Roadmap » History » Version 33
Version 32 (Marcus Pearce, 2014-07-17 08:16 PM) → Version 33/34 (Marcus Pearce, 2015-03-26 10:06 PM)
h1. Roadmap
h2. Installation
* -Remove absolute pathname in connect-to-database. (mtp-admin/music-data.lisp)-
* -Create cache directories if they don't exist.-
h2. Data import
* -It is possible to import empty datasets, which cause an error when described.-
* -running commands on an empty database produces an error (e.g., <code>mtp-admin:describe-database</code>)-
h2. Viewpoints
* Polyphonic viewpoints for modelling harmonic movement
* A straightforward language for specifying viewpoints, including viewpoint schemas (e.g. interval, interval size being target viewpoints), making system more data agnostic
* Zero barlengths sometimes cause divide by zero errors
* Implement a 'Pace' viewpoint: measure of information rate (bits/sec), analogous to flow in speech production.
h2. Viewpoint selection
* Memory errors sometimes occur with large viewpoints sets (e.g., Cpitch with no basis specified)
* When decimal places are restricted for comparison, earlier systems are preferred within a round, so use full precision to choose between ties
* Print trace information about VP sets being tested + mean IC values; record this to log file.
* -Optionally Optionally specify: min-links- min-links
* More flexible way for user to specify constraints on viewpoint search:
** Define labelled viewpoint classes
** Pairs/triples of labels/wildcards specify acceptable combinations
** User provides whitelist or blacklist spec
* Provide recommended viewpoints for more basic viewpoints than just cpitch, onset & bioi
h2. Modelling
* Predict over more than one dataset.
* Allow multiple memory stores (i.e., not restricted to just two: the LTM and STM).
* Hierarchical structure: chunk common patterns into symbols using information content and entropy as indicators of grouping structure (Pearce et al., 2010, Perception)
* Compute predictive information (PI), expected PI and PI rate as analogs to IC, entropy and entropy rate respectively (Abdallah & Plumbley, 2009).
* Offer the selection of alternative context strategies (e.g., include future context)
* Allow the user to fix the weights for particular combinations, e.g. weighted viewpoints, or weighted memory stores (LTM/STM etc.)
* Allow the user to determine order of model combination (e.g., LTM-STM first, then viewpoints vs viewpoints first then LTM/STM)
h2. Longer-term goals
Interface
* Implement as a web service to avoid installation woes.
Efficiency:
** Extend caching of results
** Use sampling to estimate mean IC during VP selection.
** Optimise viewpoint selection based on match with existing IC values.
Testing
** Include unit testing code.
h2. Installation
* -Remove absolute pathname in connect-to-database. (mtp-admin/music-data.lisp)-
* -Create cache directories if they don't exist.-
h2. Data import
* -It is possible to import empty datasets, which cause an error when described.-
* -running commands on an empty database produces an error (e.g., <code>mtp-admin:describe-database</code>)-
h2. Viewpoints
* Polyphonic viewpoints for modelling harmonic movement
* A straightforward language for specifying viewpoints, including viewpoint schemas (e.g. interval, interval size being target viewpoints), making system more data agnostic
* Zero barlengths sometimes cause divide by zero errors
* Implement a 'Pace' viewpoint: measure of information rate (bits/sec), analogous to flow in speech production.
h2. Viewpoint selection
* Memory errors sometimes occur with large viewpoints sets (e.g., Cpitch with no basis specified)
* When decimal places are restricted for comparison, earlier systems are preferred within a round, so use full precision to choose between ties
* Print trace information about VP sets being tested + mean IC values; record this to log file.
* -Optionally Optionally specify: min-links- min-links
* More flexible way for user to specify constraints on viewpoint search:
** Define labelled viewpoint classes
** Pairs/triples of labels/wildcards specify acceptable combinations
** User provides whitelist or blacklist spec
* Provide recommended viewpoints for more basic viewpoints than just cpitch, onset & bioi
h2. Modelling
* Predict over more than one dataset.
* Allow multiple memory stores (i.e., not restricted to just two: the LTM and STM).
* Hierarchical structure: chunk common patterns into symbols using information content and entropy as indicators of grouping structure (Pearce et al., 2010, Perception)
* Compute predictive information (PI), expected PI and PI rate as analogs to IC, entropy and entropy rate respectively (Abdallah & Plumbley, 2009).
* Offer the selection of alternative context strategies (e.g., include future context)
* Allow the user to fix the weights for particular combinations, e.g. weighted viewpoints, or weighted memory stores (LTM/STM etc.)
* Allow the user to determine order of model combination (e.g., LTM-STM first, then viewpoints vs viewpoints first then LTM/STM)
h2. Longer-term goals
Interface
* Implement as a web service to avoid installation woes.
Efficiency:
** Extend caching of results
** Use sampling to estimate mean IC during VP selection.
** Optimise viewpoint selection based on match with existing IC values.
Testing
** Include unit testing code.