bigframes

Features

• Add bigframes.bigquery.sql_scalar() to apply SQL syntax on Series objects (#1293) (aa2f73a)
• Add unix_seconds, unix_millis and unix_micros for timestamp series. (#1297) (e4b0c8d)
• DataFrame.join supports Series other (#1303) (ee37a0a)
• Support array output in remote_function (#1057) (bdee173)

Bug Fixes

• Dataframe sort_values Series input keyerror. (#1285) (5a2731b)
• Fix read_gbq_function issue in dataframe apply method (#1174) (0318764)
• Series sort_index and sort_values now raises when axis!=0 (#1294) (94bc2f2)

Documentation

• Add snippet to forecast future time series in the Forecast a single time series with a univariate model tutorial (#1271) (a687050)
• Update bigframes.pandas.Series docs (#1273) (0cac64f)

Features

• Add max_retries to TextEmbeddingGenerator and Claude3TextGenerator (#1259) (8077ff4)
• Bigframes.bigquery.parse_json (#1265) (27bbd80)
• Support DataFrame.astype(dict) (#1262) (5934f8e)

Bug Fixes

• Avoid global mutation in BigQueryOptions.client_endpoints_override (#1280) (788f6e9)
• Fix erroneous window bounds removal during compilation (#1163) (f91756a)

Dependencies

• Relax sqlglot upper bound (#1278) (c71ec09)

Documentation

• Add bq studio links that allows users to generate Jupiter notebooks in bq studio with github contents (#1266) (58f13cb)
• Add snippet to evaluate ARIMA plus model in the Forecast a single time series with a univariate model tutorial (#1267) (3dcae2d)
• Add snippet to see the ARIMA coefficients in the Forecast a single time series with a univariate model tutorial (#1268) (059a564)
• Update bigframes.pandas.pandas docstrings (#1247) (c4bffc3)
• Use 002 model for better scalability in text generation (#1270) (bb7a850)

Features

• Implement confirmation threshold for semantic operators (#1251) (5ba4511)

Bug Fixes

• Raise if trying to change ordering_mode after session has started (#1252) (8cfaae8)
• Reduce the number of labels added to query jobs (#1245) (fdcdc18)

Documentation

• Remove bq studio link (#1258) (dd4fd2e)
• Update bigframes.pandas.DatetimeMethods docstrings (#1246) (10f08da)
• Update semantic_operators.ipynb (#1260) (a2ed989)

Features

• Add GeoSeries.x and GeoSeries.y (#1126) (4c3548f)
• Add LinearRegression.predict_explain() to generate ML.EXPLAIN_PREDICT columns (#1190) (e13eca2)
• Add LogisticRegression.predict_explain() to generate ML.EXPLAIN_PREDICT columns (#1222) (bcbc732)
• Add write_engine parameter to read_FORMATNAME methods to control how data is written to BigQuery (#371) (ed47ef1)
• Add client side retry to GeminiTextGenerator (#1242) (8193abe)
• Add Gemini-pro-1.5 to GeminiTextGenerator Tuning and Support score() method in Gemini-pro-1.5 (#1208) (298fc73)
• Add support for LinearRegression.predict_explain and LogisticRegression.predict_explain parameter, top_k_features (#1228) (3068e19)
• Support dataframe where method (#1166) (71b4053)

Bug Fixes

• Arima model series input. (#1237) (f7d52d9)
• Json in struct destination type (#1187) (200c9bb)
• Throw an error message when setting is_row_processor=True to read a multi param function (#1160) (b2816a5)

Documentation

• Add an "open in BQ Studio" link to all BigFrames sample notebooks (#1223) (e0a8288)
• Add bq studio link for a new ipynb file called "bq_dataframes_template.ipynb" (#1239) (840aaff)
• Add example for logistic regression (#1240) (4d854fd)
• Add examples for ml PCA and SimpleImputer (#1236) (0d84459)
• Add KMeans example (#1234) (d87ab97)
• Add linear model example (#1235) (2c3e1fd)
• Add ml.model_selection examples (#1238) (50648e4)
• Add python snippet for "Create the time series model" section of the Forecast a single time series with a univariate model tutorial (#1227) (20f3190)

Features

• Add Gemini 2.0 preview text model support (#1209) (1021d57)

Documentation

• Add Gemini 2.0 text gen sample notebook (#1211) (9596b66)
• Update bigframes.pandas.index docs return types (#1191) (c63e7da)

Features

• (Series | DataFrame).plot.bar (#1152) (0fae2e0)
• bigframes.bigquery.vector_search supports use_brute_force and fraction_lists_to_search parameters (#1158) (131edc3)
• Add ARIMAPlus.predict_explain() to generate forecasts with explanation columns (#1177) (05f8b4d)
• Add client_endpoints_override to bq options (#1167) (be74b99)
• Add support for temporal types in dataframe's describe() method (#1189) (2d564a6)
• Allow join-free alignment of analytic expressions (#1168) (daef4f0)
• Series.isin supports bigframes.Series arg (#1195) (0d8a16b)
• Update llm.TextEmbeddingGenerator to 005 (#1186) (3072d38)

Bug Fixes

• Fix error loading local dataframes into bigquery (#1165) (5b355ef)
• Fix null index join with 'on' arg (#1153) (9015c33)
• Fix series.isin using local path always (#1202) (a44eafd)

Performance Improvements

• Update df.corr, df.cov to be used with more than 30 columns case. (#1161) (9dcf1aa)

Dependencies

• Remove ibis-framework by vendoring a fork of the package to bigframes_vendored. (#1170) (421d24d)

Documentation

• Add a code sample using bpd.options.bigquery.ordering_mode = "partial" (#909) (f80d705)
• Add snippet for creating boosted tree model (#1142) (a972668)
• Add snippet for evaluating a boosted tree model (#1154) (9d8970a)
• Add snippet for predicting classifications using a boosted tree model (#1156) (e7b83f1)
• Add third party pandas.Index methods and docstrings (#1171) (a970294)
• Fix Bigframes.Pandas.General_Function missing docs (#1164) (de923d0)
• Update bigframes.pandas.Index docstrings (#1144) (557ab8d)

Features

• Add astype(type, errors='null') to cast safely (#1122) (b4d17ff)

Bug Fixes

• Dataframe fillna with scalar. (#1132) (37f8c32)
• Exclude index columns from model fitting processes. (#1138) (8d4da15)
• Unordered mode too many labels issue. (#1148) (7216b21)

Documentation

• Document groupby.head and groupby.size methods (#1111) (a61eb4d)

