Conclusion and next steps（结论与后续步骤）

In this tutorial, we created a supervised machine learning project in Foundry, in which we:

• Created a project for iterative model experimentation and development,
• Performed initial feature preparation and pipelining,
• Trained a production-ready model,
• Deployed our model to a live-hosted endpoint and a batch pipeline that automatically updates.

Foundry automatically tracks the lineage of all resources you produce in the platform. At the end of this tutorial, you will have a pipeline resembling the below screenshots.

Action: Navigate to your house_price_predictions dataset, select Explore pipelines > Explore data lineage.

Next steps

The next step is to convert this example workflow to a real workflow at your organization.

This typically includes:

1. Integrate data from a range of data sources into Foundry to create a features_and_labels dataset that can be used for training and testing different models.
2. Try different model architectures, parameters, and features to get the best performance for your model.
3. Integrate your model's predictions into the Foundry Ontology for use in operational applications either through a batch deployment, live deployment, or Python transforms.
4. Create pre-release checks in your modeling objectives to ensure models are approved before release.
5. Create "writeback" actions to capture user actions as a new dataset and use this data for continuous re-training and improvement of your model.
6. Create a model inference history to improve and iterate on your model for more accurate performance and usage.

---

中文翻译

---

结论与后续步骤

在本教程中，我们在 Foundry 中创建了一个监督式机器学习项目，具体完成了以下操作：

• 创建了一个用于迭代模型实验与开发的项目，
• 执行了初始特征准备与流水线构建，
• 训练了一个可用于生产环境的模型，
• 将模型部署到实时托管端点（live-hosted endpoint）以及自动更新的批量流水线（batch pipeline）中。

Foundry 会自动追踪您在平台中生成的所有资源的血缘关系（lineage）。完成本教程后，您将获得一个类似于下方截图的流水线。

操作： 导航至您的 house_price_predictions 数据集，选择 探索流水线 > 探索数据血缘。

后续步骤

下一步是将此示例工作流转化为您所在组织的实际工作流。

通常包括以下内容：

1. 将来自多种数据源的数据集成到 Foundry 中，创建一个可用于训练和测试不同模型的 features_and_labels 数据集。
2. 尝试不同的模型架构、参数和特征，以获得模型的最佳性能。
3. 通过批量部署（batch deployment）、实时部署（live deployment）或 Python 转换（Python transforms），将模型的预测结果集成到 Foundry 本体（Ontology）中，用于运营应用。
4. 在建模目标中创建预发布检查，确保模型在发布前获得批准。
5. 创建"回写"操作，将用户操作捕获为新数据集，并利用这些数据持续重新训练和改进模型。
6. 创建模型推理历史记录，以迭代优化模型，实现更准确的性能和使用效果。