What is New in StaticFrame
2.17.0
Added inventory attribute to Bus, Yarn, and Quilt, exposing, if relevant, information about file-based data stores.
Normalized interfaces of exception subclasses to permit using copy and deepcopy.
Corrected issue in TypeBlocks that, for a single row Frame and a single column selection, could result in an unexpected 2D array.
2.16.1
Added persist(), persist[], persist.loc[] and persist.iloc[] interfaces to Bus and Yarn to selectively pre-load Frame to improve performance when loading from a store.
2.16.0
Improvements to Bus to make max_persist usage maximally lazy, i.e., referenced Frame are never loaded until extracted. Selections that produce a new Bus will no longer load Frame.
Corrected issue in Bus whereby max_persist usage can lead to unexpected StopIteration exceptions.
Bus now raises an exception if max_persist is less than 1.
Added Frame.to_sql().
Added abs() method to all containers.
Improved error message when looking up invalid datetime64 values in datetime64 indices.
Improved error messages when attempting to use __setitem__ on immutable containers.
Updated mypy to 1.13.0.
2.15.1
Corrected issues in using via_fill_value with containers using IndexHierarchy.
Corrected issue when extending a FrameGO with a column label type that is not compatible with the existing column Index.
2.15.0
Frame.unset_index(), Frame.unset_columns() now have a drop parameter.
Improvements to Quilt.to_frame() and related interfaces when contained Bus have specialized indices.
Set arraykit version to 0.10.0.
Set arraymap version to 0.4.0.
Added support for Python 3.13.
2.14.0
Added Series.via_mapping interface to provide a fully type-compliant Mapping subclass.
Performance enhancements to reindex and related IndexHierarchy operations.
2.13.2
Corrected Series and Frame assignment such assigned tuples are always treated as elements.
2.13.1
Corrected issue when using iloc_min, iloc_max, loc_min, loc_max on containers with IndexHierarchy.
Corrected issue when reindexing a Frame when both axis introduce fill values.
Improved via_fill_value interfaces to retain index and columns name attribues when defined.
2.13.0
Added Frame.merge_left(), Frame.merge_right(), Frame.merge_inner(), and Frame.merge_outer().
Updated arraykit to 0.9.0.
2.12.0
Added via_dt.year_quarter.
Corrected recursive type definition of tuples within TLabel.
Updated arraykit to 0.8.3.
Updated mypy to 1.11.1.
2.11.0
Added Frame.reduce and Batch.reduce.
Added reduce interfaces on Frame and Bus iterators.
All via_hashlib interfaces will now raise if an object dtype is encountered.
Improved identification of mappings by using collections.abc.Mapping.
2.10.3
Corrected issue in package publishing.
2.10.2
Corrected issue in Frame.pivot whereby, under some scenarios, fill_value is not applied.
2.10.1
Updated arraykit to 0.8.2.
2.10.0
Support for NumPy 2.
Updated arraykit to 0.8.0.
Updated arraymap to 0.3.0.
Updated frame-fixtures to 1.1.0.
2.9.0
Added preliminary support for DuckDB with Frame.to_duckdb(), Frame.from_duckdb(), Bus.to_duckdb(), Bus.from_duckdb(), Quilt.to_duckdb(), Quilt.from_duckdb(), Batch.to_duckdb(), Batch.from_duckdb() and Yarn.to_duckdb().
Added name parameter to Frame.from_xlsx(), Frame.from_sqlite(), and Frame.from_hdf5()`
Optimized routines for column iteration and Frame.transpose().
2.8.2
Updated arraykit to 0.7.2. Integrated enhancements in all 2D array to tuple routines.
2.8.1
Updated arraykit to 0.6.3.
2.8.0
New implementation, improved performance, and corrected issues in Frame.join_left(), Frame.join_right(), Frame.join_inner(), and Frame.join_outer().
Updated arraykit to 0.6.2.
Updated typing-extensions to 4.12.0.
2.7.0
CalllGuard and other TypeClinic interfaces now support run-time usage of TypeVar.
2.6.0
Internal redesign of Yarn to permit arbitrary reording of Frame in contained Bus.
Added reindex(), sort_index(), sort_values(), roll(), and shift() methods to Yarn.
Improved bytes representation used by Yarn.via_hashlib to exclude internal implementation configuration.
2.5.2
Improvements to Yarn to always support unique labels after retain_labels application.
2.5.1
Extended TypeClinic and CallGuard support for generic NumPy generic.
2.5.0
Removed support for Python 3.8.
Set minimum numpy to 1.22.4.
Set minimum mypy to 1.9.0.
Removed support for pre 1.0 Pandas.
Updated minimum versions of optional packages.
2.4.0
Added isna(), notna(), isfalsy(), and notfalsy() methods to all IndexBase subclasses.
2.3.0
All prod() and sum() methods now have an allna parameter to, when fillna is True, specify a value if all values are NA.
Improved type hints for interfaces that accept IO objects.
URLs given to the WWW interface now escape spaces in path components.
Performance enhancements to low-level TypeBlocks routines.
2.2.3
Corrected issue with Bus with IndexHierarchy and max_persist greater than one.
2.2.2
Corrected handling of ZIP64 extensions in zip_namelist.
2.2.1
Corrected issue with non-sequential Bus selections with max_persist greater than one, resulting in incorrectly sized generators.
Performance enhancements to Bus loading routines with max_persist.
Corrected handling of ZIP64 extensions in ZipFileRO.
2.2.0
Performance enhancements to reading NPZ files and ZIP NPZ archives.
2.1.2
ContainerBase added to the public namespace.
2.1.1
Updated release automation in CI.
2.1.0
Added TSeriesAny, TSeriesHEAny, TBusAny generic aliases.
Added Frame.to_json_typed() and Frame.from_json_typed(), supporting optimized, complete round-trip Frame serialization.
Set arraykit to 0.5.1, now with functional wheels for Mac OS arm64 / Apple Silicon.
Set arraymap to 0.2.2, now with functional wheels for Mac OS arm64 / Apple Silicon.
Added support for Python 3.12.
2.0.1
Corrected import in setup.py.
2.0.0
Nearly all containers are now generic, permitting specification of component types for static analysis and runtime validation.
Added CallGuard, exposing decorators for performing run-time function interface type check and data validation.
Added Require, a collection of classes to be used for run-time validation in tp.Annotated generics.
Added TypeClinic, ClinicResult, and ClinicError, providing type to hint conversion and checks against arbitrary hints.
Added via_type_clinic interfaces to all containers, exposing a pre-configured TypeClinic interface.
Now performing static-analysis with both MyPy (1.6.1) and Pyright (1.1.331).
1.6.5
Implemented fall-back behavior for Parquet files written with pyarrow less than 1.0.
1.6.4
Improvements to typing through usage of overload signature definitions.
Set arraykit to 0.4.10.
1.6.3
Added label_missing_skips to all iter_window and related interfaces.
1.6.2
Added label_missing_raises to all iter_window and related interfaces.
Set typing-extensions to greater than or equal to 4.7.1
1.6.1
Corrected exception raised in IndexHierarchy.level_drop() when provided count is 0.
Added typing-extensions as a dependency for back-ported typing features.
Updated mypy configuration for broader checks.
Numerous typing improvements and enhancements.
1.6.0
Comprehensive improvements and expansions to type hints using Python 3.11 and NumPy 1.25.
1.5.0
Removed support for Python 3.7.
Set minimum numpy to 1.19.5.
Improved type hints for loc and iloc interfaces on 2D containers.
Usage of from __future__ import annotations in all modules.
Normalized exceptions raised from invalid selections to raise KeyError.
1.4.6
Set arraykit to 0.4.9.
1.4.5
Imports of ThreadPoolExecutor and ProcessPoolExecutor are now deferred until used.
Multiprocessing interfaces that use ProcessPoolExecutor now accept the mp_context parameter.
1.4.4
Set arraymap to 0.1.9.
Updated TypeBlocks to use BlockIndex.iter_block().
1.4.3
Integrated arraykit.BlockIndex in TypeBlocks, offering significant Frame performance improvements.
via_dt will now raise if an NA is encountered and a fill_value is not provided.
Corrected issue in Quilt construction when component Frame have IndexHierarchy.
Corrected issue when using an Index as a selector in via_fill_value().loc[].
Added ErrorInitColumns to better indicated failures in column Index construction.
1.4.2
Improved performance when assigning an Index with FrameGO.__setitem__().
Improved typing and input handling for StoreFilter values.
Set arraymap to 0.1.8.
