Author: Drew Shea, Created: 2023-09-10

Overview

The purpose of this document is to provide all of the major SensibleAI Studio definitions and key terms.

Many of the Definitions and Key Terms reference one another. Referenced Definitions and Key Terms will show up in italics.

Definitions & Key Terms

2flatpipe

A-D

Name	Definition	Additional Information	Example
Artifact	An Artifact is an object that can be generated by a Routine Method Run that holds data, metadata, and other summary level information that can be used after its generated for downstream business processes. An Artifact is comprised of: Artifact Metadata Artifact Data (optional) Artifact Statistics (optional) Artifact Previews	Artifacts are stored in a file-system backend (like the MetaFileSystem) A Run may generate zero to N number of Artifacts. It is entirely dependent on how the Routine Developer defined their Routine . Artifacts may be accessed through the SensibleAI Studio UI or programmatically through OneStream Business Rules	anomaly_instances A set of predictions generated by an anomaly detection algorithm that labels data as anomalous or not. decision_tree_visual An image of how a decision tree was broken down within a classifier model and it decided to split the tree based on feature values.
Artifact - Artifact Definition	An Artifact Definition is what defines what a generated Artifact is anticipated to look like after it’s generated by running a Routine Method .	An Artifact Definition is broadly comprised of attribute like Artifact Key , Artifact Qualified Key , Artifact File Annotation s , Artifact In-Memory Accessible , etc.
Artifact - Artifact Key (Annotation)	The name identifier of an Artifact within a particular Routine Method .	This is not guaranteed to be unique when dealing with Nested Artifacts . See Artifact Qualified Key example for details. The Artifact Key Annotation will map to a directory name found the Routine Storage Structure within the artifacts_ directory.
Artifact - Artifact Qualified Key (Annotation)	The unique identifier of an Artifact within a particular Routine Method.	The Artifact Qualified Key is defined by the Routine Developer when writing the developing the Routine . The Artifact Qualified Key will look different from a Artifact Key when Nested Artifacts are involved. When you have Nested Artifacts , you will see the Artifact Qualified Key be comprised of both the outer Artifact Key and the inner Artifact Key separated by a “.” (ex: outer_key.inner_key )	my_dataframe : An example mapping to a dataframe my_list.0.dataframe : A nested Artifact where you have an Artifact my_list then an Artifact 0 for the 1st index, then you have an Artifact called dataframe . You would say that dataframe is the Artifact Key and my_list.0.dataframe is the Artifact Qualified Key
Artifact - Artifact Data	The actual data generated by an Artifact .	The type of data that is generated is defined by the Routine Method . Some of the common Artifact File Types include, .parquet , .html , .png , .webref , .json , and more. This file data can be read back on the .NET within Xperiflow Business Rule helper methods
Artifact - Artifact File Annotation	A file path and a description that represents one or more of the anticipated Artifact Data file types found within the Artifact.	@ = Represents an Artifact directory <int> = A directory or file variable representing an integer. <str> = A directory or file variable representing a string.	Ex1: An Interactive HTML Artifact Annotation: artifacts_/@report_content/data_/html_content.html Description: Ex2: A Tabular Artifact (as partitioned parquet files) Annotation: artifacts_/@anomaly_instances/data_/data_<int>.parquet Description: A partitioned set of parquet files where each file will have no more than 1000000 rows.
Artifact - Artifact Metadata	The metadata associated with a generated Artifact .	This is typically represented as a artifact_metadata.json file that is accessible within the metadata_ directory of the Routine Directory Structure . This holds information such as the Artifact creation time, Artifact description, Artifact title, Artifact File Annotations , and more.
Artifact - Artifact Preview	An Artifact Preview is a reviewable representation of an Artifact to better inform a Power User of as to what they will find in the Artifact data itself.	An Artifact can have 0 to N number of Artifact Previews generated for it. Generation of Artifact Previews may be turned off by the Power User
Artifact - Artifact Statistic	A statistic that is generated off of raw Artifact Data that can provide more insight into the underlying Artifact Data.	Artifact Statistics can be seen within SensibleAI Studio after the Artifact has been generated. Generation of Artifact Statistics may be turned off by the Power User
Artifact - In-Memory Json Accessible	A special attribute found on an Artifact Definition that denotes whether the Artifact can be accessed as a Json object when the Routine Method is run with a Routine Method Execution Type of “in-memory”.	Not all Artifact Data is accessible as a Json object.
Artifact - Dynamic Artifact	A special type of Artifact that is generated dynamically when the Routine Method runs. The user won’t know it exists until after the Routine Run is complete.	Within the Routine Documentation in the Signature section for any method, the user can see a field titled ,“Outputs Dynamic Artifacts”, that lets the user know if this Routine may return any Dynamic Artifact. You will not see a Dynamic Artifact show up in the Routine Documentation under the Artifacts Section (by definition, you don’t know what the definition of Dynamic Artifact is until you actually Run the Routine Method , therefore we can’t embed it “up-front” in the documentation. We can only tell you if the Routine Method may return other Dynamic Artifacts ). Typically, Routine Developers try to avoid using this capability in the Routines they develop to provide deterministic results coming out of Routine Method s.
(Sensible)AI Studio	The OneStream Solution for managing and interacting with Routines programmatically or through a user interface.	Acronym (STU) Make sure to not confuse this with the Sensible ML Feature Library which is a prebuilt data store of external data connections to pull information such as macro-economic information, etc.
Solution Exchange - Routine Store	The Routine Store is a centralized service to house OneStream developed and community developed Routines made by Routine Developers	The Routine Store allows: Routine Developers to upload their Routines Routine Developers to get their Routines approved for commercial use AI Services Users to find and import community developed Routines into their library so that they can use them This feature will not be available until a future release.