Features

• Add basic geopandas functionality (#962) (3759c63)
• Support json_extract_string_array in the bigquery module (#1131) (4ef8bac)

Bug Fixes

• Fix Series.to_frame generating string label instead of int where name is None (#1118) (14e32b5)
• Update the API documentation with newly added rep (#1120) (72c228b)

Performance Improvements

• Reduce CURRENT_TIMESTAMP queries (#1114) (32274b1)
• Reduce dry runs from read_gbq with table (#1129) (f7e4354)

Documentation

• Add file for Classification with a Boosted Treed Model and snippet for preparing sample data (#1135) (7ac6639)
• Add snippet for Linear Regression tutorial Predict Outcomes section (#1101) (108f4a9)
• Update DataFrame docstrings to include the errors section (#1127) (a38d4c4)
• Update GroupBy docstrings (#1103) (9867a78)
• Update Session doctrings to include exceptions (#1130) (a870421)

Features

• Add the ground_with_google_search option for GeminiTextGenerator predict (#1119) (ca02cd4)
• Add warning when user tries to access struct series fields with __getitem__ (#1082) (20e5c58)
• Allow fit to take additional eval data in linear and ensemble models (#1096) (254875c)
• Support context manager for bigframes session (#1107) (5f7b8b1)

Performance Improvements

• Improve series.unique performance and replace drop_duplicates i… (#1108) (499f24a)

Features

• Support series items method (#1089) (245a89c)

Documentation

• Update docstrings of DataFrame and related files (#1092) (15e9fd5)

Features

• Support regional endpoints for more bigquery locations (#1061) (45b672a)
• Update LLM generators to warn user about model name instead of raising error. (#1048) (650d80d)

Bug Fixes

• Access MATERIALIZED_VIEW with read_gbq (#1070) (601e984)
• Correct zero row count in DataFrame from table view (#1062) (b536070)
• Fix generic error message when entering an incorrect column name (#1031) (5ac217d)
• Make explode respect the index labels (#1064) (99ca0df)
• Make invalid location warning case-insensitive (#1044) (b6cd55a)
• Remove palm2 test case from llm load test (#1063) (575a10a)
• Show warning for unknown location set through .ctor (#1052) (02c2da7)

Performance Improvements

• Reduce schema tracking overhead (#1056) (1c3879d)
• Repr generates fewer queries (#1046) (d204603)
• Speedup internal tree comparisons (#1060) (4379438)

Documentation

• Add docstring return type section to BigQueryOptions class (#964) (307385f)

Features

• Add deprecation warning to PaLM2TextGenerator model (#1035) (1183b0f)
• Add DeprecationWarning for PaLM2TextEmbeddingGenerator (#1018) (4af5bbb)
• Add ml.model_selection.cross_validate support (#1020) (1a38063)
• Allow access of struct fields with dot operators on Series (#1019) (ef76f13)

Bug Fixes

• Ensure no double execution for to_pandas (#1032) (4992cc2)
• Remove pre-caching of remote function results (#1028) (0359bc8)

Documentation

• Add ml cross-validation notebook (#1037) (057f3f0)

Features

• Add bigframes.bigquery.approx_top_count (#1010) (3263bd7)
• Add bigframes.ml.compose.SQLScalarColumnTransformer to create custom SQL-based transformations (#955) (1930b4e)
• Allow multiple columns input for llm models (#998) (2fe5e48)

Bug Fixes

• Fix repr caching with partial ordering (#1016) (208a984)

Documentation

• Limit pypi notebook to 7 days and add more info about differences with partial ordering mode (#1013) (3c54399)
• Move and edit existing linear-regression tutorial snippet (#991) (4cb62fd)

Features

• Add ml.model_selection.KFold class (#1001) (952cab9)
• Support bool and bytes types in describe(include='all') (#994) (cc48f58)
• Support ingress settings in remote_function (#1011) (8e9919b)

Bug Fixes

• Fix miscasting issues with case_when (#1003) (038139d)

Performance Improvements

• Join op discards child ordering in unordered mode (#923) (1b5b0ee)

Dependencies

• Update ibis version in prerelease tests (#1012) (f89785f)

Features

• Add "include" param to describe for string types (#973) (deac6d2)
• Add subset parameter to DataFrame.dropna to select which columns to consider (#981) (f7c03dc)

Bug Fixes

• DataFrameGroupby.agg now works with unnamed tuples (#985) (0f047b4)
• Fix a bug that raises exception when re-indexing columns with their original order (#988) (596b03b)
• Make the Series.apply outcome assignable to the original dataframe in partial ordering mode (#874) (c94ead9)

Dependencies

• Limit ibis-framework version to 9.2.0 (#989) (06c1b33)
• Update to ibis-framework 9.x and newer sqlglot (#827) (89ea44f)

Features

• Add __version__ alias to bigframes.pandas (#967) (9ce10b4)
• Add Gemini 1.5 stable models support (#945) (c1cde19)
• Allow setting table labels in to_gbq (#941) (cccc6ca)
• Define list accessor for bigframes Series (#946) (8e8279d)
• Enable read_csv() to process other files (#940) (3b35860)
• Include the bigframes package version alongside the feedback link in error messages (#936) (7b59b6d)

Bug Fixes

• Astype Decimal to Int64 conversion. (#957) (27764a6)
• Make read_gbq_function work for multi-param functions (#947) (c750be6)
• Support read_gbq_function for axis=1 application (#950) (86e54b1)

Documentation

• Add docstring returns section to Options (#937) (a2640a2)
• Update title of pypi notebook example to reflect use of the PyPI public dataset (#952) (cd62e60)

Features

• Add DataFrame.struct.explode to add struct subfields to a DataFrame (#916) (ad2f75e)
• Implement bigframes.bigquery.json_extract_array (#910) (575a29e)
• Recover struct column from exploded Series (#904) (7dd304c)

Bug Fixes

• Fix issue with iterating on >10gb dataframes (#949) (2b0f0fa)
• Improve Series.replace for dict input (#907) (4208044)
• NullIndex in ML model.predict error (#917) (612271d)
• Struct field non-nullable type issue. (#914) (149d5ff)
• Unordered mode errors in ml train_test_split (#925) (85d7c21)

Performance Improvements

• Improve repr performance (#918) (46f2dd7)

Dependencies

• Re-introduce support for numpy 1.24.x (#931) (3d71913)
• Update minimum support to Pandas 1.5.3 and Pyarrow 10.0.1 (#903) (7ed3962)

