Release v1.4.2

• Changed: exclude tests/ and docs/ folders from the Python package
• Fixed: ensure pandas>=3.0.3 works with audformat

Release v1.4.1

• Added: include table ID in error messages in the methods
  drop_index(),
  extend_index(),
  pick_index(),
  update()
  of audformat.Table
  and audformat.MiscTable
  if the receiving table is assigned to a database
• Changed: black list pandas
  versions 3.0.0, 3.0.1, and 3.0.2
  as those contain a serious issue in pd.to_timedelta()
  (pandas#65150 <https://github.com/pandas-dev/pandas/issues/65150>_)
• Fixed: avoid warning when loading a table
  with data points not matching scheme labels.
  As before those points are set to None.
  This is now documented in audformat.Table.load()

Release v1.4.0

• Added: audformat.Scheme.set_labels()
  to add labels to an existing scheme
  that has none.
  To replace labels
  use audformat.Scheme.replace_labels()
• Added: store audformat version in table files.
  For parquet files it is stored under the audformat-version metadata key.
  Pickle files store now a dictionary with the keys
  audformat-version and df,
  whereas df contains the table data
• Fixed: audformat.utils.hash() for large dataframes.
  Before it resulted in the same hash
  for large dataframes
  that differed only in rows in their center.
  This is a breaking change
  and will result in different hash values.
  When using audb.publish()
  to update a database
  not published with audformat>=1.4.0,
  it will result in republishing all existing tables,
  which is the desired behavior
• Fixed: loading scheme labels from misc table,
  when using audformat.Database.load(..., load_data=True)

Release v1.3.4

• Added: support for pandas >=3.0

Release v1.3.3

• Added: support for Python 3.14
• Added: document support for (misc) tables with zero columns
  in the specification
• Fixed: ensure cache compatibility between different pandas versions
• Removed: support for Python 3.9

Release v1.3.2

• Added: support for Python 3.13
• Added: support for Python 3.12
• Fixed: audformat.Database.update() for misc tables
• Fixed: audformat.Database.get()
  if scheme is stored in a segmented table,
  and additional_schemes are stored in filewise tables.
  Before,
  values of the additional schemes were set to <NA>
• Fixed: if audformat.Column.set() assigns values
  to a column with scheme labels
  given by a misc table,
  the misc table is only converted once to a dictionary
  instead of once for each value

Release v1.3.1

• Changed: replace unmaintained iso-639 dependency
  with iso639-lang
• Fixed: ensure poetry can manage audformat

Release v1.3.0

• Added: strict argument
  to audformat.utils.hash().
  If set to True,
  the order of the data,
  and its level/column names
  are taken into account
  when calculating the hash
• Changed: store tables per default as parquet files,
  by changing the default value of storage_format
  to "parquet"
  in audformat.Table.save()
  and audformat.Database.save()
• Fixed: load csv tables with pandas.read_csv(),
  if pyarrow.csv.read_csv() fails

Release v1.2.0

• Added: expand format specifications
  to allow parquet files
  as table files
• Added: support for storing tables as parquet files
  by adding "parquet"
  (audformat.define.TableStorageFormat.PARQUET)
  as an option
  for the storage_format argument
  of audformat.Table.save()
  and audformat.Database.save()
• Added: support for numpy>=2.0
• Added: mention text files
  as potential media files
  in the documentation
• Added: mention in the documentation of audformat.utils.hash()
  that column/level names do not influence its hash value
• Added: warn in the documentation of audformat.utils.hash()
  that the hash of a dataframe or series,
  containing "Int64" as data type,
  changes with pandas>=2.2.0
• Fixed: ensure "boolean" data type
  is always used
  in indices of misc tables
  that store boolean values

Release v1.1.4

• Fixed: audformat.Database.get(),
  if its argument additional_schemes
  contains a non-existent scheme