E-K

Name	Definition	Additional Information	Example

L-O

Name	Definition	Additional Information	Example
Memory Capacity - Run Memory Capacity	The max amount of memory in gigabytes that a given Routine Method is allowed to consume at any one time.	If this memory limit is surpassed, the Run will be suspended and terminated. An error message will be provided, denoting that the memory capacity was exceeded.
Memory Capacity - Total Memory Capacity	The total memory capacity in gigabytes allowed for the for the AI Services environment that can be used at any one time.	If this memory limit is surpassed, the latest starting Run will be suspended and terminated. An error message will be provided, denoting that the memory capacity was exceeded.

P-S

Name	Definition	Additional Information	Example
Stateful Routine	A type of Routine that manages state and is comprised of multiple methods that can be executed after this routine type has been created. A Stateful Routine is defined with the python method name of init .	A Stateful Routine resembles a class that you would see in any other programming language. It has a constructor to maintain stateful member variables, has one to N methods on it, and those methods may change the state of the Routine Instance variables as those methods are executed	XGBoost Classifier Routine - This is a Routine Method because it is expects the user of the Routine to 1) Construct it 2) Call Fit (a method) which will train the XGBoost Classifier and 3) Call Predict (a method) to leverage the internal trained model to make predictions
Stateless Routine	A type of Routine that does not manage internal state and does not have a constructor method to instantiate the Routine Instance .	A Stateless Routine resembles a “static class” that you would see in any other programming language. The only difference is that you still must “create” the routine instance before you can call the Routine Methods on it.
Routine Instance	An instantiated Routine .	If the Routine is a Stateless Routine , the Routine is considered “instantiated” once the Routine is “created”. If the Routine is a Stateful Routine , the Routine is considered “instantiated” once the Routine is “created” and the Routine Constructor is Run. After creating an instance of a Routine , I can now call the various methods that are defined on that routine. Just like any programming language, I can create many instances of a Routine A Routine Constructor may specify 0 to N parameters.	I can create many instances of an CNN Image Classifier Routine. Maybe I will name and create one instance of this CNN Image Classifier Routine for classifying whether an image is a hamburger or a hotdog. I can also create another instance that can be used for classifying whether an image is a Tree or a Dog.
Routine Method	A method defined as part of a Routine. A method will take in input parameters, execute some business logic, then output (if any) results as Artifacts and potentially alter the state of the Routine Instance.	Like any standard programming language, you must instantiation the Routine prior to being able to run the Routine Method.
Routine Method - Constructor	A special type of Routine Method that must be invoked on a Stateful Routine prior to calling any of the Routine Methods available on the Routine .
Routine Method - Execution Type	The way in which a Routine Method Run is executed. The Execution Type is set to “background” by default.	There are two main Execution Types - “background” and “in-memory”. background - Executes the Routine Method Run as a background job on some system defined processing engine. in-memory - Executes the Routine Method Run “in-process” (or at the web service tier) to ensure fast execution. Note that only certain Routine Methods allow this type of execution (defined by the Routine Developer ) This should only be used when you need the Routine to return data results behind a UI interaction (4 seconds or less is a good rule of thumb) When executing in this way, any Artifact that is In-Memory Json Accessible will allow the Power User to access the Json variation of the Artifact Data as part of the In-Memory result.
Routine Method - Signature	The Routine Method’s name, return type ( Artifacts ), and parameters—essentially how the method should be called and what it returns.