Documentation

• Add Claude3 ML and RemoteFunc notebooks (#930) (cfd16c1)
• Create sample notebook to manipulate struct and array data (#883) (3031903)
• Update struct examples. (#953) (d632cd0)
• Use unstack() from BigQuery DataFrames instead of pandas in the PyPI sample notebook (#890) (d1883cc)

Features

• Add llm.TextEmbeddingGenerator to support new embedding models (#905) (6bc6a41)
• Add ml.llm.Claude3TextGenerator model (#901) (7050038)

Documentation

• Add columns for "requires ordering/index" to supported APIs summary (#892) (d2fc51a)
• Remove duplicate description for kms_key_name (#898) (1053d56)
• Update embedding model notebooks (#906) (d9b8ef5)

Features

• Implement bigframes.bigquery.json_extract (#868) (3dbf84b)
• Implement Series.str.__getitem__ (#897) (e027b7e)

Bug Fixes

• Fix caching from generating row numbers in partial ordering mode (#872) (52b7786)

Performance Improvements

• Generate SQL with fewer CTEs (#877) (eb60804)
• Speed up compilation by reducing redundant type normalization (#896) (e0b11bc)

Documentation

• Add streaming html docs (#884) (171da6c)
• Fix the DisplayOptions doc rendering (#893) (3eb6a17)
• Update streaming notebook (#887) (6e6f9df)

Features

• df.apply(axis=1) to support remote function with mutiple params (#851) (2158818)
• Allow windowing in 'partial' ordering mode (#861) (ca26fe5)
• Create a separate OrderingModePartialPreviewWarning for more fine-grained warning filters (#879) (8753bdd)

Bug Fixes

• Fix issue with invalid sql generated by ml distance functions (#865) (9959fc8)

Documentation

• Create sample notebook using ordering_mode="partial" (#880) (c415eb9)
• Update streaming notebook (#875) (e9b0557)

Features

• Add bigframes-mode label to query jobs (#832) (c9eaff0)
• Add config option to set partial ordering mode (#855) (823c0ce)
• Add stratify param support to ml.model_selection.train_test_split method (#815) (27f8631)
• Add streaming.StreamingDataFrame class (#864) (a7d7197)
• Allow DataFrame.join for self-join on Null index (#860) (e950533)
• Support remote function cleanup with session.close (#818) (ed06436)
• Support to_csv/parquet/json to local files/objects (#858) (d0ab9cc)

Bug Fixes

• Fewer relation joins from df self-operations (#823) (0d24f73)
• Fix 'sql' property for null index (#844) (1b6a556)
• Fix unordered mode using ordered path to print frame (#839) (93785cb)
• Reduce redundant remote_function deployments (#856) (cbf2d42)

Documentation

• Add partner attribution steps to integrations sample notebook (#835) (d7b333f)
• Make get_global_session/close_session/reset_session appears in the docs (#847) (01d6bbb)

Documentation

• Remove session and connection in llm notebook (#821) (74170da)
• Remove the experimental flask icon from the public docs (#820) (067ff17)

Features

• Add .agg support for size (#792) (87e6018)
• Add bigframes.bigquery.json_set (#782) (1b613e0)
• Add bigframes.streaming.to_pubsub method to create continuous query that writes to Pub/Sub (#801) (b47f32d)
• Add DataFrame.to_arrow to create Arrow Table from DataFrame (#807) (1e3feda)
• Add PolynomialFeatures support to to_gbq and pipelines (#805) (57d98b9)
• Add Series.peek to preview data efficiently (#727) (580e1b9)
• Expose gcf memory param in remote_function (#803) (014765c)
• More informative error when query plan too complex (#811) (136dc24)

Bug Fixes

• Include internally required packages in remote_function hash (#799) (4b8fc15)

Documentation

• Document dtype limitation on row processing remote_function (#800) (487dff6)

Features

• Add dataframe.insert (#770) (e8bab68)
• Add groupby head API (#791) (44202bc)
• Add ml.preprocessing.PolynomialFeatures class (#793) (b4fbb51)
• Bigframes.streaming module for continuous queries (#703) (0433a1c)
• Include index columns in DataFrame.sql if they are named (#788) (c8d16c0)

Bug Fixes

• Allow __repr__ to work with uninitialed DataFrame/Series/Index (#778) (e14c7a9)
• Df.loc with the 2nd input as bigframes boolean Series (#789) (a4ac82e)
• Ensure numpy version matches in remote_function deployment (#798) (324d93c)
• Fix temp table creation retries by now throwing if table already exists. (#787) (0e57d1f)
• Self-join optimization doesn't needlessly invalidate caching (#797) (1b96b80)

Features

• Allow functions returned from bpd.read_gbq_function to execute outside of apply (#706) (ad7d8ac)
• Support bigquery.vector_search() (#736) (dad66fd)
• Support score() in GeminiTextGenerator (#740) (b2c7d8b)
• Support bytes type in remote_function (#761) (4915424)
• Support fit() in GeminiTextGenerator (#758) (d751f5c)

Bug Fixes

• ARIMAPlus loads auto_arima_min_order param (#752) (39d7013)
• Improve to_pandas_batches for large results (#746) (61f18cb)
• Resolve issue with unset thread-local options (#741) (d93dbaf)

Documentation

• Fix ML.EVALUATE spelling (#749) (7899749)
• Remove LogisticRegression normal_equation strategy (#753) (ea5d367)

Features

• merge only generates a default index if both inputs already have an index (#733) (25d049c)
• Add +, - as unary ops, ^ binary op (#724) (968d825)
• Add GroupBy.size() to get number of rows in each group (#479) (1fca588)
• Add DataFrame ~ operator (#721) (354abc1)
• Add GeminiText 1.5 Preview models (#737) (56cbd3b)
• Add slot_millis and add stats to session object (#725) (72e9583)
• Adds bigframes.bigquery.array_to_string to convert array elements to delimited strings (#731) (f12c906)
• Allow functions decorated with bpd.remote_function() to execute locally (#704) (d850da6)
• Ensure "bigframes-api" label is always set on jobs, even if the API is unknown (#722) (1832778)
• Support ml.SimpleImputer in bigframes (#708) (4c4415f)
• Support type annotations to supply input and output types to bpd.remote_function() decorator (#717) (4a12e3c)
• Support type annotations with bpd.remote_function() and axis=1 (a preview feature) (#730) (e5a2992)

