v2024.2.1

Version 2024.2.1 is a minor release.

New features

• NCU Reader Support for RAJAPerf Variants RAJA_CUDA and Lambda_CUDA (#201)
• Timeseries and calc_temporal_pattern (#88)
• Extend metadata_column_to_perfdata to multiple columns (#216)
• Confidence Interval Stats Function (#213)

Documentation updates

• Adds a page to the documentation for publications (#214)

Bug fixes

• Fix bug in add_ncu (#196)

Internal changes

• Run Unit Tests for Different Parameters (#182)
• Expose kwargs in caliperreader (#206)
• Check And Remove Duplicates in chosen_metrics (#202)

v2024.2.0

Version 2024.2.0 is a major release.

New features

• Overhaul of Distance Based Scoring Functions (#177)
• Adds function that moves perf metric to statsframe (#198)
• Add Root Node to Graph (#192)
• Add Nodes in Slice Before Printing Tree (#181)
• Enable Filtering From List of Profiles (#173)
• Speedup thicket composition by performing profile unions in binary tree order (#170)
• Implementation of query_stats function (#157)
• Add Option to Turn Off _fill_perfdata (#168)
• Add Pickle Reader/Writer Interface (#144)
• Print thicket tree by node order (hatchet nid) instead of alphabetically (#174)
• Add support for both row and column MultiIndex in dialects (#154)
• Improve NCU Reader (#60)

Documentation updates

• Update README.md (#211)
• Re-render notebooks for docs with no binder badge (#208)
• 2024 radiuss tutorial (#205, #204)
• Built-in configuration page link extension (#176)
• Update thicket tutorial with updates to display_violin_plot() (#165)

Bug fixes

• Fix Bug When Reading Directories (#203)
• Fix Indexing Issue in Tree for MultiIndex (#197)
• Fix intersection (#193)
• Fix Deepcopy Bug (#190)
• Moves re-exports of display_violinplot into the try-except-else guard (#160)

Internal changes

• Reset cache before pickle dump (#207)
• Remove Optional Argument from _fill_perfdata (#194)
• Pin numpy version to less than 2.0.0 (#191)
• Optimize Loop in Unify (#185)
• Remove Print in Groupby (#180)
• Check if Metric Exists in add_ncu (#178)
• Add new positional argument to unit test (#179)
• Change Filter Shallow Copies to Deep Copies (#175)
• Improve Reader Error Messages (#163)
• Check If filter_metadata Function Results in Empty Metadata Table (#164)
• Reinements to display_violinplot_thicket Function (#156)
• Remove 'not' from check as we want to skip if python version is below 3.8 (#162)
• Codecov support (#161)
• Avoid FutureWarning in ensemble.py (#171)

v2024.1.0

Version 2024.1.0 is a major release.

New features

• Add GroupBy Class and Aggregation Mechanism #85
• Updated percentile.py to allow user defined percentiles, and separating calculated values into separate columns #103
• Moves stats imports from top-level __init__.py to stats folder #132
• Replaces th. for th.stats. in various places #138
• Add per-node scoring functions #127
• Adds violin plot functionality along with testing #141
• Print Tree on Performance Data #118
• Add Progress Bar to Reader #131

Documentation updates

• update output of jupyter notebooks in docs #117
• Add output of statsframe tutorial notebook #119
• Add output of query language notebook #120
• render notebooks for 2023.3.1 release #121, #122
• fix apidocs rendering #124
• fix URL links in README #143
• update tree and performance table figures in user guide #126
• add executed notebooks for hatchet v2024.1.0 #146, #149
• execute notebooks remove legacy data references, formatting #150

Bug fixes

• fix type checks in groupby #123
• Disable Node Ordering if All Profiles Don't Have It #128
• Expand Check to All Inner Indices #130
• Synchronize Profile Components #145
• Fix filter_stats #148

Internal changes

• Update Extra-P Unit Tests to Match Latest Release #115
• Add Testing for Display Functions #87
• Improve Compatibility in concat_thickets #89
• Rework _unify #129
• update test data with newer rajaperf #125
• Improve Validation Checks #142
• Prevent test failure if Extra-P is not installed #136

v2023.1.1

v2023.3.1

bump version to fix missing external module

v2023.3.0

Version 2023.3.0 is a major release.

New features

• Thicket-specific tree renderer
• StatsFrame MultiIndex Squash
• One Profile per Hash
• Fill Performance DataFrame

Documentation updates

• add new page with tutorial materials
• add CITATION.cff for HPDC submission
• change figs to match tree and dataframe
• update readme to point to docs
• Docs fixes
• add thicket documentation and tutorial through read the docs

Bug fixes

• Add Pandas Groupby Functionality to Stats Functions
• Fix tree indexing to correspond with statsframe
• Intersection Bugfix
• Fix unify_ensemble handling of None
• Bugfix reader_dispatch
• query: return new copy of thicket

Internal changes

• Refactor Ensemble Functions & Change Ensemble APIs
• add initial tests for stats
• Preference Redesign
• Separate .set_title() from initial sns.heatmap call
• Relax Verify Thicket Structure Check for Stats Functions
• StatsFrame change name to a column index, not column
• Return dictionary instead of list in groupby
• Adds automatic ownership test and fixes
• Sort performance data index after columnar_join
• Use Deterministic Hash Algorithm
• Adds usability of show_metric_columns() to statistic functions
• Fixes the dependencies in setup.py and adds a script to autogenerate new requirements.txt files from setup.py

v2023.2.0

This is the first official release of Thicket, a toolkit for exploratory data analysis of multi-dimensional performance data.