Version 0.11.0

• Breaking Change: Python 2 support has been dropped. This is to align
  with the pandas package which dropped Python 2 support at the end of 2019.
  (#268)

Enhancements

• Ensure table_schema argument is not modified inplace. (:issue:278)

Implementation changes

• Use object dtype for STRING, ARRAY, and STRUCT columns when
  there are zero rows. (#285)

Internal changes

• Populate user-agent with pandas version information. (#281)
• Fix pytest.raises usage for latest pytest. Fix warnings in tests.
  (#282 )
• Update CI to install nightly packages in the conda tests. (#254)

Version 0.10.0

Documentation

• Document BigQuery data type to pandas dtype conversion for read_gbq. ( #269 )

Dependency updates

• Update the minimum version of google-cloud-bigquery to 1.9.0. ( #247 )
• Update the minimum version of pandas to 0.19.0. ( #262 )

Internal changes

• Update the authentication credentials. Note: You may need to set reauth=True in order to update your credentials to the most recent version. This is required to use new functionality such as the BigQuery Storage API. ( #267 )
• Use to_dataframe() from google-cloud-bigquery in the read_gbq() function. ( #247 )

Enhancements

• Fix a bug where pandas-gbq could not upload an empty DataFrame. ( #237 )
• Allow table_schema in to_gbq to contain only a subset of columns, with the rest being populated using the DataFrame dtypes ( #218 ) (contributed by @JohnPaton)
• Read project_id in to_gbq from provided credentials if available (contributed by @daureg)
• read_gbq uses the timezone-aware DatetimeTZDtype(unit='ns', tz='UTC') dtype for BigQuery TIMESTAMP columns. ( #269 )
• Add use_bqstorage_api to read_gbq. The BigQuery Storage API can be used to download large query results (>125 MB) more quickly. If the BQ Storage API can't be used, the BigQuery API is used instead. ( #133, #270 )

Version 0.9.0

• Warn when deprecated private_key parameter is used. #240
• New dependency Use the pydata-google-auth package for authentication. #241

PyPI

Version 0.8.0

Breaking changes

• Deprecate private_key parameter to pandas_gbq.read_gbq and pandas_gbq.to_gbq in favor of new credentials argument. Instead, create a credentials object using google.oauth2.service_account.Credentials.from_service_account_info or google.oauth2.service_account.Credentials.from_service_account_file. See the authentication how-to guide for examples. (#161, #231 )

Enhancements

• Allow newlines in data passed to to_gbq. (#180)
• Add pandas_gbq.context.dialect to allow overriding the default SQL syntax dialect. (#195, #235)
• Support Python 3.7. (#197, #232)

Internal changes

• Migrate tests to CircleCI. (#228, #232)

Version 0.7.0

• int columns which contain NULL are now cast to float, rather than object type. (#174)
• DATE, DATETIME and TIMESTAMP columns are now parsed as pandas' timestamp objects (#224)
• Add :class:pandas_gbq.Context to cache credentials in-memory, across calls to read_gbq and to_gbq. (#198, #208)
• Fast queries now do not log above DEBUG level. (#204) With BigQuery's release of clustering querying smaller samples of data is now faster and cheaper.
• Don't load credentials from disk if reauth is True. (#212) This fixes a bug where pandas-gbq could not refresh credentials if the cached credentials were invalid, revoked, or expired, even when reauth=True.
• Catch RefreshError when trying credentials. (#226)

Version 0.6.1

• Improved read_gbq performance and memory consumption by delegating DataFrame construction to the Pandas library, radically reducing the number of loops that execute in python (#128)
• Reduced verbosity of logging from read_gbq, particularly for short queries. (#201)
• Avoid SELECT 1 query when running to_gbq. (#202)

Version 0.6.0

• Warn when dialect is not passed in to read_gbq. The default dialect
  will be changing from 'legacy' to 'standard' in a future version.
  (#195 )
• Use general float with 15 decimal digit precision when writing to local
  CSV buffer in to_gbq. This prevents numerical overflow in certain
  edge cases. (#192)

Version 0.5.0

• Project ID parameter is optional in read_gbq and to_gbq when it can inferred from the environment. Note: you must still pass in a project ID when using user-based authentication. (#103)
• Progress bar added for to_gbq, through an optional library tqdm as dependency. (#162)
• Add location parameter to read_gbq and to_gbq so that pandas-gbq can work with datasets in the Tokyo region. (#177)

Version 0.4.1

PyPI release

• Only show verbose deprecation warning if Pandas version does not populate it. #157

Version 0.4.0

PyPI release, Conda Forge release

• Fix bug in read_gbq when building a dataframe with integer columns on Windows. Explicitly use 64bit integers when converting from BQ types. (#119)
• Fix bug in read_gbq when querying for an array of floats (#123)
• Fix bug in read_gbq with configuration argument. Updates read_gbq to account for breaking change in the way google-cloud-python version 0.32.0+ handles query configuration API representation. (#152)
• Fix bug in to_gbq where seconds were discarded in timestamp columns. (#148)
• Fix bug in to_gbq when supplying a user-defined schema (#150)
• Deprecate the verbose parameter in read_gbq and to_gbq. Messages use the logging module instead of printing progress directly to standard output. (#12)