Bug Fixes

• Correct index labels in multiple aggregations for DataFrameGroupBy (#723) (6a78c89)
• Fix Null index assign series to column (#711) (ffb4b57)
• Set bpd.remote_function()s input_types and output_types default to None to allow omitting them when type annotations are present (#729) (0e25a3b)
• Warn and disable time travel for linked datasets (#712) (085fa9d)

Performance Improvements

• Optimize dataframe-series alignment on axis=1 (#732) (3d39221)

Documentation

• Add examples to DataFrameGroupBy and SeriesGroupBy (#701) (e7da0f0)

Features

• read_gbq_query supports filters (9386373)
• read_gbq suggests a correct column name when one is not found (9386373)
• Add DefaultIndexKind.NULL to use as index_col in read_gbq*, creating an indexless DataFrame/Series (#662) (29e4886)
• Bigframes.bigquery.array_agg(SeriesGroupBy|DataFrameGroupby) (#663) (412f28b)
• To_datetime supports utc=False for string inputs (#579) (adf9889)

Bug Fixes

• read_gbq_table respects primary keys even when filters are set (#689) (9386373)
• Fix type error in test_cluster (#698) (14d81c1)
• Improve escaping of literals and identifiers (#682) (da9b136)
• Properly identify non-unique index in tables without primary keys (#699) (6e0f4d8)
• Remove a usage of the resource package when not available, such as on Windows (#681) (96243f2)
• The imported samples error and use peek() (#688) (1a0b744)

Performance Improvements

• Don't run query immediately from read_gbq_table if filters is set (9386373)
• Use a LIMIT clause when max_results is set (9386373)

Documentation

• Add code snippets for imported onnx tutorials (#684) (cb36e46)
• Add code snippets for imported tensorflow model (#679) (b02c401)
• Use class_weight="balanced" in the logistic regression prediction tutorial (#678) (b951549)

Features

• Add DataFrame.__delitem__ (#673) (2218c21)
• Add Series.case_when() (#673) (2218c21)
• Add strategy="quantile" in KBinsDiscretizer (#654) (c6c487f)
• Add Series.combine (#680) (2fd1b81)
• Series.str.split (#675) (6eb19a7)
• Suggest correct options in bpd.options.bigquery.location (#666) (57ccabc)
• Support axis=1 in df.apply for scalar outputs (#629) (f6bdc4a)
• Support gcf vpc connector in remote_function (#677) (9ca92d0)
• Warn with a more specific DefaultLocationWarning category when no location can be detected (#648) (e084e54)

Bug Fixes

• Include index_col when selecting columns and filters in read_gbq_table (#648) (e084e54)

Dependencies

• Add jellyfish as a dependency for spelling correction (57ccabc)

Documentation

• Add code snippets for llm text generatiion (#669) (93416ed)
• Add logistic regression samples (#673) (2218c21)
• Address lint errors in code samples (#665) (4fc8964)
• Document inlining of small data in read_* APIs (#670) (306953a)

Features

• bigframes.options and bigframes.option_context now uses thread-local variables to prevent context managers in separate threads from affecting each other (#652) (651fd7d)
• Add ARIMAPlus.coef_ property exposing ML.ARIMA_COEFFICIENTS functionality (#585) (81d1262)
• Add a unique session_id to Session and allow cleaning up sessions (#553) (c8d4e23)
• Add the bigframes.bigquery sub-package with a bigframes.bigquery.array_length function (#630) (9963f85)
• Always do a query dry run when option.repr_mode == "deferred" (#652) (651fd7d)
• Custom query labels for compute options (#638) (f561799)
• Warn with DefaultIndexWarning from read_gbq on clustered/partitioned tables with no index_col or filters set (#631, #658) (2715d2b, 73064dd)
• Support index_col=False in read_csv and engine="bigquery" (73064dd)
• Support gcf max instance count in remote_function (#657) (36578ab)

Bug Fixes

• Don't raise UnknownLocationWarning for US or EU multi-regions (#653) (8e4616b)
• Fix bug with na in the column labels in stack (#659) (4a34293)
• Use explicit session in PaLM2TextGenerator (#651) (e4f13c3)

Documentation

• Add python code sample for multiple forecasting time series (#531) (16866d2)
• Fix the Palm2TextGenerator output token size (#649) (c67e501)

Features

• Add .cache() method to persist intermediate dataframe (#626) (a5c94ec)
• Add transpose support for small homogeneously typed DataFrames. (#621) (054075d)
• Allow single input type in remote_function (#641) (3aa643f)
• Expose gcf max timeout in remote_function (#639) (dfeaad0)
• Series binary ops compatible with more types (#618) (518d315)
• Support the score method for PaLM2TextGenerator (#634) (3ffc1d2)

Bug Fixes

• Allow to_pandas to download more than 10GB (#637) (ce56495)
• Extend row hash to 128 bits to guarantee unique row id (#632) (9005c6e)
• Llm fine tuning tests (#627) (4724a1a)
• Llm palm score tests (#643) (cf4ec3a)

Performance Improvements

• Automatically condense internal expression representation (#516) (03c1b0d)
• Cache transpose to allow performant retranspose (#635) (44b738d)

Documentation

• Add supported pandas apis on the main page (#628) (8d2a51c)
• Add the first sample for the Single time-series forecasting from Google Analytics data tutorial (#623) (2b84c4f)
• Address more technical writers' feedback (#640) (1e7793c)

Features

• Add Series.struct.dtypes property (#599) (d924ec2)
• Add fine tuning fit() for Palm2TextGenerator (#616) (9c106bd)
• Add quantile statistic (#613) (bc82804)
• Expose max_batching_rows in remote_function (#622) (240a1ac)
• Support primary key(s) in read_gbq by using as the index_col by default (#625) (75bb240)
• Warn if location is set to unknown location (#609) (3706b4f)

Bug Fixes

• Address technical writers fb (#611) (9f8f181)
• Infer narrowest numeric type when combining numeric columns (#602) (8f9ece6)
• Use exact median implementation by default (#619) (9d205ae)

Documentation

• Fix rendering of examples for multiple apis (#620) (9665e39)
• Set index_cols in read_gbq as a best practice (#624) (70015b7)

Features

• Add hasnans, combine_first, update to Series (#600) (86e0f38)
• Add MultiIndex subclass. (#596) (5d0f149)
• Add pivot_table for DataFrame. (#473) (5f1d670)
• Add Series.autocorr (#605) (4ec8034)
• Support list of numerics in pandas.cut (#580) (290f95d)

