mining structure exists on the target example and the mining structure is te an unprocessed state.
Both the structure and monster voorwaarde be reprocessed.
The mining structure exists on the target example and the mining structure has bot processed. Only the mining proefje wasgoed exported.
The monster can be used without processing.
The mining structure definition wasgoed also exported by using the WITH DEENDENCIES keyword.
Both the structure and prototype voorwaarde be reprocessed.