Model connectivity and development（模型连接与开发）¶
In the Palantir platform, models are artifacts that encapsulate any machine learning logic. Models can be leveraged across workflows in data pipelines, Ontology, and application layers to enable a wide variety of different use cases.
Palantir takes an expansive approach to enabling model integration. Models can either be developed within the platform on integrated data, or externally and imported as libraries, artifacts, containers, or as APIs.
Once integrated, models can be used with platform tools for inference, deployment, governance, ML Ops, and operationalization. Every step in the model creation and consumption process is subject to platform guarantees around lineage, security, versioning, reproducibility, and auditing.
Sources¶
Models of any form can be developed in-platform (for example, via open-source tools such as scikit-learn, TensorFlow, and OR-tools, or via custom libraries), imported from an external environment (such as notebooks, third-party data science products, or container registries), or configured as externally hosted APIs.
Learn more about model development within the platform and integration of external models.
Modeling Objectives¶
Modeling Objectives serve as the "mission control" for streamlining model management, evaluation, review, release, and deployment for a defined problem. Beyond a convenient user interface, Modeling Objectives provide a governance and permissions layer, an automation layer (e.g. uniform evaluation of model candidates), and a CI/CD layer for continuous and downtime-free deployment for models.
Objectives enable the complete model lifecycle for any modeling problem, including those not traditionally addressed by ML Ops tools, such as simulation and optimization.
Using Objectives, organizations can go beyond simply creating a data pipeline to safely deploy models in an operational capacity for users and systems that make decisions. Once operationalized, Foundry enables ML feedback loops from production data, outcomes, applications, and user actions. These feedback loops provide modeling teams with a powerful data asset for monitoring, understanding, and improving production performance, as well as adapting to new circumstances.
Learn more about Modeling Objectives.
Models in the Ontology¶
The ontology is the operational layer of the Palantir platform. The Ontology connects digital assets to their real-world counterparts to enable all types of different use cases.
Once a model has been integrated into the platform, the model can be deployed using Modeling Objectives and registered for use within the ontology layer. This allows operational interactive workflows to be backed by high-trust models to power, quick, critical decision-making.
Learn more about the Ontology and how to integrate models with it.
中文翻译¶
模型连接与开发¶
在 Palantir 平台中，模型（models）是封装了任意机器学习逻辑的制品。模型可被应用于数据管道、本体（Ontology）和应用层的工作流中，从而支持多种不同的使用场景。
Palantir 采用开放式的模型集成策略。模型既可以在平台内基于集成数据进行开发，也可以从外部导入为库、制品、容器或 API。
集成后，模型可与平台工具配合使用，用于推理、部署、治理、ML Ops 及运营化。模型创建与消费过程中的每一步都需遵循平台在血缘、安全、版本控制、可复现性及审计方面的保障要求。
来源¶
任何形式的模型都可以在平台内开发（例如通过 scikit-learn、TensorFlow、OR-tools 等开源工具，或自定义库），从外部环境导入（如笔记本、第三方数据科学产品或容器仓库），或配置为外部托管的 API。
建模目标¶
建模目标（Modeling Objectives）作为"任务控制中心"，用于简化针对特定问题的模型管理、评估、审核、发布和部署。除了便捷的用户界面外，建模目标还提供治理与权限层、自动化层（例如对候选模型进行统一评估）以及 CI/CD 层，支持模型的持续无中断部署。
建模目标能够覆盖任何建模问题的完整模型生命周期（model lifecycle），包括传统 ML Ops 工具未涉及的领域，如仿真和优化。
通过建模目标，组织不仅能创建数据管道，还能安全地将模型以运营能力部署给做出决策的用户和系统。一旦实现运营化，Foundry 可从生产数据、结果、应用和用户行为中获取 ML 反馈循环。这些反馈循环为建模团队提供了强大的数据资产，用于监控、理解和改进生产性能，并适应新环境。
了解更多关于建模目标的信息。
本体中的模型¶
本体是 Palantir 平台的运营层。本体将数字资产与其现实世界对应物相连接，以支持各种类型的使用场景。
模型集成到平台后，可通过建模目标进行部署，并注册用于本体层。这使得运营交互工作流能够由高信任度模型驱动，从而支持快速、关键性的决策。