Skip to content

Latest commit

 

History

History
111 lines (74 loc) · 5.17 KB

DATABASE_CONTENT.md

File metadata and controls

111 lines (74 loc) · 5.17 KB
Raw

The Content of the Posterior Database

Github Actions is used to check that the database conforms to the following structure.

Posteriors posteriors

Posteriors are stored as json files as posteriors/[posterior_name].json.

All posteriors in the database contain (at minimum):

  • data_name: Data (name)
  • model_name: Model (name)
  • added_by: Name of the person that added the posterior
  • added_date: Date of the person that added the posterior

Also, the posterior can contain slots on

  • references: What references should be cited if the posterior is used. These are using BibTeX short names. You can find the actual references in references/references.bib.
  • reference_posterior_name: Reference posterior name if it exists/has been computed. Otherwise is null.
  • dimension: Dimensions of different parameter names. The sum is the total dimension of the posterior. Hence the dimension of the Dirichlet vector of length K is K - 1.
  • keywords: Keywords for the data (see keywords below)

Data data

All datasets are stored as zipped JSON objects in data/data, and information on the specific data is stored in data/info. The raw datasets and reproducible scripts can be found data/data-raw.

data/data

  1. All data are stored as zipped JSON files as data/data/[data_name].json.zip.
  2. Matrices are stored as a list of row vectors.

data/info

The data info file contains information on the data used and is stored as data/info/[data_name].info.json.

All data in the database contain (at minimum):

  • name: The dataset name
  • data_file: Path to data JSON in the database
  • title: The title for the dataset (used for printing)
  • added_by: Name of the person who added the dataset.
  • added_date: Date when the dataset was added.

Also, the posterior can contain slots on

  • references: What references should be cited if the dataset is used. These are using BibTeX short names. You can find the actual references in references/references.bib.
  • description: A short description of the data used.
  • urls: URLs to the data with more information.
  • keywords: Keywords for the data (see keywords below)

data/data-raw

We can find the raw dataset and conversion scripts from raw format to the usable JSON files in data-raw/[data_name].

The script should be executable (in R or Python) to reproduce the dataset at the given location and be called [data_name].r or [data_name].py.

Models models

models/info

The model info file contains information on the model and is stored as models/info/[model_name].info.json.

All models in the database contain (at minimum):

  • name: The model name
  • model_code: A named JSON object with one element per framework. For example, the "stan" points to a model representation as stan code.
  • title: The title for the model (used for printing)
  • added_by: Name of the person who added the dataset.
  • added_date: Date when the dataset was added.

Also, the models can contain slots on

  • references: What references should be cited if the model is used. These are using BibTeX short names. You can find the actual references in references/references.bib.
  • description: A short description of the model used.
  • urls: urls to the model to read more.
  • keywords: Keywords for the model (see keywords below)

models/stan

Contain models as stan files stored as models/stan/[model_name].stan.

Posterior reference draws reference_draws

Contain reference draws information are stored as JSON in reference_posteriors/draws/info/[posterior_name].info.json. The actual draws are stored in reference_posteriors/draws/[posterior_name].json.zip.

See REFERENCE_POSTERIOR_DEFINITION.md for details on what is considered to be reference posterior draws in the posteriordb.

All reference draws in the database contains (at minimum) the following information:

  • name: The reference_posterior name. "comments", "added_by", "added_date", "versions"
  • inference:
    • method: How were the draws created (e.g. "stan_sampling" or "analytical")
    • method_arguments: arguments used to create reference_posterior, such as chains, iter, warmup, algorithm, seed etc., for full reproducibility.
  • diagnostics: Stored diagnostic values of the reference posterior
  • checks_made: Checks made on the reference posterior to asses it complies with the reference posterior definition. See REFERENCE_POSTERIOR_DEFINITION.md.
  • comments: Additional comments on the
  • added_by: Added by (name)
  • added_date: Date the file was added
  • versions:

References references

It contains a bib-tex file with all references in the databases for quickly assembling bibliographies of a specific set of posteriors, models or data.

Schemas schemas

JSON-schemas for all elements in the database. Currently not included.

Keyword

In the database, multiple keywords are used (and can be added by users). Some relevant keywords of interest:

  • stan_benchmarks: Models included in the Stan Benchmarks