Peewit Change Log
2014-3-21, changes in 0.9.5
- bugfix aggregation: change labels on aggregator-dim only
- do not peel (singleton) cubes from s-ancestores or t-successors, rather use get_value(s)
- more readable error for failed vcube-read
- bugfix build_plots
- bugfix partially_permute
- order contraint in vcube-chain drob
- bugfix baseclass: call ancestore init iff no of positional args fits
- mention indirect deps in echo production
- bugfix optional inputs
- shell as module function
- fixes for python2.7
- bugfix merge_straight
- updated example-packages
- bugfix test selection
- labels from var-names as e-node option
- bugfix arg-aggregation
- skip production for old input-indexes
- clean-up dependency resolution
- rewrite deploy
- joint dependencies for descent and ascent
- first implementation of looping: repeatedly deploy e-node for indefinite descent sizes
- bug-fix input-refs
- sibling-values in proper object instead of dict
- labels from var-names
- bugfix on producing semi-uniform v-cubes
- bugfix in vcube.deepcopy()
- example_project uses current libsvm interface
- allow absolute base-dir
- first implementation of descent by vcubes
- sibling-value access fixed
- rewrite plotting from v-cubes
- basic unit-tests
- core rewrite: new e-tree model
- dropped features:
- parent experiments
- multi-dim descent values
- run-numbers on plain-text-output and plots
- pull_peewit() working again
- For technical reasons we had to swtich from mercurial to git. archiving now uses git and development repo is a git-repo too.
- archivist uses private repo-dir
- removed guest access
- run_number in latex_snippets
- restore by run_number
- default prune_ascent and prune_condition
- overwrite base-dir by file in .peewit/
- test-install script in doc/
- archivist.export_project()
- bugfix import plotting
- bugfix pruned-ascent
- separate plot-module
- synchronization.py was missing in distro
- example_1.py contained faulty calls
- read v-cube by basename
- more v-cube manipulation: bold_merge, drop_singleton, partialy_permute_values
- bugfix in archiving
- explicit ascent makes leaf v-cubes holding a-value
- optional subtree dropping on ascent
- method repeat singleton value over given dim
- sync considers foreign states before pushing logs and v-cubes
- several renamings
- better separation core vs. paralellization
- state sensitive sync with remote head for logs and v-cubes
- bugfix: do not overwrite remote logs when lauching another experiment.
- removed needless spec of architecture for launching
- removed needless path exports for PBS-launching
- better example user-module on data reading
- removed some unneeded inter-dependencies
- bugfix v-cube merge
- bugfix parallelization: inconsitent command line args
- bugfix: omit statefile if no launcher is used
- double dashes for name part separation
- first implementation of semi-uniform v-cubes
- bugfix parallelization: get abs path of etree module when called with rel path
- bugfix remove current file in combination with launcher
- log build time for pv-tree
- verbose descent: argument is list of verbose nodes
- simpler loading of v-cubes
- default partname 'main' with and without launcher
- bugfix on relative paths to sync executables
- more robust job distribution through user space paths to sync executables
- ascent function can access v-cube of the computed subtree
- ancestor inputs get by default a box with pnode, vcube and d-values
- arg-aggregation can yield the objects from the pv-tree not only their value_names
- changed run call: launcher specification now separated from run
- remove_current_files() excludes log files
- launcher creates relevant dirs on remote side
- history files: no homedir but user in runs and user in filename
- better integration of path handling for launcher
- exclude history files when pulling from server
- history files are know commit with each run
- archiving service bug-fix: bug in reading history files
- archiving service bug-fix: history dir was ignored by hg
- archiving services: integration of version control system hg
such that older code can easily be restored.
The history of peewit itself is available through a public accessible repo for increasing the chances to experiments runnable. - log-files for single run experiemnts
- parallelization services: distribute partial computations and collect partial results into one v-cube
uses unison for syncronizing - clean-up service: remove run files of current experiment version
- platform-independent path-handling
- changed convention: e-nodes defined with lower-case letters
- run with verbose_descent-flag: prints descent calls with arguments given by value-names
- incremental e-tree definition with reuse of descent values
- rough merge of similar v-cubes
for examples v-cubes that stem from experiments related by incremental definition
- self-referential descent inputs
- alternate production dropped
Instead we plan to facilitate the run and merge of similar e-trees and vcubes.
- argaggregation rewrite