SIERRA Hooks

SIERRA allows a number of different elements of its pipeline to be customized and extend by a Project. To override part of a stage of the SIERRA pipeline, you must create a pipeline/stageX directory structure in your project directory, where stageX is the stage you want to override part of.

Note

This is the most advanced feature of SIERRA, and strange things may happen or things might not work right if you don’t call the super() version of whatever functions you are overriding in your hook.

Warning

Don’t try to override parts of the pipeline which are not listed below. It will probably not work, and even if it does there is every chance that it will break stuff.

Multi-Stage Hooks

Base Batch Criteria

Suppose you want to extend one of the following core SIERRA classes to add additional attributes/methods you want to be accessible to all Batch Criteria in your project:

To do this, do the following:

Create variables/batch_criteria.py in the root directory for your project.
Override one or more of the classes above, and SIERRA will then select your version of said override classes when running. Exactly where SIERRA is looking/what module it uses when a given class is requested can be seen with --log-level=TRACE.

Warning

Don’t override or extend any of the interfaces! It will causes static analysis and/or runtime errors.

Stage 3 Hooks

Experimental Run Collation

In order to generate additional inter-experiment graphs, you have to also collate additional CSV files by:

Create pipeline/stage3/run_collator.py.

Override the sierra.core.pipeline.stage3.run_collator.ExperimentalRunCSVGatherer class:

import sierra.core.pipeline.stage3.run_collator as run_collator
import pathlib

class ExperimentalRunCSVGatherer(run_collator.ExperimentalRunCSVGatherer):
    def gather_csvs_from_run(self,
                             run_output_root: pathlib.Path) -> tp.Dict[tp.Tuple[str, str], pd.DataFrame]:
        ...

Stage 4 Hooks

Tiered YAML Config

Suppose you have some graphs which are common to multiple SIERRA projects, and you don’t want to have to duplicate the graph definitions in the .yaml files. You can put those definitions in a single location and then add them to the .yaml graph definitions that are unique to the Project as follows:

Create pipeline/stage4/yaml_config_loader.py.

Override the sierra.core.pipeline.stage4.yaml_config_loader.YAMLConfigLoader class:

import sierra.core.pipeline.stage4.yaml_config_loader as ycl
from sierra.core import types

class YAMLConfigLoader(ycl.YAMLConfigLoader):
    def __call__(self, cmdopts: types.Cmdopts) -> tp.Dict[str, types.YAMLDict]:
        ...

Intra-Experiment Graph Generation

You way want to extend the set of graphs which is generated for each experiment in the batch, based on what batch criteria is selected, or for some other reason. To do so:

Create pipeline/stage4/intra_exp_graph_generator.py.

Override the sierra.core.pipeline.stage4.inter_exp_graph_generator.InterExpGraphGenerator class:

import sierra.core.pipeline.stage4 as stage4

class IntraExpGraphGenerator(stage4.intra_exp_graph_generator.IntraExpGraphGenerator):
    def __call__(self, criteria: bc.IConcreteBatchCriteria) -> None:
        ...

Inter-Experiment Graph Generation

You way want to extend the set of graphs which is generated across each each experiment in the batch (e.g., to create graphs of summary performance measures). To do so:

Create pipeline/stage4/inter_exp_graph_generator.py.

Override the sierra.core.pipeline.stage4.inter_exp_graph_generator.InterExpGraphGenerator class:

import sierra.core.pipeline.stage4 as stage4
import sierra.core.batch_criteria as bc

class InterExpGraphGenerator(stage4.inter_exp_graph_generator.InterExpGraphGenerator):
    def __call__(self, criteria: bc.IConcreteBatchCriteria) -> None:
        ...