Bug Fixes

• Address more technical writers feedback (#581) (4b08d92)
• Error for object dtype on read_pandas (#570) (8702dcf)
• Inverting int now does bitwise inversion rather than sign flip (#574) (5f1db8b)
• Loc setitem dtype issue. (#603) (b94bae9)
• Toc menu missing plotting name (#591) (eed12c1)

Documentation

• (Series|Dataframe).dtypes (#598) (edef48f)
• Add code samples for str accessor methdos (#594) (a557ea2)
• Add docs for DataFrame and Series dunder methods (#562) (8fc26c4)
• Add examples for at/iat (#582) (3be4a2e)

Features

• (Series|DataFrame).explode (#556) (9e32f57)
• Add DataFrame.eval and DataFrame.query (#361) (5e28ebd)
• Add ColumnTransformer save/load (#541) (9d8cf67)
• Add ml.metrics.mean_squared_error (#559) (853c25e)
• Add support for numpy expm1, log1p, floor, ceil, arctan2 ops (#505) (e8e66cf)
• Add transformers save/load (#552) (d805241)
• Allow DataFrame binary ops to align on either axis and with loc… (#544) (6d8f3af)
• Expose DataFrame.bqclient to assist in integrations (#519) (0be8911)
• Read_pandas accepts pandas Series and Index objects (#573) (f8821fe)
• Support ML.GENERATE_EMBEDDING in PaLM2TextEmbeddingGenerator (#539) (1156c1e)
• Support max_columns in repr and make repr more efficient (#515) (54e49cf)

Bug Fixes

• Assign NaN scalar to column error. (#513) (0a4153c)
• Don't download 100gb onto local python machine in load test (#537) (082c58b)
• Exclude list-like s parameter in plot.scatter (#568) (1caac27)
• Fix case where df.peek would fail to execute even with force=True (#511) (8eca99a)
• Fix error in Series.drop(0) (#575) (75dd786)
• Include all names in MultiIndex repr (#564) (b188146)
• Plot.scatter s parameter cannot accept float-like column (#563) (8d39187)
• Product operation produces float result for all input types (#501) (6873b30)
• Reloaded transformer .transform error (#569) (39fe474)
• Rename PaLM2TextEmbeddingGenerator.predict output columns to be backward compatible (#561) (4995c00)
• Respect hard stack size limit and swallow limit change exception. (#558) (4833908)
• Restore string to date/time type coercion (#565) (4ae0262)
• Sync the notebook with embedding changes (#550) (347f2dd)
• Use bytes limit on frame inlining rather than element count (#576) (659a161)

Performance Improvements

• Add multi-query execution capability for complex dataframes (#427) (d2d7e33)

Dependencies

• Include pyarrow as a dependency (#529) (9b1525a)

Documentation

• bigframes.options.bigquery.project and location are optional in some circumstances (#548) (90bcec5)
• Add "Supported pandas APIs" reference to the documentation (#542) (74c3915)
• Add General Availability banner to README (#507) (262ff59)
• Add opeartions in API docs (#557) (ea95761)
• Add progress_bar code sample (#508) (92a1af3)
• Add the code samples for metrics{auc, roc_auc_score, roc_curve} (#520) (5f37b09)
• Address more comments from technical writers to meet legal purposes (#571) (9084df3)
• Fix docs of ARIMAPlus.predict (#512) (3b80f95)
• Include Index in table-of-contents (#564) (b188146)
• Mark Gemini model as Pre-GA (#543) (769868b)
• Migrate the overview page to Bigframes official landing page (#536) (a0fb8bb)

⚠ BREAKING CHANGES

• rename model parameter min_rel_progress to tol
• early_stop setting no longer supported, always uses True
• rename model parameter n_parallell_trees to n_estimators
• rename class_weights to class_weight
• rename learn_rate to learning_rate
• PCA n_components supports float value and None, default to None
• rename various ml model parameters for consistency with sklearn (https://github.com/googleapis/python-bigquery-dataframes/pull/491)

Features

• Add configuration option to read_gbq (#401) (85cede2)
• Add ml ARIMAPlus model params (#488) (352cb85)
• Add ml KMeans model params (#477) (23a8d9a)
• Add ml LogisticRegression model params (#481) (f959b65)
• Add ml PCA model params (#474) (fb5d83b)
• Add params for LinearRegression model (#464) (21b2188)
• Add support for Python 3.12 (#231) (df2976f)
• Allow assigning directly to Series.name property (#495) (ad0e99e)
• Ensure Series.str.len() can get length of array columns (#497) (10c0446)
• Option to use bq connection without check (#460) (0b3f8e5)
• PCA n_components supports float value and None, default to None (65c6f47)
• Rename class_weights to class_weight (65c6f47)
• Rename learn_rate to learning_rate (65c6f47)
• Rename model parameter min_rel_progress to tol (65c6f47)
• Rename model parameter n_parallell_trees to n_estimators (65c6f47)
• Rename various ml model parameters for consistency with sklearn (https://github.com/googleapis/python-bigquery-dataframes/pull/491) (65c6f47)
• Support BQ regional endpoints for europe-west9, europe-west3, us-east4, and us-west1 (#504) (fbada4a)
• Support dataframe.cov (#498) (c4beafd)
• Support Series.dt.floor (#493) (2dd01c2)
• Support Series.dt.normalize (#483) (0bf1e91)
• Update plot sample to 1000 rows (#458) (60d4a7b)

Bug Fixes

• early_stop setting no longer supported, always uses True (65c6f47)
• Fix -1 offset lookups failing (#463) (2dfb9c2)
• Plot.scatter c argument functionalities (#494) (d6ee994)
• Properly support format param for numerical input. (#486) (ae20c35)
• Renable to_csv and to_json related tests (#468) (2b9a01d)
• Sampling plot cannot preserve ordering if index is not ordered (#475) (a5345fe)
• Use actual BigQuery types rather than ibis types in to_pandas (#500) (82b4f91)

Dependencies

• Support pandas 2.2 (#492) (e2cf50e)

Documentation

• Add code samples for metrics.{accuracy_score, confusion_matrix} (#478) (3e3329a)
• Add code samples for metrics.{recall_score, precision_score, f11_score} (#502) (370fe90)
• Improve API documentation (#489) (751266e)
• Update bigquery connection documentation (#499) (4bfe094)
• Update LLM + K-means notebook to handle partial failures (#496) (97afad9)

⚠ BREAKING CHANGES

• exclude remote models for .register() (#465)