Routine	A modular software component comprised of one or more functions that accept inputs and produce outputs, such as data results or interactive web applications.	The python script format and interface that a Routine Developer must implement in order to have their data science functionality accepted into the AI Studio. BaseRoutine , ParameterBaseModel , and ArtifactBaseModel act as the backbone of the interface to develop your script against There is a test harness to allow developers to validate that their script conforms to the interface.
Routine Type	A type of Routine. There are two types of routines that can be defined by Routine Developer and executed by users. The Stateful Routine and the Stateless Routine.	The two Routine Types are modeled off the standard programming concepts of Classes and Static Classes. Classes have instance variables, constructors, and methods that take in input, change state, and output results. On the other hand, Static Classes just define 1 to N methods and simply take in inputs and output results.
Routine Test Harness	A Python PyTest framework that can validate and provide feedback to Routine Developer on if their Routine conforms to the standards and the interface that the AI Studio expects.	This feature will not be available until a future release.
Routine Run (Routine Method Run)	A Run denoted as any invocation of a Routine Method or Routine Constructor .	A Routine Run will generate a storage directory structure within the MetaFileSystem A Routine Run’s progress can be monitored within SensibleAI Studio or programmatically. A Routine Run can be run in “standard mode” or “in-memory” mode. “in-memory” mode allows the Routine Run to be executed on the AI Engine web service to enable faster execution. In this mode, Artifacts that are JSON-serializable are accessible via a JSON representation of the Artifact Data .
Routine Storage Structure	A file-based directory structure that outlines how inputs, metadata, and outputs are stored to represent a Routine Instance , and Routine Runs .	Whenever a user creates a Routine Instance or Routine Run is created, a new underlying directory structure is created in the MetaFileSystem to represent the data of the Routine Instance or Routine Run. For more information on what this directory structure looks like, please see: Routine Storage Structure

T-Z

Name	Definition	Additional Information	Example
Workflow	An instance of a Workflow Definition	Another analogy to use here is think of a Workflow Instance as an instance of an object.
Workflow Definition	A Workflow Definition defines a series of workflow steps, transitions, triggers, validations, that can be taken to execute some use case. A Workflow Definition can have many Workflows (instances)	A Workflow Definition is expected to accompany a Routine to define how a user will be guided through the process on the front-end of filling out the necessary parameters to fire of a Routine.
User - Routine Developer	A Routine Developer develops and contributes new Routines to the AI Studio that other the Power User can build with.	The Routine Developer Builds and Deploys AI Studio Routines in Python. The typical background of a Routine Developer is a Python developer, data scientist, or computer scientist The Routine Developer writes code in Python and implements the BaseRoutine interface. Over time, third party developers will be able to develop Routines. For the short term, only OneStream licensed developers will be allowed to act as Routine Developers
User - Power User	A Power User is leveraging the outputs (typically relational tables, docs, reports, etc.) of AI Studio Routines to power other processes in OneStream.	The Power User integrates AI Studio Routines into business processes in OneStream by leveraging the AI Studio APIs and core OneStream business rule APIs. The typical background of a Power User is OneStream Solution developers, solution consultants, Sensible ML consultants.
User - End User	The typical OneStream End User . It is expected that the End User will not spend any time inside of the AI Studio. Rather, they will rely on dashboards / reports built by the Power User	The End User simply just leverages the supercharged business processes inside of OneStream that was built for them by the Routine Developers and the Power Users.