Databricks Bundle guide index

For all Databricks, MLflow, Spark and preprocessing guides with short page descriptions, open the integrations documentation index.

Start with the end-to-end project guide to create a project, train a model and publish predictions. Before you generate one, choose a project shape: layout, prediction path and what runs on Spark. Use the topic pages below for a specific change. Generated project Markdown contains short instructions and links to these references.

The project pins its Core version in deployment/requirements.txt. These public pages describe the current template; use the upgrade guide when the installed release differs. Training fits in one bounded Python process; distributed inference scores an admitted saved model on Spark workers.

Configuration and operations

Task	Detailed guide
Choose a layout and setup mode	Open guide
Files you edit	Open guide
What a configuration change requires	Open guide
Training and inference boundaries	Open guide
Distributed inference settings	Open guide
Monitoring investigation and training lineage	Open guide
Guided setup and offline preview	Open guide
Custom preprocessing recipes	Open guide
Writing your own step	Open guide
Write your own pre-split filter	Open guide
Preprocessing diagnostics	Open guide
Source layout and existing projects	Open guide
Local input limits	Open guide
Record identity and result availability	Open guide
Choose the evaluation split and result availability	Open guide
Search and cross-validation	Open guide
Optional classification decision thresholds	Open guide
Ensemble recipes	Open guide
Explicit training sampling	Open guide
Source windows are independent of splitting and scheduling	Open guide
Source date formats and timezones	Open guide
Configuration validation and migration	Open guide
Upgrade integration code	Open guide
Where the workflow lives	Open guide
Understand files in an experiment's Artifacts tab	Artifact reference
Independent scoring and promotion policies	Open guide
Operator actions through Run now	Open guide
Approve an existing candidate without training	Open guide
Reject a candidate or roll back a promotion	Open guide
Recent challenger history	Open guide
Recover an interrupted alias change	Open guide
Train, approve and score a multi-target model set	Open guide
Set the model_set block	Open guide
Choose a model-set output mode	Open guide
Compose model-set outputs with declarative rules	Open guide
Model-set quality policy and replacement rule	Open guide
Score a model set	Open guide
Find and read SHAP explanations	Open guide
What is created	Open guide
Additional quality gates	Open guide
Reading notebook results and failures	Open guide
First run	Open guide
Runtime package and compute configuration	Open guide
Explicit retry policy	Open guide
Job resource files	Open guide
Remove a deployed project	Open guide
Fit, inspect and score from Python	Open guide
Publish one calendar period of predictions to Delta	Open guide

For basic or nested cross-validation, see CV and total work.

Monitoring and serving

Detailed setup asks prediction_delivery: batch creates the score job, endpoint creates only the serving job, and both creates both. Quick setup defaults to batch. All choices keep monitoring. Logged endpoints created by the serving job enroll automatically after verification; observations still require requests and a monitoring run. See delivery choices and setup.

Task	Detailed guide
Understand how monitoring runs and what to read first	Monitoring overview
Create or upgrade the shared monitoring store	Store setup
Enroll model identities and prepare reference data	Enrollment
Match later outcomes to predictions	Delayed labels
Investigate input distribution and data quality	Drift
Detect observed model performance loss	Performance
Configure guarded automatic retraining	Retraining
Read and filter saved observations	Dashboard
Create an endpoint and query it	Serving
Publish and look up online feature values	Online features
Observe requests, latency and incomplete logs	Serving monitoring
Increase challenger traffic each day	Daily rollout

Preprocessing step catalog

The preprocessing catalog explains recipes and phases and indexes every registered step. Each family has its own page:

Family	Page
Missing values, cleaning and types	Open page
Scaling and numeric transformations	Open page
Bounds and outliers	Open page
Categorical encoding	Open page
Binning	Open page
Feature generation and selection	Open page
Time and geography	Open page
Text and embeddings	Open page
Split, sampling and diagnostic Core nodes	Open page
Custom factory contracts and ready functions	Open page

Reference pages

For exact fields and selectable operations, use the training configuration, model catalog, preprocessing catalog, scoring reference and feature table guide. The glossary defines shared terms; troubleshooting maps validation messages to fixes.