Removed deprecated JSON interfaces: Frame.from_json(), Frame.from_json_url().
1.4.1
Optimizations to reduce TypeBlocks size and initialization time.
1.4.0
Replaces automap with arraymap 0.1.7, offering optimized Index performance.
1.3.2
Corrected issue with Frame.iter_element().apply() whereby columns.name are not preserved.
Corrected issue when selecting a single row from a Frame that has object columns that contain sequences.
Set arraykit version to 0.3.4.
1.3.1
Set arraykit version to 0.3.1.
1.3.0
The via_dt interface now supports usage of a fill_value argument for handling missing values.
Added unique_enumerated() method to Series and Frame.
1.2.7
Set arraykit version to 0.3.0: replaced usage of np.nonzero() with first_true_1d() and first_true_2d().
Specifiers of dtypes given to astype() and related interfaces will now permit “object” to be used (in addition to object).
1.2.6
Corrected issue when selecting IndexHierarchy labels that contain tuples.
IndexHiearchy.astype() now properly delegates the name attribute.
Better usage of np.concatenate in TypeBlocks.transpose() and HierarchicalLocMap.build_offsets_and_overflow().
1.2.5
Specifiers of dtypes given to astype() and related interfaces will now raise if NumPy will implicitly convert the argument to object.
Batch.to_frame() now exposes index_constructor, columns_constructor arguments.
Improvements to usage of index_constructor and columns_constructor arguments in Frame.from_concat when Series are provided as components.
When using a Batch to write to an archive, non-unique labels will now raise a StoreLabelNonUnique exception.
1.2.4
IndexHierarchy.__setstate__ now properly sets indexers to be immutable.
Bus with associated Store instances are now pickleable after removing _weak_cache in Store.__getstate__().
1.2.3
Series.isna(), Series.dropna(), and related functions now properly handle arrays that contain objects such as Frame or np.ndarray that raise for usage of __bool__().
Set arraykit version to 0.2.9: isna_element now identifies pd.Timestamp('nat') as a NA value, and invalid datetime64 strings given to Frame.from_delimited and related interfaces now properly raise exceptions.
1.2.2
Corrected issue in IndexHierarchy.label_widths_at_depth() which caused incorrect cell merging in XLSX output when merge_hierarchical_labels is True.
1.2.1
Corrected issue in Series.dropna() whereby full drops would not retain the index class.
Performance enhancement to TypeBlocks.equals() and related routines using arrays_equal().
1.2.0
Significant performance optimizations to IndexHierarchy set operations, as well as optimized pathways for determining TypeBlocks equality.
JSON metadata in NPY and NPZ encodings of Frame data now properly encodes and decodes datetime64 and datetime.date objects.
Corrected issue on Python 3.11 in the creation of memory displays due to usage of Enum.
Corrected issue in Frame.relabel_shift_out() where index_constructors are not assigned to subset index.
Extended Frame.iter_tuple() constructor argument to support dataclass-created classes.
1.1.1
Corrected handling of 0-sized containers in Frame.insert_before() and Frame.insert_after().
Corrected issue with some IndexHierarchy formations when using slices in an HLoc with more than one depth selection.
1.1.0
Added Frame.consolidate interface, including Frame.consolidate[] and Frame.consolidate.status.
Added Quilt.bus property.
IndexHierarchy.rehierarch() and related routines now correctly reorder index constructors by default.
Added index_constructors arguments to IndexHierarchy.rehierarch(), Series.rehierarch(), Bus.rehierarch(), Yarn.rehierarch().
Added index_constructors, columns_constructors arguments to Frame.rehierarch().
1.0.1
Parameters dtypes, fill_value, and format (given to via_str.format) now properly work with defaultdict, infinite iterators, and mappings indexed by position (when columns are not defined or created with IndexAutoFactory).
1.0.0
API change: IndexHierarchy numerical and statistical methods, such as sum(), var(), std(), cumprod(), cumsum(), now raise NotImplementedError.
API change: Frame.astype[] calls now set consolidate_blocks to False by default.
API change: composite_index and composite_index_fill_value parameters removed from Frame.join_left, Frame.join_right, Frame.join_inner, and Frame.join_outer; added include_index parameter.
API change: IndexDefaultFactory renamed IndexDefaultConstructorFactory.
Added via_dt.year_month.
0.9.23
Added via_hashlib interface to all containers.
Added Frame.iter_group_other(), Frame.iter_group_other_items(), Frame.iter_group_other_array(), Frame.iter_group_other_array_items().
Added Series.iter_group_other(), Series.iter_group_other_items(), Series.iter_group_other_array(), Series.iter_group_other_array_items().
Set arraykit version to 0.2.6.
0.9.22
IndexYear now accepts selection by integers for years.
Added Series.loc_notna_first(), Series.loc_notna_last(), Series.loc_notfalsy_first(), Series.loc_notfalsy_last().
Added Frame.loc_notna_first(), Frame.loc_notna_last(), Frame.loc_notfalsy_first(), Frame.loc_notfalsy_last().
Set arraykit version to 0.2.4.
0.9.21
Set arraykit version to 0.2.3.
Set automap version to 0.6.2.
Improvements to delimited file parsing when index_depth is greater than zero and quoted fields or escaped characters are used.
Corrected issue where Pandas MutliIndex are not converted to IndexHierarchy in Frame.from_pandas() or Series.from_pandas().
0.9.20
Added via_str.format() for applying Python formatting mini-language strings to elements.
Added skip_initial_space, quoting, quote_double, and escape_char parameters to Frame.from_delimited() and related interfaces.
Added Frame.from_json_index(), Frame.from_json_columns, Frame.from_json_split(), Frame.from_json_records(), Frame.from_json_values().
Added Frame.to_json_index(), Frame.to_json_columns, Frame.to_json_split(), Frame.to_json_records(), Frame.to_json_values.
Added axis argument to Series.count() for compatibility with Frame interface.
0.9.19
Frame.from_delimited now powered by arraykit.delimited_to_arrays, offering vastly improved performance for all delimited file reading.
Added Series.from_delimited.
Added WWW interface for downloading network resources to provide to constructors.
Added columns_select, skip_initial_space, quoting, quote_double, escape_char, thousands_char, and decimal_char parameters to Frame.from_delimited and derived interfaces.
Corrected issue in converting to IndexHierarchy from Pandas MultiIndex when the MultiIndex is bloated.
0.9.18
Improved layout of default memory display.
Corrected issue where a non matching Series assignment or reindex might use fill_value to determine the type of the returned values.
Certain invalid Frame selections now properly raise KeyError.
0.9.17
Unified implementation of IndexBase set and concatenation operations.
Added IndexHierarchy.from_values_per_depth() constructor.
Set automap version to 0.6.1.
Extended Batch.astype interface to fully cover Frame.astype interface.
0.9.16
Added memory property to display memory usage via MemoryDisplay interfaces.
Implemented the DataFrame Interchange Protocol export via the Frame.__dataframe__() interface.
Implemented __repr__() for ILoc.
Updated Batch.__repr__().
Added name parameter to Frame.to_frame(), Series.to_frame(), and related methods.
Improved error reporting for invalid IndexHierarchy.
IndexHierarchy.from_index_items() now supports items of IndexHierarchy.
0.9.15
Added __repr__() for HLoc.
Added IndexHierarchy.index_at_depth()
Added IndexHierarchy.indexer_at_depth()
Added depth_level and order_by_occurrence parameters to unique() on IndexBase subclasses.
Corrected issue calling Frame.to_pandas() with an empty Frame.
Implemented argument checking on all IndexHierarchy depth selection parameters.
IndexHierarchy.astype() now accepts a dtypes argument to assign dtypes by tuple or mapping.
Corrected return type of via_str.contains() to return Booleans.
0.9.14
Added Frame.corr(), Series.corr(), and Batch.corr().
Added compression argument to StoreClientMixin exporters that write ZIPs.
Corrected issue with selection on zero-sized IndexHierarchy.
0.9.13
Added to_zip_npy() and from_zip_npy() interfaces to Bus, Batch, Yarn, and Quilt.
Added test_example_gen.py to test and enforce automatic example generation.
Improved usage of __slots__ throughout.
Continuous integration quality checks now using pylint 2.15.0 and isort 5.10.1
Minimum pyarrow set to 0.17.0
0.9.12
ErrorNPYEncode exceptions raised during authoring NPZ files or NPY directories now remove those files or directories.
ErrorNPYEncode exceptions raised during to_zip_npz() now remove the archive.
