Node Activation Mode
Node activation options
Batch processing of workflows
You can set the following execution options for each node when running a batch process:
- Don't execute the node
- Execute the node without retraining the model.
- Execute the node and retrain the model.
To select one of these options for a node's behavior during batch processing, consider the following:
-
The context in which you launch the workflow execution. In batch processing, the context includes the following launch parameters:
- Launching the package with or without the /Teach parameter.
- Launching the package with the /Node parameter.
-
The node activation configuration of the Supernode or Loop that contains the node (if applicable).
- The activation mode set for the node itself.
Note: For more detailed information on batch processing and the /Teach and /Node parameters, see batch processing of workflows.
Running nodes interactively
When editing (debugging) a workflow, you can run an individual node in one of two ways:
- Run the node without retraining the model.
- Run the node and retrain the model.
A node's behavior during an interactive run is determined by the following settings:
-
The execution command: The command that initiates the execution of the node itself or the Supernode (Loop) that includes it. The available options are:
- Retrain node
- Run node
-
The activation of a parent node: If the node is inside a Supernode (Loop), its behavior is determined by the activation mode of that parent mode.
-
The node's activation mode: If the node is not inside a Supernode (Loop), its behavior is determined by its own activation mode setting.
Configuring node activation mode
Valid modes of batch processing
This group of settings only applies to node activation in the batch mode.
- Run the node and all its subnodes: The checkbox defines whether the node will be activated in case of no-retraining batch workflow launch (without the /Teach parameter). Disable this mode to skip the node and all its subnodes during batch processing.
- Retrain the node and all its subnodes: The checkbox defines whether the node will be activated in case of batch launch with retraining (with the /Teach parameter set). Disable this checkbox, if you don't need the node and its subnodes to be retrained.
By default, both options are selected for new nodes.
Specified mode of node activation
This setting determines if a node runs with or without retraining its model. It applies to both interactive and batch workflow executions.
Select one of the following options:
- Always run: The node is always executed without retraining the model. This setting overrides all other contexts such as the Retrain node command or the /Teach parameter.
- Always retrain: The node always runs and retrains the model. This setting also overrides all other contexts.
-
Determined by the context of current processing: This setting is used by default. The node’s behavior (i.e., Always run or Always retrain) is determined by the execution context. The system checks the following factors in order of precedence:
-
Parent node settings. If the node is inside a supernode or loop, it inherits the execution mode from that parent node.
Note: If supernodes or loops are nested, the node inherits its setting from the outermost supernode (loop) in the hierarchy.
-
The execution command.
Interactive mode: Behavior is determined by the command you use (Retrain node or Run node).
Batch mode: Behavior is determined by the setting of the /Teach parameter.
Batch processing execution logic
During batch processing, the system determines whether to run or retrain a node by checking its configuration. The logic follows the order of operations shown in the diagram:
Execution exceptions
A node might not execute during a batch run, even if its settings indicate it should, in the following situations:
-
Another node is specified in the launch parameters. If you use the /Node) parameter, only the specified node and any preceding nodes that generate its input data are executed. All other nodes are skipped.
-
A preceding node fails to run. A node cannot be executed if the input data it depends on is not generated.
Read on: Node Training (Retraining)