Features

• (Series|DataFrame).plot (#438) (1c3e668)
• read_gbq_table supports LIKE as a operator in filters (#454) (d2d425a)
• Add DataFrame.pipe() method (#421) (95f5a6e)
• Set force=True by default in DataFrame.peek() (#469) (4e8e97d)
• Support datetime related casting in (Series|DataFrame|Index).astype (#442) (fde339b)
• Support Series.dt.strftime (#453) (8f6e955)

Bug Fixes

• Any() on empty set now correctly returns False (#471) (f55680c)
• Df.drop_na preserves columns dtype (#457) (3bab1a9)
• Disable to_json and to_csv related tests (#462) (874026d)
• Exclude remote models for .register() (#465) (73fe0f8)
• Fix broken link in covid notebook (#450) (adadb06)
• Fix broken multiindex loc cases (#467) (b519197)
• Fix grouping series on multiple other series (#455) (3971bd2)
• Groupby aggregates no longer check if grouping keys are numeric (#472) (4fbf938)
• Raise ValueError when read_pandas() receives a bigframes DataFrame (#447) (b28f9fd)
• Series.(to_csv|to_json) leverages bq export (#452) (718a00c)
• Warn when read_gbq / read_gbq_table uses the snapshot time cache (#441) (e16a8c0)

Documentation

• Add code samples for ml.metrics.r2_score (#459) (85fefa2)
• Add the docs for loc and iloc indexers (#446) (14ab8d8)
• Add the pages for at and iat indexers (#456) (340f0b5)
• Add version information to bug template (#437) (91bd39e)
• Indicate that project and location are optional in example notebooks (#451) (1df0140)

Features

• (Series|DataFrame).plot.(line|area|scatter) (#431) (0772510)
• Support CMEK for remote_function cloud functions (#430) (2fd69f4)

⚠ BREAKING CHANGES

• read_parquet uses a "pandas" engine to parse files by default. Use engine="bigquery" for the previous behavior

Features

• (Series|Dataframe).plot.hist() (#420) (4aadff4)
• Add detect_anomalies to ml ARIMAPlus and KMeans models (#426) (6df28ed)
• Add engine parameter to read_parquet (#413) (31325a1)
• Add ml PCA.detect_anomalies method (#422) (8d82945)
• Support BYOSA in remote_function (#407) (d92ced2)
• Support CMEK for BQ tables (#403) (9a678e3)

Bug Fixes

• Move third_party.bigframes_vendored to bigframes_vendored (#424) (763edeb)
• Only do row identity based joins when joining by index (#356) (76b252f)
• Read_pandas inline respects location (#412) (ae0e3ea)

Documentation

• Add predict sample to samples/snippets/bqml_getting_started_test.py (#388) (6a3b0cc)
• Document minimum IAM requirement (#416) (36173b0)
• Fix the note rendering for DataFrames methods: nlargest, nsmallest (#417) (38bd2ba)

Features

• Add ml.metrics.pairwise.euclidean_distance (#397) (1726588)
• Add TextEmbedding model version support (#394) (e0f1ab0)

Bug Fixes

• Code exception in remote_function now prevents retry and surfaces in the client (#387) (dd3643d)
• Docs link for metrics.pairwise (#400) (a60aba7)

Dependencies

• Update ibis to version 8.0.0 and refactor remote_function to use ibis UDF method (#277) (350499b)

Documentation

• Update README to point to new summary pages (#402) (bfe2b23)

⚠ BREAKING CHANGES

• rename cosine_similarity to paired_cosine_distances (#393)
• move model optional args to kwargs (#381)

Features

• Add DataFrames.corr() method (#379) (67fd434)
• Add ml.metrics.pairwise.manhattan_distance (#392) (9d31865)
• Enable regional endpoints for me-central2 (#386) (469674d)

Bug Fixes

• Avoid ibis warning for "database" table() method argument (#390) (a0490a4)
• Correct the numeric literal dtype (#365) (93b02cd)
• Rename cosine_similarity to paired_cosine_distances (#393) (81ece46)

Performance Improvements

• Inline read_pandas for small data (#383) (59b446b)

Dependencies

• Add minimum version constraint for sqlglot to 19.9.0 (#389) (8b62d77)

Documentation

• Add a code sample for creating a kmeans model (#267) (4291d65)
• Fix bigframes.pandas.concat documentation (#382) (234b61c)

Miscellaneous Chores

• Release 0.22.0 (#396) (8f73d9e)

Code Refactoring

• Move model optional args to kwargs (#381) (4037992)

Features

• Add Series.cov method (#368) (443db22)
• Add ml.llm.GeminiTextGenerator model (#370) (de1e0a4)
• Add ml.metrics.pairwise.cosine_similarity function (#374) (126f566)
• Add XGBoostModel (#363) (d5518b2)
• Limited support of lambdas in Series.apply (#345) (208e081)
• Support bigframes.pandas.to_datetime for scalars, iterables and series. (#372) (ffb0d15)
• Support read_gbq wildcard table path (#377) (90caf86)

Bug Fixes

• Error message fix. (#375) (930cf6b)

Documentation

• Clarify ADC pre-auth in a non-interactive environment (#348) (99a9e6e)

Performance Improvements

• Make repr cache the block where appropriate (#350) (068879f)

Documentation

• Add a sample to demonstrate the evaluation results (#364) (cff0919)
• Fix the DataFrame.apply code sample (#366) (1866a26)

Features

• Add DataFrame.peek() as an efficient alternative to head() results preview (#318) (9c34d83)
• Add ARIMA_EVAULATE options in forecasting models (#336) (73e997b)
• Add Index constructor, repr, copy, get_level_values, to_series (#334) (e5d054e)
• Improve error message for drive based BQ table reads (#344) (0794788)
• Update cut to work without labels = False and show intervals as dict (#335) (4ff53db)

Bug Fixes

• Chance default connection name in getting_started.ipnyb (#347) (677f014)
• Series iteration correctly returns values instead of index (#339) (2c6af9b)

Documentation

• Add code samples for Series.{between, cumprod} (#353) (09a52fd)

Bug Fixes

• Read_gbq large response issue (#332) (b8178b9)
• Use object dtype for ARRAY columns in to_pandas() with pandas 1.x (#329) (374ddb5)

Documentation

• Add DataFrame.applymap documentation (#326) (bd531a1)
• Add code samples for series methods (#323) (32cc6fa)
• Add remote model requirements (#333) (c91f70c)

Bug Fixes

• Handle multi-level columns for df aggregates properly (#305) (5bb45ba)
• Update max_output_token limitation. (#308) (5cccd36)

Documentation

• Add code samples for Series.corr (#316) (9150c16)