Authoring NPYs to a pre-existing directory will now raise a RuntimeError.
0.9.11
Supplying an IndexDatetime subclass as an explicit_constructor to an IndexAutoFactory now raises.
Corrected issue with fillna and fillfalsy functions when non-elemental fill values are supplied.
Updated arraykit to 0.1.13.
0.9.10
Corrected single depth selection issue with IndexHierarchy.
0.9.9
Updated arraykit to 0.1.12.
Set minimum numpy to 1.18.5.
Added index_constructors parameter to Frame.relabel_shift_in().
Corrected issue when Frame.astype() called with an empty Frame.
Extended via_values property to take optional consolidation arguments via __call__() constructor; usage of via_values instance returns same-typed, same-sized container.
0.9.8
Added via_values property to Series, Index, Frame, IndexHierarchy and Batch; permits applying functions to complete containers with apply() and supports usage as arguments in arbitrary NumPy functions with __array_ufunc__().
Corrected usage of IterNodeDelegate with iterator endpoints that do not iterate hashables; added IterNodeDelegateMapable for usage with iterators of hashables.
Improved type and dtype preservation in concatenation and set operations on IndexHierarchy.
Normalized ordering of results from Frame.bloc[] selections to row-major ordering without sorting labels.
Added via_str.contains().
Corrected issue in ArchiveZIP when __del__ is called when no archive is set.
Frame.to_sqlite() now requires a named Frame or an explicit label; Frame.from_sqlite(), Frame.from_hdf5() now make label a required argument.
API Documentation re-organized, now using procedurally generated code examples.
0.9.7
Corrected issue in Series.from_overlay() that prematurely aborted processing all Series
Normalized ordering of results from Frame.bloc[] selections.
0.9.6
Corrected issue in Quilt creation when given a Bus with Frame with datetime64 indices.
Extended IndexAutoConstructorFactory to evaluate dtype from arbitrary iterables, not just arrays.
Improvements to consistency and performance of loc_to_iloc.
Implemented max and min methods on IndexHierarchy; related statistical methods now raise.
0.9.5
Updated AutoMap to 0.5.1
Removed “performance” package from setup.py.
0.9.4
Enhanced support for fill_value as a FillValueAuto, a mapping, or a sequence of fill values (per column) where appropriate.
Added Index.dropna(), Index.dropfalsy(), IndexHierarchy.dropna(), IndexHierarchy.dropfalsy().
Added Index.fillfalsy(), IndexHierarchy.fillfalsy().
Performance improvements for Frame.iter_group_labels(), Frame.iter_group_labels_items(), Frame.iter_group_labels_array(), Frame.iter_group_labels_array_items().
Fixed usage of dtypes argument when encountering zero-sized data in Frame.from_records() and Frame.from_pandas().
Improved Frame.iter_tuple to not coerce types through arrays.
Added Frame.set_columns(), Frame.set_columns_hierarchy(), and Frame.unset_columns().
0.9.3
Added apply_element() and apply_element_items() methods to FrameAssign and SeriesAssign interfaces.
Added implementation of __array__() and __array_ufunc__() to all containers for better support with NumPy objects and binary operators.
Added Series.iter_group_array(), Series.iter_group_array_items(), Series.iter_group_labels_array(), Series.iter_group_labels_array_items().
Added Frame.iter_group_array(), Frame.iter_group_array_items(), Frame.iter_group_labels_array(), Frame.iter_group_labels_array_items().
Corrected issue when using binary operators with a FrameGO and a Series.
Corrected issue and performance of name assignment when extracting Series from Frame with an IndexHierarchy.
Added IndexAutoConstructorFactory for automatic constructor selection based on NumPy dtype.
0.9.2
Corrected more issues when calling IndexHierarchy.loc[] with another IndexHierarchy, or when calling Frame.assign.apply when that frame has IndexHierarchy columns.
Corrected undesirable type coercion from happening in single-row selections from IndexHierarchy.
0.9.1
Corrected issue when calling IndexDatetime.loc[] with an empty list.
Corrected issue when calling IndexHierarchy.loc[] with another IndexHierarchy
0.9.0
API change: Bus no longer accepts a Series on initialization; use Bus.from_series().
API change: Batch no longer normalizes containers after each step in processing; use Batch.via_container to force elements or arrays to Frame or Series.
API change: Index objects can no longer be created with datetime64 arrays; such labels must use an IndexDatetime subclass instead. If this is happening implicitly with an operation, that operation should expose a parameter for index_constructor or index_constructors.
API change: IndexAutoFactory is no longer accepted as an index_constructor argument in Series.from_pandas() and Frame.from_pandas(); IndexAutoFactory should be passed as an index or columns argument instead.
Minimum Python version is now 3.7
New implementation of IndexHierarchy, offering significantly improved performance and removal of the requirement of tree hierarchies.
Added Batch.to_series().
Fixed issue when using Frame.from_npz with an NPZ created with a FrameGO.
Fixed issue when supplying overspecified mappings to Frame.astype.
0.8.38
Further improved handling of binary equality operators with IndexDatetime subclasses.
0.8.37
Improved handling of binary equality operators with IndexDatetime subclasses.
0.8.36
Extended interface of Batch to include all methods for handling missing values, as well as all via_* interfaces.
Silenced all NumPy warnings where the issue raised in the warning is being explicitly handled in the code.
0.8.35
Performance enhancements to Frame.pivot(), Frame.iter_group(), and Frame.iter_group_items().
Frame.pivot() func parameter can now be set to None to perform no aggregation.
Extended Series.from_overlay() and Frame.from_overlay() to support func and fill_value arguments; func can be used to optionally specify what elements are available for assignment in overlay.
Extended via_fill_value() interfaces to implement __getitem__ and loc selection interfaces on Series and Frame for selections that potentially contain new labels filled with the fill value.
0.8.34
Added NPY and NPZ interfaces for creating NPY and NPZ archvies from arrays and Frame components.
Added index_constructors argument to IndexHierarchy.from_product()
Added index_constructor argument to Index.level_add(), IndexHierarchy.level_add(), and Frame.relabel_level_add().
Added index_constructor, columns_constructor arguments to Frame.relabel().
Added Series.to_frame_he().
Added index, index_constructor, columns, columns_constructor arguments to Series.to_frame(), Series.to_frame_go(), Series.to_frame_he().
Improvement to Frame.from_concat to avoid creating one-element indices from Series when an index is provided along the appropriate axis.
Added index_constructor argument to Series.relabel().
Added index_constructor argument to Series.from_concat().
Added index argument to Series.from_pandas().
Added index and columns argument to Frame.from_pandas().
Improvement to Index initialization to raise ErrorInitIndex if given a single string as labels.
Set operations on labels of different datetime64 units now raise an Exception.
0.8.33
Performance enhancements to Frame.from_npy and Frame.from_npz.
0.8.32
Added Frame.to_pickle(), Frame.from_pickle().
Added index_constructor, columns_constructor to Frame.from_concat.
Fixed issue in Frame.insert_after(), Frame.insert_before(), Series.insert_after(), Series.insert_before() with negative ILoc labels.
0.8.31
Added Frame.from_npy_mmap; removed memory_map option from Frame.from_npy.
0.8.30
Performance enhancements to Frame.from_npy and Frame.from_npz.
0.8.29
Added consolidate_blocks Boolean parameter to Frame.to_npz() and Frame.to_npy.
0.8.28
Added Frame.to_npy(), Frame.from_npy() with a memory_map option.
Improvements to Frame.to_npz() to support large files and buffered writes.
Performance enhancements to all _StoreZip subclasses through usage of WeakValueDictionary caching.
Added IndexHiearchy.relabel_at_depth().
Added support for string slicing and selection with Series.via_str[] and Frame.via_str[].
0.8.27
Reimplemented Frame.to_npz(), Frame.from_npz(), removing support for object arrays (and pickles) and improving performance.
Added Bus.to_zip_npz(), Bus.from_zip_npz(), Quilt.to_zip_npz(), Quilt.from_zip_npz(), Batch.to_zip_npz(), Batch.from_zip_npz() and Yarn.to_zip_npz().
Implemented Series.fillfalsy_forward(), Series.fillfalsy_backward(), Series.fillfalsy_leading(), Series.fillfalsy_trailing().
Implemented Frame.fillfalsy_forward(), Frame.fillfalsy_backward(), Frame.fillfalsy_leading(), Frame.fillfalsy_trailing().
Added Quilt.equals().
Frame.from_pandas() now supports zero-sized DataFrame.
