Home
Javier Sánchez edited this page Dec 20, 2017
·
14 revisions
Roadmap TODO and Roadmap DONE
This is mandatory to verify code correctness and ease within-versions and portability work (including Octave and Windows support).
- Method level: parameter(s) + dataset + expected performance
- Script level: scripts with CV parameter(s) + dataset + expected performance (TODO performance reference)
Tested platforms:
- Matlab's versions (DONE)
- Windows support
- Tested on Windows 10: Matlab 2017b and Octave 4.2
- Windows 8.1: Matlab 2016b and Octave 4.2
- Linux support:
- Tested on Linux Mint/Ubuntu 17.1: Matlab 2009a, 2013b, 2016b, 2017b and Octave 4.0
- Tested on Debian 7: Matlab 2009a
We have some errors and limitations to improve:
- [ Handling folds execution failure #8 ] (https://github.com/ayrna/orca/issues/8) (DONE)
- Parameters processing can confuse parameters with similar name #7 (DONE)
- Errors with highly imbalanced datasets, see Issue #9 (DONE)
- Add datasets loading tools (TODO?)
- Add k-fold and hold-out experimental designs provided a dataset name (TODO?)
- Perform experiments in all the datasets of a folder (DONE)
- Instalation and quick instalation (DONE)
- Tutorials:
- scripts (DONE)
- code example (DONE)
- how to add a method or a metric (TODO)
We need to standardize several aspects of methods (https://github.com/ayrna/orca/issues/6)
- Define common 'abstract' interface for all the methods, including functions that are not used in the experimenter but can be used externally (for instance train/test methods).
- Code documentation, indentation, variable naming convention...
- We have some methods that have to be divided in several methods to simplify the code (an example is experiments post-processing steps)
- Matlab has a coding guide that we may follow:
- Create a tool to generate data partitions for ORCA from a single file.