Features

• Add 'columns' as an alias for 'col_order' (#298) (a01b271)
• Add Series dt.tz and dt.unit properties (#303) (2e1a403)
• Add to_gbq() method for LLM models (#299) (dafbc1b)
• Allow manually set clustering_columns in dataframe.to_gbq (#302) (9c21323)
• Support assigning to columns like a property (#304) (f645c56)
• Support upcasting numeric columns in concat (#294) (e3a056a)

Bug Fixes

• DF.drop tuple input as multi-index (#301) (21391a9)
• Fix bug converting non-string labels to sql ids (#296) (a61c5fe)

Documentation

• Add code samples for Series.ffill and DataFrame.ffill (#307) (1c63b45)

Features

• Add dataframe.to_html (#259) (2cd6489)
• Add IntervalIndex support to bigframes.pandas.cut (#254) (6c1969a)
• Add replace method to DataFrame (#261) (5092215)
• Specific pyarrow mappings for decimal, bytes types (#283) (a1c0631)

Bug Fixes

• Dataframes to_gbq now creates dataset if it doesn't exist (#222) (bac62f7)
• Exclude pandas 2.2.0rc0 to unblock prerelease tests (#292) (ac1a745)
• Fix DataFrameGroupby.agg() issue with as_index=False (#273) (ab49350)
• Make Series.str.replace work for simple strings (#285) (ad67465)
• Update dataframe.to_gbq to dedup column names. (#286) (746115d)
• Use setuptools.find_namespace_packages (#246) (9ec352a)

Dependencies

• Migrate to ibis-framework >= "7.1.0" (#53) (9798a2b)

Documentation

• Add code snippets for explore query result page (#278) (7cbbb7d)
• Code samples for astype common to DataFrame and Series (#280) (95b673a)
• Code samples for DataFrame.copy and Series.copy (#290) (7cbc2b0)
• Code samples for drop and fillna (#284) (9c5012e)
• Code samples for isna, isnull, dropna, isin (#289) (ad51035)
• Code samples for rename , size (#293) (eb69f60)
• Code samples for reset_index and sort_values (#282) (acc0eb7)
• Code samples for sample, get, Series.round (#295) (c2b1892)
• Code samples for Series.{add, replace, unique, T, transpose} (#287) (0e1bbfc)
• Code samples for Series.{map, to_list, count} (#290) (7cbc2b0)
• Code samples for Series.{name, std, agg} (#293) (eb69f60)
• Code samples for Series.groupby and Series.{sum,mean,min,max} (#280) (95b673a)
• Code samples for DataFrame set_index, items (#295) (c2b1892)
• Fix the rendering for get_dummies (#291) (252f3a2)

Features

• Add filters argument to read_gbq for enhanced data querying (#198) (034f71f)
• Add module/class level api tracking (#272) (4f3db3d)
• Deprecate use_regional_endpoints (#199) (319a1f2)

Bug Fixes

• Increase recursion limit, cache compilation tree hashes (#184) (b54791c)
• Replaced raise NotImplementedError with return NotImplemented (#258) (a133822)

Documentation

• Add code samples for values and value_counts (#249) (f247d95)
• Add sample for getting started with BQML (#141) (fb14f54)

Features

• Add ARIMAPlus.predict parameters (#264) (99598c7)
• Add DataFrame from_dict and from_records methods (#244) (8d81e24)
• Add DataFrame.select_dtypes method (#242) (1737acc)
• Add nunique method to Series/DataFrameGroupby (#256) (c8ec245)
• Support dataframe.loc with conditional columns selection (#233) (3febea9)

Bug Fixes

• Enfore pandas version requirement <2.1.4 (#265) (9dd63f6)
• Exclude pandas 2.1.4 from prerelease tests to unblock e2e tests (b02fc2c)
• Fix value_counts column label for normalize=True (#245) (d3fa6f2)
• Migrate e2e tests to bigframes-load-testing project (8766ac6)
• Ml.sql logic (#262) (68c6fdf)
• Update the llm_kmeans notebook (#247) (66d1839)

Documentation

• Add code samples for shape and head (#257) (5bdcc65)
• Add example for dataframe.melt, dataframe.pivot, dataframe.stac… (#252) (8c63697)
• Add example to dataframe.nlargest, dataframe.nsmallest, datafra… (#234) (e735412)
• Add examples for dataframe.cummin, dataframe.cummax, dataframe.cumsum, dataframe.cumprod (#243) (0523a31)
• Add examples for dataframe.nunique, dataframe.diff, dataframe.a… (#251) (77074ec)
• Correct the docs for option_context (#263) (d21c6dd)
• Correct the params rendering for ml.remote and ml.ensemble modules (#248) (c2829e3)
• Fix return annotation in API docstrings (#253) (89a1c67)

⚠ BREAKING CHANGES

• model.predict returns all the columns (#204)

Features

• Add info and memory_usage methods to dataframe (#219) (9d6613d)
• Add remote vertex model support (#237) (0bfc4fb)
• Add the recent api method for ML component (#225) (ed8876d)
• Model.predict returns all the columns (#204) (416171a)
• Send warnings on LLM prediction partial failures (#216) (81125f9)

Bug Fixes

• Add df snapshots lookup for read_gbq (#229) (d0d9b84)
• Avoid unnecessary row_number() on sort key for io (#211) (a18d40e)
• Dedup special character (#209) (dd78acb)
• Invalid JSON type of the notebook (#215) (a729831)
• Make to_pandas override enable_downsampling when sampling_method is manually set. (#200) (ae03756)
• Polish the llm+kmeans notebook (#208) (e8532b1)
• Update the llm+kmeans notebook with recent change (#236) (f8917ab)
• Use anonymous dataset to create remote_function (#205) (69b016e)

Documentation

• Add code samples for index and column properties (#212) (c88d38e)
• Add code samples for df reshaping, function, merge, and join methods (#203) (010486c)
• Add examples for dataframe.kurt, dataframe.std, dataframe.count (#232) (f9c6e72)
• Add examples for dataframe.mean, dataframe.median, dataframe.va… (#228) (edd0522)
• Add examples for dataframe.min, dataframe.max and dataframe.sum (#227) (3a375e8)
• Code samples for Series.dot and DataFrame.dot (#226) (b62a07a)
• Code samples for Series.where and Series.mask (#217) (52dfad2)
• Code samples for dataframe.any, dataframe.all and dataframe.prod (#223) (d7957fa)
• Make the code samples reflect default bq connection usage (#206) (71844b0)

Miscellaneous Chores

• Release 0.15.0 (#241) (6c899be)

