HAL changelog - 1.0.0 to 1.1.0
-------------------------------------------------------------------------------

User interface:

 * A help page has been added, with links from some of the more confusing pieces of the UI (blue boxes!)
 * Algorithm specifications can be cloned from a previously-specified algorith!
 * Initial movement toward bootstrap.js in some of the UI. More to come.
 * Target algorithm runs now have run status pages, accessible from the experiment run status page.
 * All child runs of an experiment are viewable in the run status page, with auto-refresh.
 * The Log table is now browseable/filterable in the UI.
 * Users can now give names to configuration experiment results, and use those names in further experiments.
 * Fixed some database contention with AJAX requests on experiment monitor pages.
 * HAL status page was refactored slightly.
 * HAL status page now polls adaptively, and the requests will never outrun the server.
 * Complete overhaul of algorithm and experiment specification, with more to come.
 * Some minor changes to the plots (titles, axes, tics)

Data management:

 * Many HAL objects are now populated lazily when retrieved from the database.
 * Query optimisation on most of the more complicated retrievals.
 * Some API cleanup to remove unused arguments and keys.
 * Retrieval of instance distributions and metaproblem instances is much speedier.
 * The log table is now circular (but large), with buffered commits, to avoid flooding.
 * Imports of objects with relative paths should now work, without causing a huge tree under imports/ to be created.


Execution management:
 * Retry scheduling has been made less aggressive.
 * Support for TORQUE clusters.
 * Runtime accounting has been made more consistent in some edge cases.
 * Execution managers can now decorate active runs with annotations (e.g. cluster job ids).
 * Local and SSH execution managers now log run output by default (into <hal directory>/runlogs)


HAL primitives:

 * Friedman test has been added to the statistics API.
 * Domains of type 'Algorithm' are now available for meta-algorithm parameters, in order to enforce constraints.
 * Metrics have been overhauled and divided into cases based on various termination criteria.
 * Initial support for instance features and feature extractors.


Other bug fixes:

 * Many edge cases with run termination, cluster/ssh execution, and output monitoring have been fixed.
 * Many potential deadlocks when using SQLite have been fixed.
 * Terminated runs will not be used to fulfill new run requests with the same parameters. Only runs that finished successfully will be used.