Fixed issue in Frame.set_index() where column is passed as None.
Removed TypeBlocks._block_slices.
0.8.26
Frame.to_pandas() now creates pd.RangeIndex for IndexAutoFactory-created indices.
Performance enhancements to Frame.from_concat().
0.8.25
Corrected issue extracting containers stored in Series.
0.8.24
Improved dtype resoltion on Frame methods that reduce dimensionality.
0.8.23
Corrected issue where summing a Frame of Booleans along axis 0 resulted in Booleans instead of integers.
0.8.22
Performance enhancements to Frame.iter_group() and Frame.iter_group_items().
0.8.21
Added Frame.to_npz(), Frame.from_npz().
Performance enhancements to Frame.iter_group() and Frame.iter_group_items().
Performance enhancements to Frame.pivot().
Added drop parameter to Frame.iter_group() and Frame.iter_group_items().
Introduction of TypeBlocks._block_slices as lazily derived and persistently stored.
Fixed issue with Frame.from_overlay when called with FrameGO.
Added index_constructor argument to apply, apply_pool, map_any, map_fill, map_all.
0.8.20
Added dtypes parameter to Frame.from_pandas().
Added index_constructors, columns_constructors to the following interfaces: Frame.from_sql(), Frame.from_structured_array(), Frame.from_delimited(), Frame.from_csv(), Frame.from_clipboard, Frame.from_tsv(), Frame.from_xlsx(), Frame.from_sqlite(), Frame.from_hdf5(), Frame.from_arrrow(), Frame.from_parquet().
StoreConfig now exposes index_constructors and columns_constructors arguments.
Incorrectly formed Batch iterables will now, upon iteration, raise a BatchIterableInvalid exception.
Added Quilt.sample().
all() and any() on Series and Frame no longer raise when NA values are present and skipna is False.
Performance enhancements to Bus loading routines when using max_persist by refactoring internal architecture of Bus to no longer hold a reference to a Series but instead use a mutable array.
0.8.19
Optimization of Bus.items(), Bus.values, Bus.iter_element(), and Bus.iter_element_items() when max_persist is greater than one.
Added Yarn.iter_element(), Yarn.iter_element_items().
Added Yarn.drop[]
Added Yarn.reindex(), Yarn.relabel_flat(), Yarn.relabel_level_add(), Yarn.relabel_level_drop(), Yarn.rehierarch().
Added Bus.unpersist(), Yarn.unpersist(), and Quilt.unpersist().
Improvements to standard string representation of Quilt.
Added is_month_start(), is_month_end(), is_year_start(), is_year_end(), is_quarter_start(), is_quarter_end() to via_dt interfaces.
Added hour, minute, second properties to via_dt interfaces.
Improved implementation of weekday(), added quarter() to via_dt interfaces.
Fixed issue when using iter_window_* methods on two-dimensional containers where the opposite axis is not a default index constructor.
Fixed issue when selecting rows from Frame with 0-length columns.
0.8.18
Implementation of Yarn(), a container that presents numerous Bus as a uniform, 1D interface.
Fixed issue in Frame.astype[] when selecting targets with a Boolean Series or arrays.
Fixed unnecessary type coercion in the Frame returned by Frame.drop_duplicated().
Improved handling of reindexing and lookups between datetime64 and date / datetime objects.
Frame.equals(), Series.equals(), Index.equals(), IndexHiearchy.equals() and all related routines now distinguish by datetime64 unit in evaluating basic equality.
0.8.17
Extended Series.count() and Frame.count() with skipfalsy and unique parameters.
Added Series.isfalsy(), Series.notfalsy(), Series.dropfalsy(), Series.fillfalsy().
Added Frame.isfalsy(), Frame.notfalsy(), Frame.dropfalsy(), Frame.fillfalsy().
Exposed isna_element() via arraykit on root namespace.
Added Bus.from_concat().
Added Bus.to_series().
Bus.reindex(), Bus.relabel(), Bus.relabel_flat(), Bus.relabel_level_add(), Bus.relabel_level_drop(), Bus.rehierarch() now, if necessary, load all contents from the associated Store and return a Bus without a Store association.
Added index_constructor argument to Series.from_concat_items().
Added index_constructor, columns_constructor arguments to Frame.from_concat_items().
Introduced IndexDefaultConstructorFactory to permit specifying index name attributes with default index constructors.
0.8.16
Added Frame.to_series().
Frame.sort_values(), Frame.sort_index(), Frame.sort_columns, Series.sort_index(), and IndexHierarchy.sort() now accept ascending as an iterable of Booleans to specify value per vector.
FrameGO.via_fill_value() now supports providng a fill value in __setitem__() assignment.
IndexAutoFactory can now be instantiated with a size parameter to pre-set the size of an auto-index, such as when used to initialize a FrameGO.
0.8.15
Added support for loading containers into specialized VisiData Sheet and IndexSheet subclasses; added to_visidata() exporter to all containers.
Added StyleConfig class for configuring display characteristics. Added default StyleConfigCSS for improved default HTML presentation.
Added Series.rank_ordinal, Series.rank_dense, Series.rank_mean, Series.rank_min, Series.rank_max.
Added Frame.rank_ordinal, Frame.rank_dense, Frame.rank_mean, Frame.rank_min, Frame.rank_max.
Fixed issue in Series.from_element() and Frame.from_element() that would broadcast some iterables instead of treat them as an element.
Extended Frame.unset_index() to support unsetting IndexHierarchy.
0.8.14
Added index_continuation_token and columns_continuation_token to Frame.from_delimited() and related methods.
Added via_re() interfaces to Index, IndexHierarchy, Series, Frame.
Updated arraykit to 0.1.8
0.8.13
Integration with arraykit; replacement of numerous utility methods with arraykit implementations.
Added via_fill_value() interface to Series and Frame.
0.8.12
Performance enhancements to Quilt.iter_series().apply(), Quilt.iter_tuple().apply(), Quilt.iter_array().apply().
0.8.11
Fixed issue when supplying dtype arguments to apply methods with string dtypes.
Added parameters argument to Frame.from_sql to perform SQL parameter substitution.
In group-by operations where the group key is a hashable, the returned Index.name will be set to that key.
Performance enhancements to Bus.iter_element().apply() and Bus.iter_element_items().apply()`.
0.8.10
Performance enhancements to Index initialization.
Performance enhancements to Series.iter_element().apply(), Series.iter_element().map_any(), Series.iter_element().map_all(), and Series.iter_element().map_fill().
Performance enhancements to Frame.iter_series().apply(), Frame.iter_tuple().apply(), Frame.iter_array().apply().
0.8.9
Performance enhancements to Series.dropna().
Series.relabel() and Frame.relabel() now raise if given a set or frozenset.
Fixed issue in Frame.assign.loc[] when using a Boolean array as a column selector.
0.8.8
Added Frame.cov(), Series.cov(), and Batch.cov().
Performance enhancements to loc selections by element.
0.8.7
Implemented support for multiprocessing Frame writing from StoreZip subclasses used by Bus, Batch, and Quilt.
Enabled write_max_workers and write_chunksize in StoreConfig.
Added py.typed file to package.
Improved exceptions raised when attempting to write to a file at an invalid path.
Improved handling of reading files with columns but no data with Frame.from_delimited.
0.8.6
Frame.rename now accepts optional arguments for index and columns renaming.
Series.rename now accepts an optional argument for index renaming.
Added Frame.relabel_shift_in() and Frame.relabel_shift_out().
Fixed issue where Frame.dropna() fails on single-columns Frame.
Extended IndexHierarchy.level_drop to perform corresponding drops on name when name is an appropriately sized tuple.
Extended Frame.set_index to support creating a 1D index of tuples when more than one column is selected.
0.8.5
Frame.from_sql now properly applies dtypes to columns used by index_depth selections.
Added Index.unique and IndexHierarchy.unique, both taking a depth_level specifier for selecting one or more depths.
Fixed issue with Frame.bloc selections that result in a zero-sized Series.
0.8.4
Refined Frame.bloc selections to reduce type coercion.
Improved Frame.assign.bloc when assigning with Series and Frame.
0.8.3
Added iloc_searchsorted() and loc_searchsorted() to Index, IndexDatetime, and IndexHierarchy.
Added ddof parameter to all containers that expose std and var.
Fixed issue with Frame.assign where there was a dependency on the order of column labels given in selection.
Improved handling for NumPy Boolean types stored in SQLite DBs via StoreSQLite interfaces.
