Open access & transparency
of data and models

Jan Philipp Dietrich
dietrixyzch@pik-xyzpotsdam.de

Open Source

Open Access of models & tools

What means Open Source?

opensource.org/osd

6. No Discrimination Against Fields of Endeavor

1. Free Redistribution

2. Source Code

3. Derived Works

...

5. No Discrimination Against Persons or Groups

...

Copyleft

  • Code derivatives must be published under the same license
  • What is open stays open
  • Examples:



  • useful to..
    • ..establish software as common good
    • ..prevent the use in proprietary software

Non-Copyleft

  • Code derivatives can have a different license
  • What is open can be used in any way
  • Examples:

 

 

 

  • useful to ..
    • ..avoid license conflicts
    • ..broaden the range of applications

Two types of Open Source

  • technically yes, otherwise no
  • in addition code should be..
    • ..owned by you
    • ..properly accessible
    • ..documented
    • ..citable
  • license of code contributions should be clarified


Applying a license and thats it?

  • all code owner need to agree to the license
  • ownership has to be clarified
  • code owner are depending on the contract..
    • ..code developer
    • ..employer (institute, university, company...)

Code ownership

Source: Mimi and Eunice  | CC-BY-SA Nina Paley

Contributors License Agreement (CLA)

  • clarifies inbound license
  • useful to bundle code ownership for..
    • .. change of license (e.g. to solve compatibility problems)
    • .. legal representation 
  • in many cases not required

Code ownership - inbound license

GitHub default: inbound = outbound

Outbound license
(code sharing)

Inbound license
(code contributions)

\neq
  • version management strongly recommended
    • development history
    • option to revert changes
  • publicly accessible code repository (e.g. GitHub project)
    • simplifies collaboration
    • provides some feedback about usage
  • decentralized repo (git) better for sharing than centralized (svn)




     
  • recommendation: git in combination with GitHub

Code accessibility

Repository

User

Documentation

powered by goxygen » github.com/pik-piam/goxygen

  • documentation important for code transparency
  • in-code documentation recommended (e.g. doxygen, roxygen, goxygen)
*' @equations

q13_cost_tc(i2) ..
  v13_cost_tc(i2) =e= sum(ct, i13_land(i2) * i13_tc_factor(ct,i2)
                     * vm_tau(i2)**i13_tc_exponent(ct,i2)
                     * (1+pm_interest(i2))**15);

*' Relative technological change costs `v13_cost_tc` are calculated as a
*' heuristically derived power function of the land use intensity `vm_tau` 
*' for the investment-yield-ratio (see figure below) multiplied by the 
*' initial, regional crop areas in 1995 `pm_land_start` and shifted 15 
*' years into the future using the region specific interest rate 
*' `pm_interest`.
*'
*' ![Investment-yield ratio in relation to $\tau$-factor
*' [@dietrich_forecasting_2014]](tcc_regression.png){ width=60% }
*'
*' The shifting is performed because investments into technological change
*' require on average 15 years of research before a yield increase is 
*' achieved, but the model has to see costs and benefits concurrently in 
*' order to take the right investment decisions (see also 
*' @dietrich_forecasting_2014). Investment costs scale with crop area as a 
*' wider areal coverage means typically also higher variety in biophysical
*' conditions and therefore more research required for the same overall 
*' intensity boost.

q13_tech_cost_annuity(i2) ..
 v13_tech_cost_annuity(i2) =e= (vm_tau(i2)/pc13_tau(i2)-1) * v13_cost_tc(i2)
                               * pm_interest(i2)/(1+pm_interest(i2));

*' In order to get the full investments required for the desired 
*' intensification the relative technological change costs are multiplied 
*' with the given intensification rate. These full costs are then 
*' distributed over an infinite time horizon by multiplication with the 
*' interest rate `pm_interest(i)` (annuity with infinite time horizon).

q13_tech_cost(i2) ..
  vm_tech_cost(i2) =e= v13_tech_cost_annuity(i2) + pc13_tech_cost_past(i2);

*' Additionally, the technological change costs coming from past investment
*' decisions are added to the technological change costs of the current 
*' period.

Code citation

> citation("madrat")

Dietrich J, Baumstark L and Giannousakis A (2018). _madrat: May All Data 
be Reproducible and Transparent (MADRaT)_. doi: 10.5281/zenodo.1115490 
(URL: http://doi.org/10.5281/zenodo.1115490), R package version 1.44.0,
<URL: https://github.com/pik-piam/madrat>.

get your work cited!

Digital Object Identifiers (DOI) key for data or software citations

Services such as zenodo.org host code/data and provide DOI

GitHub Integration allows automatic DOI generation for every release

Open Data

Open Access of Data

Open Data vs Open Source

same idea but different licenses

Copyleft

  • Copyleft = Share-Alike (SA)
  • Examples:


     
  • BY = Attribution (Author needs to be mentioned)

 

Non-Copyleft

  • No Share-Alike
  • Examples:
  • central term: database
  • a data point cannot be protected, a database can!
  • CreativeCommons.org (CC) data license toolkit

ODbL

Databases and Derivatives

problems with fuzzy definition

Often unclear whether copyright rules apply or not:

  • Is it a database or just a data set?
  • Is it a derivative of a database or not?
    (e.g. model outputs)

Workaround for ambiguous cases: publish data processing routine instead of data

Source: Mimi and Eunice  | CC-BY-SA Nina Paley

Standardized input processing

powered by madrat » github.com/pik-piam/madrat

retrieveData: bundle data as required by the model

readSource: read source data and (dis-)aggregate to ISO country level

calcOutput: calculate data sets as required by the model

Open Access  and Transparency in FABLE

What is required?

Decisions to take

Decision

Open vs Non-Open

Copyleft vs Non-Copyleft

 

Recommended licenses

Contributors License Agreement (CLA)

Code distribution

Data distribution

Input data treatment

Output data license

 

 

MAgPIE (as example)

Open

Copyleft for model
Non-Copyleft for tools

copyleft: AGPL | non-copyleft: BSD2

model: yes | tools: no

GitHub

not decided

publication of data preparation routines

not specified (depends on application)

 

 

Further Reading

This presentation - slides.com/jandietrich/fable-open-access

 

PIK Guidelines

pik-potsdam.de/members/dietrich/GuidelinesOpenSource

pik-potsdam.de/members/dietrich/GuidelinesOpenData

 

Open Source

opensource.org/osd

gnu.org/licenses

 

Open Data

creativecommons.org

opendatacommons.org/licenses/odbl

How can FABLE ensure open access & transparency of data and models

By Jan Dietrich

How can FABLE ensure open access & transparency of data and models

  • 154

More from Jan Dietrich