Samplers output standardization

[lucasgautheron]

Is your feature request related to a problem? Please describe.

Goals:

• We should be able to traceback each sample to the parameters that help generate it, and the version of the package that was used.
• The samples themselves should be standardized
• Ideally, it should be possible to deliver the samples and their parameters with the dataset...

Describe the solution you'd like

For the storage:

We could use a nested tree just like w/ annotations :

samples / 
   vetted/
       segments.csv
   high-volubility/
       lena/
           segments.csv
           parameters.csv
       vtc/
           segments.csv
           parameters.csv
    energy1
        segments.csv
        parameters.csv

[alecristia]

this assumes the name is interpretable, right? wouldn't it make more sense to --yes-- give it a name, but either append parameters to the name and/or add a random or sequential number (eg random2 the second time I generate a random sample), and/or include the yaml of the parameters in the same folder?

[lucasgautheron]

Well, we could save the runs under samples/<name>/parameters_%Y-%m-%d_%H_%M_%S.csv and samples/<name>/segments_%Y-%m-%d_%H_%M_%S.csv, that would do the trick right ?

[alecristia]

yes, naming files in that way ensures that there is no over-writing, and connects parameters to data. It also allows for errors in our judgment in terms of the parameters we may have now versus in the future, as well as "harvesting" of data & parameters by using csvs for both (an alternative would have been yaml for the parameters, but they sound equivalent to me in this case).