Acumos "model" creation

AI4EU computing assets are called "models".

This task tracks the work related to the creation of a tool that will allow the creation of docker images out of BEAT algorithms.

Note that not all BEAT algorithms are eligible for exportation to an Acumos compatible Docker image.

Based on the current state of the platform as of 18.01.2021, here is a list of the constraints that must be taken into account:

The current Acumos orchestrator follows a single input single output sequential execution implementation. This means that each data frame coming from the entry data broker will have to pass through all the blocks from a solution up to its final block before processing the next data frame.

This means that there's no training currently possible within the Acumos platform.

Therefore in terms of BEAT algorithms, this means that only algorithms matching the following criteria can be used:

• Only sequential algorithms can be used as autonomous ones uses pull rather than a push to get data which is not available currently in Acumos
• Only algorithms which write as many outputs as they gets inputs as otherwise the orchestrator won't be able to carry on a full run

These criteria may change in the future based on the features implemented on Acumos.

It has also been decided that in the first iteration, the tool that will export the BEAT algorithm will only be used locally. Therefore no online platform data extraction will happen.

This means that a user will first have to run a full experiment using the algorithm(s) destinated to be exported before the tool can be used.