Improved loc_to_iloc() methods to raise for missing keys in Index created where loc_is_iloc.
0.8.2
Added Series.iloc_searchsorted() and Series.loc_searchsorted().
Interfaces of Frame.to_delimited(), Frame.to_csv(), Frame.to_tsv(), and Frame.to_clipboard() are extended with parameters for control of quoting and escaping delimiters and other characters. The standard library’s csv module is now used for writing.
0.8.1
API change: Frame.from_element_loc_items() renamed Frame.from_element_items; Frame.from_element_iloc_items is removed.
Frame.assign now returns a FrameAssign instance with an apply method to permit using the assignment target, after function application, as the assignment value.
Series.assign now returns a SeriesAssign instance with an apply method to permit using the assignment target, after function application, as the assignment value.
IndexDatetime subclasses now properly assign name attrs from an Index given as an initializer.
Series.items() now returns labels of IndexHierarchy as tuples instead of np.ndarray.
Added Batch.apply_except and Batch.apply_items_except to permit ignore exceptions on function application to contained Frames.
Added Batch.unique().
Batch now supports operations on Frame that return an np.ndarray.
Added Quilt.from_items() and Quilt.from_frames().
Bus.sort_index() and Bus.sort_values() now return a Bus instance.
Improvements to Bus.items(), Bus.values for optimal Store reads when max_persist is None.
Implemented Bus.rename() to return a Bus instance.
Implemented Bus.drop[] to return a Bus instance.
Implemented Bus.reindex(), Bus.relabel(), Bus.relabel_flat(), Bus.relabel_level_add(), Bus.relabel_level_drop(), Bus.rehierarch().
Implemented Bus.roll(), Bus.shift().
0.8.0
API change: Frame.sort_values() now has a label positional argument that replaces the former key positional argument.
API change: Frame.sort_values() now requires multiple labels to be provided as a list to permit distinguishing selection of single tuple labels.
API change: iter_labels.apply() on Index and IndexHierarchy now returns an np.ndarray rather than a Series.
API change: iter_tuple and iter_tuple_items interfaces now require axis to be kwarg-only.
API change: iter_tuple, iter_tuple_items methods now require an explicit tuple as constructor if fields are invalid NamedTuple attrs.
API change: iter_array, iter_array_items, iter_series, and iter_series_items now require axis to be kwarg-only.
Added key argument for sort pre-processing to Frame.sort_values().
0.7.15
Added key argument for sort pre-processing to Index.sort(), IndexHierarchy.sort(), Series.sort_index(), Series.sort_values(), Frame.sort_index(), Frame.sort_columns
Implemented support for multiprocessing Frame loading from StoreZip subclasses used by Bus, Batch, and Quilt.
Added read_max_workers, read_chunksize, write_max_workers, write_chunksize to StoreConfig.
Added include_index_name, include_columns_name parameters to Frame.to_arrow
Added include_index_name, include_columns_name parameters to Frame.to_parquet
Added index_name_depth_level, columns_name_depth_level parameters to Frame.from_arrow
Added index_name_depth_level, columns_name_depth_level parameters to Frame.from_parquet
Fixed issue where non-optimal dtype would be used for new columns added in reindexing.
0.7.14
Added immutable, hashable containers SeriesHE and FrameHE.
Implemented read_many for all Store subclasses; Bus now uses these interfaces for significantly faster reads of multi-Frame selections.
Improved handling of connection object given to Frame.from_sql.
Improved type-preservation and performance when assigning Frame into Frame.
Added Bus.from_items() constructor.
0.7.13
Improved handling for using Frame.iter_group on zero-sized Frame.
Series can now be used as arguments to dtypes in Frame constructors.
Added via_dt.strptime and via_dt.strpdate for parsing strings to Python date, datetime objects, respectively.
0.7.12
Bus indices are no longer required to be string typed.
StoreConfig adds label_encoder, label_decoder parameters for translating hashables to strings and strings to hashables when writing to / from Store formats.
Frame.from_sql now supports a columns_select parameter.
StoreConfig now supports a columns_select parameter; columns_select parameters from StoreConfig are now used in StoreZipParquet, StoreSQLite.
Extended via_str.startswith() and via_str.endswith() functions to support passing an iterable of strings to match.
Improved IndexHierarchy.loc_to_iloc to support Boolean array selections.
0.7.11
Corrected issue in Frame.iter_series due to recent optimization.
0.7.10
Improvements to Quilt extraction routines.
0.7.9
Improved handling of invalid file paths given to constructors.
Improved implementations of Bus.items(), Bus.values, and Bus.equals() that deliver proper results when max_persist is active.
Implementation of Quilt, a container that presents the contents of a Bus as either vertically or horizontally stacked Frame.
Implemented __deepcopy__() on all containers.
0.7.8
Frame.iter_tuple_items() now exposes a constructor argument to control creation of axis containers.
Added Batch.apply_items.
Added Frame.count, Series.count, Batch.count.
Added Frame.sample, Series.sample, Index.sample, IndexHierarchy.sample, Batch.sample.
Added Frame.via_T and IndexHierarchy.via_T accessors for opposite axis binary operator application of 1D operands.
0.7.7
IndexHierarchy.iter_label now defaults to iterating full depth labels.
Batch.__repr__() is no longer a display that exhausts the stored generator.
Frame.iter_tuple() now exposes a constructor argument to control creation of axis containers.
0.7.6
Fixed issue in using Frame.extend with zero-length Frame.
0.7.5
Implemented Frame.isin on TypeBlocks.
Implemented Frame.clip on TypeBlocks.
0.7.4
Series.from_element now works correctly with tuples
Batch element handling now avoids diagonal formations; Batch.apply() now handles elements correctly
dtypes parameters can now be provided with dict_values instances.
Frame.to_parquet, Frame.to_arrow now convert datetime64 units to nanosecond if not supported by PyArrow.
0.7.3
Bus now exposes max_persist parameter to define the maximum number of loaded Frame retained by the Bus.
Added len() to via_str interfaces.
Frame.iter_element now takes an axis argument to determine element order, where 0 is row major, 1 is column major.
Silenced NaturalNameWarning via tables in StoreHDF5.
StoreSQLite will now re-write, rather than update, a file path where an SQLite DB already exists.
Improved handling for iterating zero-sized Frame.
Improved type detection when performing operations on Frame.iter_element iterators.
Frame.shift() file_value parameter is now key-word argument only.
Frame.roll() include_index, include_columns is now key-word argument only.
0.7.2
Extended application of binary equality operators to permit comparison with arrays of single elements.
0.7.1
Refined application of binary equality operators to permit comparison with strings or elements that are not sequences.
0.7.0
API change: __bool__ of all containers now raises a ValueError.
API change: IndexHierarchy.iter_label now iterates over realized labels.
API change: IndexBase.union, IndexBase.intersection no longer automatically unpack values from ContainerOperand subclasses.
API change: Container operands used with binary equality operators will raise if sizes are not equivalent.
API change: Frame.from_xlsx, as well as StoreConfig now set trim_nadir to False by default.
API change: Series.relabel_add_level to Series.relabel_level_add, Series.relabel_drop_level to Series.relabel_level_drop, Frame.relabel_add_level to Frame.relabel_level_add, Frame.relabel_drop_level to Frame.relabel_level_drop, Index.add_level to Index.level_add, IndexHierarchy.add_level to IndexHierarchy.level_add, IndexHierarchy.drop_level to IndexHierarchy.level_drop.
0.6.38
Frame.dtype interface now takes TDtypesSpecifier, permitting setting dtype by mapping, iterable, or single value.
dtypes can be given as a single TDtypeSpecifier for specifying dtype of all columns.
Series of Frame can now be created without specifying dtype arguments.
Frame now supports usage as a weakref.
Frame.from_parquet now raises when columns_select names columns not found in the file.
0.6.37
Fixed issue in implementation of trim_nadir when reading XLSX files.
0.6.36
Fixed issue in Frame.from_pandas when the columns have mixed types including integers.
Improved dtype preservation in zero-sized Series extraction from Frame.
Added trim_nadir parameter to StoreConfig and Frame.from_xlsx: permits removing all-None trailing rows and columns resulting from XLSX styles being applied to empty cells.
0.6.35
Added a name parameter to Series.from_pandas and Frame.from_pandas.
Added Frame.from_msgpack and Frame.to_msgpack.
Refactored Bus and Batch to use the mixin class StoreClientMixin to share exporters and constructors.
Added StoreClientMixin.to_zip_parquet and StoreClientMixin.from_zip_parquet.
