Benchmarks should be Samples #56304

New Issue

Francesco Siddi · 2018-08-11T01:48:13+02:00

Francesco Siddi commented

2018-08-11 01:48:13 +02:00

The Open Data platform aims at hosting more than benchmark results. For this reason we should look into introducing a more generic data structure, called 'sample'.

Sample schema on My Data

field	type	description
id	int	the PK
aid	str	Alphanumeric representation of the PK
manage_token	str	Token issued to the data owner to update/remove the data from Open Data
raw_data	json	The sample as submitted by the client
date_created	datetime	Submission date
is_redacted	bool	(to discuss - optionally specify which fields to redact?)
weight	int	(to discuss)
user	int	FK to the User
serie	int	FK to the Serie - e.g. benchmark, blender_org downloads, telemetry, etc

A similar schema could be adopted by Open Data

field	type	description
id	int	the PK
aid	str	Alphanumeric representation of the PK
manage_token	str	Token issued to the data owner to update/remove the data from Open Data
data	json	The (redacted data) which should be indexed
date_created	datetime	Submission date
weight	int	(to discuss)
serie	str	The Serie - e.g. benchmark, blender_org downloads, telemetry, etc

AlphaID vs UUID

While it would be fantastic to be able to reference a sample with a 6-chars alphanumeric string, which would be build starting from the My Data sample id, we understand that the Open Data portal will get data input from various sources. This issue can be solved by providing a 'serie' name with the sample.

Francesco Siddi commented

2018-08-11 01:48:13 +02:00

Added subscriber: @fsiddi

Sem Mulder commented

2020-01-14 10:58:03 +01:00

Added subscriber: @SemMulder

Sem Mulder commented

2020-01-14 10:58:03 +01:00

more generic data structure

I think a better approach would be to create different specific (i.e. non-generic) data structures for different kind of samples. I believe this to be a better option because statically known data structures are easier to reason about than dynamic ones (especially if combined with type checking), resulting in more robust code.

> more generic data structure I think a better approach would be to create different specific (i.e. non-generic) data structures for different kind of samples. I believe this to be a better option because statically known data structures are easier to reason about than dynamic ones (especially if combined with type checking), resulting in more robust code.

Sign in to join this conversation.