Bug Fixes

• Correctly handle null values when initializing fingerprint ordering (#210) (8324f13)

Documentation

• Add an example notebook about line graphs (#197) (f957b27)

Features

• Add 'cross' join support (#176) (765446a)
• Add 'index', 'pad', 'nearest' interpolate methods (#162) (6a28403)
• Add series.sample (identical to existing dataframe.sample) (#187) (37914a4)
• Add unordered sql compilation (#156) (58f420c)
• Log most recent API calls as recent-bigframes-api-xx labels on BigQuery jobs (#145) (4ea33b7)
• Read_gbq creates order deterministically without table copy (#191) (8ab81de)
• Support date_series.astype("string[pyarrow]") to cast DATE to STRING (#186) (aee0e8e)
• Support series.at[row_label] = scalar (#173) (0c8bd33)
• Temporary resources no longer use BigQuery Sessions (#194) (4a02cac)

Bug Fixes

• All sort operation are now stable (#195) (3a2761f)
• Default to 7 days expiration for read_csv, read_json, read_parquet (#193) (03606cd)
• Deprecate the remote_service_type in llm model (#180) (a8a409a)
• For reset_index on unnamed multiindex, always use level_[n] label (#182) (f95000d)
• Match pandas behavior when assigning listlike to empty dfs (#172) (c1d1f42)
• Use anonymous dataset instead of session dataset for temp tables (#181) (800d44e)
• Use random table for read_pandas (#192) (741c75e)
• Use random table when loading data for read_csv, read_json, read_parquet (#175) (9d2e6dc)

Documentation

• Add code samples for read_gbq_function using community UDFs (#188) (7506eab)
• Add docstring code samples for Series.apply and DataFrame.map (#185) (c816d84)
• Add llm kmeans notebook as an included example (#177) (d49ae42)
• Use head() to get top n results, not to preview results (#190) (87f84c9)

Features

• to_gbq without a destination table writes to a temporary table (#158) (e1817c9)
• Add DataFrame.__iter__, DataFrame.iterrows, DataFrame.itertuples, and DataFrame.keys methods (#164) (c065071)
• Add Series.__iter__ method (#164) (c065071)
• Add interpolate() to series and dataframe (#157) (b9cb55c)
• Support 32k text-generation and multilingual embedding models (#161) (5f0ea37)

Bug Fixes

• Update default temp table expiration to 7 days (#174) (4ff26cd)

Features

• Add DataFrame.melt (#113) (4e4409c)
• Add DataFrame.to_pandas_batches() to download large DataFrame objects (#136) (3afd4a3)
• Add bigframes.options.compute.maximum_bytes_billed option that sets maximum bytes billed on query jobs (#133) (63c7919)
• Add pandas.qcut (#104) (8e44518)
• Add pd.get_dummies (#149) (d8baad5)
• Add unstack to series, add level param (#115) (5edcd19)
• Implement operator @ for DataFrame.dot (#139) (79a638e)
• Populate ibis version in user agent (#140) (c639a36)

Bug Fixes

• Don't override the global logging config (#138) (2ddbf74)
• Fix bug with column names under repeated column assignment (#150) (29032d0)
• Resolve plotly rendering issue by using ipython html for job pro… (#134) (39df43e)
• Use indexee's session for loc listlike cases (#152) (27c5725)

Documentation

• Add artithmetic df sample code (#153) (ac44ccd)
• Fix indentation on read_gbq_function code sample (#163) (0801d96)
• Link to ML.EVALUATE BQML page for score() methods (#137) (45c617f)

Features

• Add back reset_session as an alias for close_session (#124) (694a85a)
• Change query parameter to query_or_table in read_gbq (#127) (f9bb3c4)

Bug Fixes

• Expose bigframes.pandas.reset_session as a public API (#128) (b17e1f4)
• Use series's own session in series.reindex listlike case (#135) (95bff3f)

Documentation

• Add runnable code samples for DataFrames I/O methods and property (#129) (6fea8ef)
• Add runnable code samples for reading methods (#125) (a669919)

Features

• Implement DataFrame.dot for matrix multiplication (#67) (29dd414)

⚠ BREAKING CHANGES

• rename bigframes.pandas.reset_session to close_session (#101)

Features

• Add bigframes.options.bigquery.application_name for partner attribution (#117) (52d64ff)
• Add AtIndexer getitems (#107) (752b01f)
• Rename bigframes.pandas.reset_session to close_session (#101) (36693bf)
• Send BigQuery cancel request when canceling bigframes process (#103) (e325fbb)
• Support external packages in remote_function (#98) (ec10c4a)
• Use ArrowDtype for STRUCT columns in to_pandas (#85) (9238fad)

Bug Fixes

• Support multiindex for three loc getitem overloads (#113) (68e3cd3)

Performance Improvements

• If primary keys are defined, read_gbq avoids copying table data (#112) (e6c0cd1)

Documentation

• Add documentation for Series.struct.field and Series.struct.explode (#114) (a6dab9c)
• Add open-source link in API doc (#106) (db51fe3)
• Update ML overview API doc (#105) (1b3f3a5)

⚠ BREAKING CHANGES

• The default behavior of to_parquet is changing from no compression to 'snappy' compression.

Features

• Support compression in to_parquet (a8c286f)

Bug Fixes

• Create session dataset for remote functions only when needed (#94) (1d385be)

Features

• Add aliases for several series properties (#80) (c0efec8)
• Add equals methods to series/dataframe (#76) (636a209)
• Add iat and iloc accessing by tuples of integers (#90) (228aeba)
• Add level param to DataFrame.stack (#88) (97b8bec)
• Allow df.drop to take an index object (#68) (740c451)
• Use default session connection (#87) (4ae4ef9)

Bug Fixes

• Change the invalid url in docs (#93) (969800d)

Documentation

• Add more preprocessing models into the docs menu. (#97) (1592315)

Features

• Add df.unstack (#63) (4a84714)
• Add idxmin, idxmax to series, dataframe (#74) (781307e)
• Add ml.preprocessing.KBinsDiscretizer (#81) (24c6256)
• Add multi-column dataframe merge (#73) (c9fa85c)
• Add update and align methods to dataframe (#57) (bf050cf)
• Support STRUCT data type with Series.struct.field to extract child fields (#71) (17afac9)

Bug Fixes

• Avoid 403 response too large to return error with read_gbq and large query results (#77) (8f3b5b2)
• Change return type of Series.loc[scalar] (#40) (fff3d45)
• Fix df/series.iloc by list with multiindex (#79) (971d091)