Performance improvements to Frame.to_pandas when a Frame has unified TypeBlocks.
0.6.34
Updated all delimited text output formats to include a final line termination.
Frame.from_overlay now takes optional index and columns arguments; Series.from_overlay now takes an optional index argument.
Improvements to union/intersection index formation in Frame.from_overlay and Series.from_overlay.
0.6.33
Performance improvements to Frame.pivot.
Frame.from_xlsx now exposes skip_header and skip_footer parameters.
0.6.32
Added Frame.from_overlay, Series.from_overlay constructors.
Added support for dataclass as records in Frame.from_records and Frame.from_records_items.
Additional delegated Frame methods added to Batch.
0.6.31
Fixed issue when loading pickled containers where Boolean selection would not be properly identified.
0.6.30
Added via_dt.fromisoformat() to all containers, supporting creation of date/datetime objects from ISO 8601 strings.
Batch.to_frame now returns a Frame with an IndexHierarchy if all Batch operations retain one or more Frame.
Batch interface extended with core Frame methods.
Restored parameter name in Series.relabel to be index.
Support for writing date, datetime, and np.datetime64 via Frame.to_xlsx.
Exposed store_filter parameter in Frame.from_xlsx,``Frame.to_xlsx``.
Removed format_index, format_columns attributes from StoreConfig.
0.6.29
Fixed issue in Series.drop when the Series has an IndexHierarchy.
Calling Frame.from_series with something other than a Series will now raise.
Calling Index.from_pandas, Series.from_pandas, and Frame.from_pandas now raise when given a non-Pandas object.
StoreConfig given to Bus.to_xlsx, Bus.to_sqlite, and Bus.to_hdf5 are now properly used.
0.6.28
Introduced the Batch, a lazy, parallel processor of groups of Frame.
Index and IndexHierarchy intersection() and union() now accept *args, performing the set operation iteratively on all arguments.
Revised default aggregation function to Frame.pivot.
Fixed issue in writing SQLite stores from Frame labelled with strings containing hyphens.
Added include_index_name, include_columns_name to Frame.to_delimited.
Added include_index_name, include_columns_name to StoreConfig and Frame.to_xlsx interfaces.
Added index_name_depth_level and columns_name_depth_level to Frame.from_delimited and related methods.
Added index_name_depth_level, columns_name_depth_level to StoreConfig and Frame.from_xlsx interfaces.
0.6.27
Improved implementation of Frame.pivot.
0.6.26
Removed class-level documentation injection, permitting better static analysis.
Corrected issue in appending tuples to an empty IndexGO.
0.6.25
Added Frame.from_clipboard() and Frame.to_clipboard().
Added Frame.pivot_stack() and Frame.pivot_unstack().
0.6.24
Fixed flaw in difference operations on IndexDatetime subclasses of equivalent indices.
0.6.23
Frame.from_parquet and Frame.from_arrow now accept a dtypes argument.
All PathLike path objects now accepted wherever Path objects were previously.
Added fillna methods to Index, IndexHierarchy.
Added to StoreFilter the following parameters: value_format_float_positional, value_format_float_scientific, value_format_complex_positional, value_format_complex_scientific.
Index and IndexHierarchy will reuse instances for set operations on equivalent indices.
Added IndexHierarchy.from_names constructor for creating zero-length IndexHierarchy.
Refinements to IndexHierarchy to support grow-only mutation from zero length.
0.6.22
Fixed flaw in IndexLevel for handling of zero-length levels.
Fixed flaw in TypeBlocks.iloc that caused an undesirable reference cycle.
0.6.21
IndexHierarchy set operations will now delegate Index types when they are equivalent between operands at corresponding depth levels.
Frame.from_concat now delegates returned index input index name, type, IndexHierarchy contained types, if aligned on all indices per axis.
Fixed issue when calling relabel_add_level() from a FrameGO.
0.6.20
Extended functionality of HLoc selections in IndexHierarchy to properly handle selection lists, Boolean arrays, and nested ILoc selections.
Corrected issue in Frame.from_concat whereby, when given inputs with IndexHierarchy, IndexHierarchy were not returned.
0.6.19
Extended name propagation to applications of binary operators where an operand is a scalar.
Binary operators now work with Frame and same-shaped NumPy arrays.
0.6.18
Extended support for step arguments in loc interfaces.
Implemented Frame.join_left, Frame.join_right, Frame.join_inner, and Frame.join_outer.
Implemented Frame.insert_before, Frame.insert_after.
Implemented Series.insert_before, Series.insert_after.
IndexHierarchy.from_labels now enforces all labels to have the same depth.
Fixed issue where, when passing an array to Frame.from_records, the name parameter is not passed to the constructor.
0.6.17
Implemented equals() methods on all containers.
Added defensive check against assigning a Pandas Series to a FrameGO as an unlabeled iterator.
Added proper handling of types multiple-inherited from str (or any other type) and Enum.
Implemented support for operator overloading of addition and multiplication on string dtypes.
0.6.16
Implemented via_str and via_dt accesors on all ContainerOperand.
When writing to XLSX, the shape of the Frame is validated to fit within the limits of XLSX sheets.
0.6.15
Added support for round() on Frame.
Added name parameter to all methods of IterNodeDelegate that produce a new container, including map_any(), map_fill(), map_all(), apply(), and apply_pool().
Support for include_index and include_columns in DisplayConfig instances and Display output.
Performance improvements to iterating tuples from IndexHierarchy.
Performance improvements for IndexHierarchy transformations, including adding or dropping levels and rehierarch.
0.6.14
Added explicit handling for binary operators applied to differently-sized IndexHierarchy.
0.6.13
Refined behavior of Frame.from_concat_items when given tuples as labels; implemented support for tuples as labels in IndexLevels.values_at_depth.
0.6.12
Refined behavior of names attribute on IndexBase to ensure that an appropriately sized iterable of labels is always returned.
0.6.11
Added IndexHour and IndexHourGO indices.
Added IndexMicrosecond and IndexMicrosecondGO indices.
Added support for round() on Series.
Index.astype now returns specialized datetime64 Index objects when given an appropriate dtype.
IndexHierarchy.astype now produces an IndexHierarchy with specialized datetime64 Index objects when given an appropriate dtype.
Added IndexLevels.dtypes_at_depth() and IndexLevels.dtype_per_depth() to capture resolved dtypes per depth.
Added IndexLevels.values_at_depth() to capture resolved typed arrays per depth.
Updated IndexHierarchy.display() to display proper types per depth.
Refactored IndexLevel to lazily cache depth and length attributes.
Refactored IndexHierarchy to store a TypeBlocks instance instead of 2D array, permitting reuse of TypeBlocks functionality, columnar type preservation, and immutable array reuse.
Fixed flaw in IndexHierarchy.label_widths_at_depth.
Fixed flaw in Frame.from_records and related routines whereby a NamedTuple in an iterable of length 1 was converted to a single-row, two-dimensional array.
Fixed flaw in Frame function application on iterators for some Index type configurations.
API documentation now shows full signatures for all functions.
0.6.10
Improvements to interface display, including in inclusion of function arguments and new “Assignment” category; improvements to API documentation.
Fixed issue in not handling mismatched size between index and values on Series initialization.
Fixed issue creating a datetime64 Index from another datetime64 Index when their dtypes differ.
Fixed an issue when passing an immutable Index as columns in FrameGO.reindex.
0.6.9
Series default constructor now efficiently handles Series given as values.
Frame default constructor now efficiently handles Frame given as data.
AutoMap now serves as the core mapping structure for all Index object, offering better performance, immutability, and internal uniqueness checks.
0.6.8
Fixed issue in using relabel() on columns in FrameGO.
Fixed issue in using Frame.drop with IndexHierarchy on either axis.
Unified to_frame and to_frame_go interfaces on Frame, FrameGO, and IndexHierarchy.
Enabled include_index, include_columns parameters for Frame.to_parquet.
Added columns_select parameter to Frame.from_parquet.
Updated requirements: pyarrow==0.16.0
Refined Frame.from_arrow usage of ChunkedArray, disabling date_as_object, enabling self_destruct, and improving handling of NumPy array extraction.
Added STATIC attribute to ContainerBase and all subclasses.
0.6.7
Fixed issue in assigning a column to a FrameGO from a generator that raises an exception.
0.6.6
Added difference method to all Index subclasses.
Added index_constructor and columns_constructor parameters to Frame.from_pandas; index_constructor added to Series.from_pandas.
0.6.5
Refined IndexBase.from_pandas.
