ispirt/depa-training

How ispirt/depa-training works

DEPA for Training is a reference implementation of confidential clean rooms for privacy-preserving analytics and model training. A scenario supplies data preparation and pipeline configuration; the core framework supports dataset loading, joining, and differential-privacy-aware training, with evaluation and model outputs. Contracts, policy, and encrypted-data handling provide important governance and data-protection capabilities. The interactive demo discovers scenarios and exposes deployment and status controls. Edges are limited to README statements and relationships visible in the excerpts or source index; unsampled runtime wiring is left unasserted.

Main entry points: pipeline_executor.py Pipeline Executor · xgb_train.py XGBoost Training · app.py Demo Backend · dl_train.py Deep Learning

Components 23

Interactive Demo

Demo UI gui-demo/static/js/app.js
Demo Backend gui-demo/app.py

Scenario Workflows

Scenario Definitions scenarios
Pipeline Config scenarios
Data Preprocessing scenarios
Model Setup scenarios
Scenario Data scenarios
Trained Models scenarios

Training Engine

Pipeline Executor src/train/pytrain/pipeline_executor.py
Dataset Construction src/train/pytrain/utilities/dataset_constructor.py
Dataset Joining src/train/pytrain/join.py
Task Base src/train/pytrain/task_base.py
Deep Learning src/train/pytrain/dl_train.py
XGBoost Training src/train/pytrain/xgb_train.py
Model Construction src/train/pytrain/utilities/model_constructor.py
Loss Construction src/train/pytrain/utilities/loss_constructor.py
Evaluation src/train/pytrain/utilities/eval_tools.py

Privacy And Governance

Encrypted Data Loader src/encfs/encfs.sh
Policy Rules src/policy/policy.rego
Scenario Contracts scenarios

External Services

Contract Service
Azure Services

Other components

Scenario Operator

Connections 26

Scenario Operator → Demo UI: operates README.md
Demo UI → Demo Backend: requests scenarios gui-demo/static/js/app.js
Demo Backend → Scenario Definitions: discovers gui-demo/app.py
Demo UI → Demo Backend: runs scripts gui-demo/static/js/app.js
Demo Backend → Azure Services: checks account gui-demo/app.py
Demo Backend → Pipeline Config: reads scenario gui-demo/app.py
Pipeline Config → Pipeline Executor: configures
Pipeline Executor → Dataset Construction: loads datasets
Pipeline Executor → Dataset Joining: dispatches join
Pipeline Executor → Deep Learning: dispatches training
Pipeline Executor → XGBoost Training: dispatches training
Dataset Construction → Scenario Data: reads data
Dataset Joining → Scenario Data: joins datasets
XGBoost Training → Task Base: inherits from src/train/pytrain/xgb_train.py
XGBoost Training → Dataset Construction: imports src/train/pytrain/xgb_train.py
XGBoost Training → Evaluation: imports src/train/pytrain/xgb_train.py
Deep Learning → Model Construction: uses
Deep Learning → Loss Construction: uses
Evaluation → Trained Models: evaluates outputs
Deep Learning → Trained Models: produces models
XGBoost Training → Trained Models: produces models
Encrypted Data Loader → Scenario Data: loads encrypted data
Scenario Contracts → Contract Service: works with README.md
Scenario Operator → Contract Service: signs contract
Pipeline Config → Policy Rules: sets policy
Azure Services → Encrypted Data Loader: hosts clean room

Drawn by GitDiagram on November 10, 2026 from the repository's file tree, README and source files. Read as Markdown ·