Euclid DSPS SHINE

Synthetic Diffsky/FENIKS closure and photometric inference workflows.

Euclid DSPS SHINE wraps native dsps in a small, testable interface for synthetic closure catalogue generation, catalog preparation, forward photometry, MAP/posterior inference, and reproducible diagnostics.

Where to Start

Goal

Read

Main command or config

Install the package

Installation

python -m pip install -e .

Run the production workflow

Production Runbook, Diffsky FENIKS DSPS Closure Dataset

configs/diffsky_synthetic_feniks_260617_50k.yaml

Validate closure data

Production Runbook, Diffsky FENIKS DSPS Closure Dataset

diffsky-validate-dsps-closure

Train supervised priors

Supervised Diffsky Prior Learning

configs/prior_diffsky_synthetic_feniks_full_realnvp.yaml

Run amortized inference

Amortized Inference

configs/amortized_diffsky_synthetic_feniks_full_gpu.yaml

Use HLTDS/FS2 debug references

Data And Assets, HLTDS Reference Data

configs/diffsky_dataset_hltds_04_14.yaml

Workflow Map

Stage

What it owns

Primary outputs

Data contract

FENIKS closure truth, prepared parquet rows, units, truth semantics, and error-model provenance. HLTDS/FS2 are debug references.

Manifests, schema files, inventory reports, integrity reports.

Forward model

The DSPS boundary, filter handling, SSP assets, calibration terms, and photometric likelihood setup.

Predicted fluxes, residuals, closure diagnostics.

Inference

MAP fits, posterior sampling, supervised truth priors, and amortized encoder/prior experiments.

CSV/JSON summaries, checkpoints, posterior samples.

Assessment

Population realism, redshift calibration, forward closure, and truth-vs-posterior diagnostics.

Markdown reports, metrics tables, plots.

Project Boundaries

  • euclid_dsps.model is the only module that calls native DSPS.

  • euclid_dsps.io owns parquet row contracts and photometry unit conversions.

  • euclid_dsps.fit and euclid_dsps.mcmc own point estimates and posterior sampling.

  • euclid_dsps.workflows composes command-line workflows.

  • euclid_dsps.reporting writes CSV, JSON, Markdown, and plot artifacts.

Scientific Guardrails

A good photometric fit is not enough to claim physical recovery. Use same-parameter forward closure, supervised truth-prior diagnostics, posterior calibration, and derived-quantity comparisons before interpreting recovered parameters physically.

Getting Started

Science Workflows

Reference