0.6.4
Fixed issue introduced into Frame.iter_group and Frame.iter_group_items when selecting a single column with an object dytpe.
Fixed mapping lookups to use single-argument tuples in map_any_iter_items and map_fill_iter_items and related methods.
0.6.3
Improvements to any and all methods on all containers when using skipna=True and NAs are presernt; now, a TypeError will now be raised when NAs are found and skipna=False.
When converting from Pandas 1.0 extension dtypes, proper NumPy types are used if no pd.NA are present; if pd.NA are present, they are replaced with np.nan in the resulting object array.
0.6.2
Frame.sort_values now accepts multiple labels given as any iterable.
loc selection on Series or Frame with IndexAutoFactory-style indices now treat the slice stop as inclusive.
Removed creation of internal mapping object for IndexAutoFactory indices, or where Index are created where loc_is_iloc.
Improved induction of dtype for labels array stored in Index.
0.6.1
The bloc and assign.bloc selectors on Frame now use [] instead of (), aligning the interface with other selectors.
Added IndexNanosecond and IndexNanosecondGO indices.
All iter_* interfaces now explictly define arguments.
Frame.fillna() and Series.fillna() now accept Frame and Series, respectively, as arguments.
Series.sort_index, Series.sort_values, Frame.sort_index, Frame.sort_columns, and Frame.sort_values now retain index/columns name after sorting.
Renamed Series.iter_group_index(), Series.iter_group_index_items(), Frame.iter_group_index(), Frame.iter_group_index_items() to Series.iter_group_labels(), Series.iter_group_labels_items(), Frame.iter_group_labels(), Frame.iter_group_labels_items()
Fixed issue in Frame display where, when at or one less than the count of display_rows, would display different numbers of rows for the Index and the body of the Frame.
Zero-sized Frame now return zero-sized Series from selection where possible.
0.6.0
Removed deprecated Frame and Series non-specialized constructor usage; removed support for providing mapping types to apply.
Improved support for using tuples in Frame.__getitem__ and FrameGO.__setitem__ with IndexHierarchy and Index with tuple labels.
0.5.13
Made Frame.clip, Frame.duplicated, Frame.drop_duplicated key-word argument only. Made Series.clip, Series.duplicated, Series.drop_duplicated key-word argument only.
Frame.iter_series now sets the name attribute of the Series from the appropriate index.
Added Index.head(), Index.tail(), IndexHierarchy.head(), IndexHierarchy.tail().
Frame.from_records and related routines now do full type induction per column; all type induction on untyped iterables now examines all values.
0.5.12
All Index subclasses now use PositionsAllocator to share immutable positions arrays, increasing Index performance.
Fixed issue in using FrameGO.relabel with a non grow-only IndexBase.
IndexHiearchy.from_labels now accepts a reorder_for_hierarchy Boolean option to reorder labels for hierarchical formation.
FrameGO.from_xlsx, FrameGO.from_hdf5, FrameGO.from_sqlite now return the FrameGO instances. Updated all Store.read methods to accept a containter_type arguement.
Added consolidate_blocks parameter to StoreConfig.
Added consolidate_blocks parameter to Frame.from_xlsx, Frame.from_hdf5, Frame.from_sqlite, Frame.from_pandas.
Implemented IndexYearGO, IndexYearMonthGO, IndexDateGO, IndexMinuteGO, IndexSecondGO, IndexMillisecondGO grow-only, derived classes of datetime64 indices.
Added Frame constructors: Frame.from_series, Frame.from_element, Frame.from_elements. Deprecated creating Frame from an untyped iterable or element.
Added Series constructors: Series.from_element. Deprecated creating Series from an element with the default intializer.
Added index_constructor, columns_constructor arguement to Frame.from_items, Frame.from_dict.
NP-style methods on Series and Frame no longer accept arbitrary keywork arguments.
Removed keys() and items() methods from Index and IndexHierarch; default iterators from IndexHierarchy now iterate tuples instead of arrays.
Added to IterNodeDelegate the following methods for applying mapping types to iterators: map_all, map_any, and map_fill. Generator versions are also made available: map_all_iter, map_all_iter_items, map_any_iter, map_any_iter_items, map_fill_iter, map_fill_iter_items.
0.5.11
Fixed issue in Frame.assign when assigning iterables into a single column.
0.5.10
Improvements to Frame.assign to handle unordered column selectors and preserve columnar types not affected by assignment.
Restored application of default column and index formattng in StoreXLSX.
0.5.9
Fixed issue in __slots__ usage of derived Containers.
Implemented StoreConfig and StoreConfigMap classes, and updated all Store and Bus interfaces to use them.
Implemented tracking of Store file modification times, and implemented raising exceptions for any unexpected file modifications.
Improved handling of reading XLSX files with trailing all-empty rows resulting from style formatting across empty data.
Improved HDF5 reading so as to reduce memory overhead.
0.5.8
Fixed issue in Frame.sort_values() when axis=0 and underlying block structure is homogenous.
Improved performance of Frame.iter_group and related methods.
Fixed issue raised when calling built-in help() on SF containers.
Improved passing of index names in IndexHierarchy.to_pandas.
Improved propagation of name in methods of Index and IndexHierarchy.
0.5.7
StoreFilter added to the public namespace.
names argument added to Frame.unset_index.
Improved handling of ILoc usage within loc calls.
Improved input and output from/to XLSX.
0.5.6
Frame.from_concat, Series.from_concat now accept empty iterables.
Frame.iter_group.apply and related routines now handle producing a Series from a multi-column group selection.
0.5.5
Index objects based on datetime64 now accept Python datetime.date objects in loc expressions.
Fixed index formation when using apply on Frame.iter_group and Frame.iter_group_items (and related interfaces) when the Frame has an IndexHierarchy.
Fixed issue in a Frame.to_frame_go() not creating a fully decoupled Index for columns in the returned Frame.
0.5.4
Index objects based on datetime64 now return empty Series when a partial loc selection does not match any values found in the Index.
0.5.3
Frame.set_index_hiearchy passes on name to returned Frame.
Index objects based on datetime64 now accept Python datetime.datetime objects in loc expressions.
Exposed interface attribute on ContainerBase subclasses.
0.5.2
Refinements to Series.isin(), Frame.isin(), Index.isin(), and IndexHierarchy.isin() to better identify cases of unique elements.
Added IndexMinute datetime index subclass.
0.5.1
Implemented handling in Frame.from_delimited for column-only files.
Frame.iter_tuple and Frame.iter_tuple_items will return tuple instead of NamedTuple if fields are not valid identifiers.
Frame.from_records now supports empty records if columns is provided.
Frame.from_concat now implements better type preservation in vertical concatenation of arrays.
0.5.0
Introduced the Bus, a Series-like container of mulitple Frame, supporting lazily reading from and writing to XLSX, SQLite, and HDF5 data stores, as well as zipped pickles and delimited files.
Added interface attribute to all containers, providing a hierarchical presentation of all interfaces.
Added display_tall() and display_wide() convenience methods to all containers.
Added label_widths_at_depth() on Index and IndexHierarchy.
Added Series.from_concat_items() and Frame.from_concat_items().
Added Frame.to_xarray().
Added Frame.to_xlsx(), Frame.from_xlsx().
Added Frame.to_sqlite(), Frame.from_sqlite().
Added Frame.to_hdf5(), Frame.from_hdf5().
Added Frame.to_rst().
Added Frame.to_markdown().
Added Frame.to_latex().
The interface of Frame.from_delimited (as well as Frame.from_csv and Frame.from_tsv) has been updated to conform to the common usage of index_depth and columns_depth. IndexHierarchy is now supported when index_depth or columns_depth is greater than one. The former parameter index_column is renamed index_column_first.
Added IndexHierarchy.from_index_items and IndexHierarchy.from_labels_delimited.
Added IndexBase.names attribute to provide normalized names equal in length to depth.
The DisplayConfig parameter type_show now, if False, hides, native class types used as headers. This is the default display for all specialized string output via Frame.to_html, Frame.to_rst, Frame.to_markdown, Frame.to_latex, as well as Jupyter display methods.
Added Frame.unset_index().
Added Frame.pivot().
Added Frame.iter_window, Frame.iter_window_items, Frame.iter_window_array, Frame.iter_window_array_items.
Added Series.iter_window, Series.iter_window_items, Series.iter_window_array, Series.iter_window_array_items.
Added Frame.bloc and Frmae.assign.bloc
Added IndexHierarchy.rehierarch, Series.rehierarch, and Frame.rehierarch.
Defined __bool__ for all containers, where the result is determined based on if the underlying NumPy array has size greater than zero.
Improved Frame.to_pandas() to preserve columnar types.
Frame.set_index_hierarchy now accepts a reorder_for_hierarchy argument, reordering the rows to support hierarchability.
Added Frame.from_dict_records and Frame.from_dict_records_items; when given records, the union of all keys is used to derive columns.
0.4.3
Fixed issues in FrameGO setitem and using binary operators between Frame and FrameGO.
0.4.2
Corrected flaw in axis 1 statistical operations with Frame constructed from mixed sized TypeBlocks.
Added Series.loc_min, Series.loc_max, Series.iloc_min, Series.iloc_max.
Added Frame.loc_min, Frame.loc_max, Frame.iloc_min, Frame.iloc_max,
0.4.1
iter_element().apply now properly preserves index and column types.
Using Frame.from_records with an empty iterable or iterator will deliver a ErrorInitFrame.
Matrix multiplication implemented for Index, Series, and Frame.
Added Frame.from_records_items constructor.
Improved dtype selection in FrameGO set item and related functions.
IndexHierarchy.from_labels now accepts an index_constructors argument.
Frame.set_index_hierarchy now accepts an index_constructors argument.
IndexHierarhcy.from_product() now attempts to use ``name of provided indicies for the IndexHierarchy name, when all names are non-None.
Added IndexHierarchy.dtypes and IndexHierarchy.index_types, returning Series indexed by name when possible.
0.4.0
Improved handling for special cases Series initialization, including initialization from iterables of lists.
The Series initializer no longer accepts dictionaries; Series.from_dict is added for explicit creation from mappings.
IndexAutoFactory suport removed from Series.reindex and Frame.reindex and added to Series.relabel and Frame.relabel.
The following Series and Frame methods are renamed: reindex_flat, reindex_add_level, and reindex_drop_level are now relabel_flat, relabel_add_level, and relabel_level_drop.
Implemented Frame.from_sql constructor.
0.3.9
IndexAutoFactory introduced to consolidate creation of auto-incremented integer indices, and provide a single token to force auto-incremented integer indices in other contexts where index arguments are taken.
IndexAutoFactory support implemented for the index argument in Series.from_concat and Series.reindex.
IndexAutoFactory support implemented for the index and columns argument in Frame.from_concat and Frame.reindex.
Added new DisplyaConfig parameters to format floating-point values: value_format_float_positional, value_format_float_scientific, value_format_complex_positional, value_format_complex_scientific,
Set default value_format_float_scientific and value_format_complex_scientific to avoid truncation of scientific notation in output displays.
0.3.8
All duplicate-handling functions now support heterogenously typed object arrays with unsortable (but hashable) types.
Operations on all indices now preserve order when indices are equal.
Functions with the skipna argument now properly skip None in Frames with built with object arrays.
Frame.to_csv now uses the argument name delimiter instead of sep, aligning with the usage in Frame.from_csv.
0.3.7
Completed implementation of Frame.fillna_forward, Frame.fillna_backward, Frame.fillna_leading, Frame.fillna_trailing.
Fixed issue exposed in FrameGO.sort_values() due to NumPy integers being used for selection.
IndexHierarchy.sort(), IndexHierarchy.isin(), IndexHierarchy.roll() now implemented.
Series.sort_index() now properly propagates IndexBase subclasses.
Frame.sort_index() and Frame.sort_columns() now properly propagate IndexBase subclasses.
All containers now derive from ContainerOperand, simplyfying inheritance and ContainerOperandMeta application.
Index objects based on datetime64 now accept datetime64 objects in loc expressions.
All construction from Python iterables now better handle array creation from diverse Python objects.
0.3.6
Frame.to_frame_go now properly handles IndexHierarchy columns.
Improved creation of IndexHierarchy from other IndexHierarchy or IndexHierarchyGO.
Frame initializer now exposes index_constructor and columns_constructor arguments.
Frame.from_records now efficiently uses dict_view objects containing row records.
Frame now supports shapes of all zero and non-zero combinations of index and column lengths; Frame construction will raise an exception if attempting to set a value in an unfillable Frame shape.
Frame, Series, Index, and IndexHierarchy all have improved implementations of cumprod and cumsum methods.
0.3.5
Improved type handling of datetime64 typed columns in Frame.
Added median method to all MetaOperatorDelegate classes, inlcuding Series, Index, and Frame.
Frame and Series sort methods now propagate name attributes.
Index.from_pandas() now correctly collects name / names attributes from Pandas indexes.
Implemented Series.fillna_forward, Series.fillna_backward, Series.fillna_leading, Series.fillna_trailing.
Fixed flaw in dropping columns from a Frame (via Frame.set_index or the Frame.drop interface), whereby sometimes (depending on TypeBlocks structure) the drop would not be executed.
Index objects based on datetime64 now limit __init__ arguments only to those relevant for those derived classes.
Index objects based on datetime64 now support transformations from both datetime.timedelta as well as np.timedelta64.
Index objects based on datetime64 now support selection with slices with datetime64 units different than those used in the Index.
0.3.4
Added dtypes argument to all relevant Frame constructors; dtypes can now be specified with a dictionary.
Deprecated instantiating a Frame from dict; added Frame.from_dict for explicit Frame creation from a dict.
0.3.3
Improvements to all datetime64 based indicies: direct creation from labels now properly parses values into datetime64, and loc-style lookups now handle partial matches on lower-resolution datetimes. Added IndexSecond and IndexMillisecond Index classes.
Index can now be constructed directly from an IndexHierarchy (resulting in an Index of tuples)
Improvements to application of ellipsis when normalizing width in Display string representations.
Frame.values now always returns a 2D NumPy array.
Series.iloc, when a non-mulitple selection is given, now returns a single element, not a Series.
0.3.2
IndexHierarchy.level_drop() and related methods have been updated such that negative integers drop innermost levels, and postive integers drop outermost levels. This is an API breaking change.
Fixed missing handling for all-missing in Series.dropna.
Improved loc and HLoc usage on Series with IndexHierarchy to insure a Series is returned when a multiple selection is used.
IndexHierarchy.from_labels() now returns proper error message for invalid tree forms.
0.3.1
Implemented Series.iter_group_index(), Series.iter_group_index_items(), Frame.iter_group_index(), Frame.iter_group_index_items() for producing iterators (and targets of function application) based on groupings of the index; particularly useful for IndexHierarhcy.
Implemented Series.from_concat; improved Frame.from_concat in concatenating indices with diverse types. Frame.from_concat() now accepts Series.
Added Index.iter_label() and IndexHierarchy.iter_label(), for variable depth label iteration, particularly useful for IndexHierarchy.
Improved initializer behavior of IndexDate, IndexYearMonth, IndexYear to apply expected dtype when creating arrays from non-array initializers, allowing conversion of string date representations to proper date types.
Added Index.to_pandas and specialized methods on IndexDate and derived classes. Added IndexHierarchy.to_pandas.
Added support for Series as an argument to FrameGO.extend().
Added Series.to_frame() and Series.to_frame_go().
The name attribute is now implemented for all containers; all constructors now take a name argument, and a rename method is available. Extracting columns, rows, and setting indices on Frame all propagate name attributes appropriately.
The default Series display has been updated to show the “<Series>” label above the index, consistent with the presentation of Frame.
The Frame.from_records() method has been extended to support explicitly passing dtypes per column, which permits avoiding type discovery through observing the first record or relying on NumPy’s type discovery in array creation.
The Frame.from_concat() constructor now handles hierarchical indices correctly.
0.3.0
The Index.keys() method now returns the underlying KeysView from the Index’s dictionary.
All primary containers (i.e., Series, Frame, and Index) now display HTML tables in Jupyter Notebooks. This is implemented via the _repr_html_() methods.
All primary containers now feature a to_html() method.
All primary containers now feature a to_html_datatables() method, which authors a complete HTML file with DataTables/JavaScript-powered table viewing, sorting, and searching.
StaticFrame’s display infrastructure now permits individually coloring types by category, as well as different display formats for supporting HTML output.
StaticFrame’s display infrastructure now shows hierarchical indices, used for either indices or columns, in the same display grid used for other display components.
The DisplayConfig class has been expanded to permit definition of colors, specified in hexadecimal integers or string codes, for all type categories, as well as independent settings for type delimiters, and a new setting for display_format.
The following DisplayFormats have been created and implemented: terminal, html_datatables, html_table, and html_pre.