Changes to the Snowpark Python API¶

Added support for displaying source code as comments in the generated scripts when registering stored procedures. This is enabled by default, turn off by specifying source_code_display=False at registration.

Added a parameter if_not_exists when creating a UDF, UDTF or Stored Procedure from Snowpark Python to ignore creating the specified function or procedure if it already exists.

Accept integers when calling snowflake.snowpark.functions.get to extract value from array.

Added functions.reverse in functions to open access to Snowflake built-in function REVERSE.

Added parameter require_scoped_url in snowflake.snowflake.files.SnowflakeFile.open() (in Private Preview) to replace is_owner_file, which is marked for deprecation.

Fixed a bug that overwrote paramstyle to qmark when creating a Snowpark session.

Fixed a bug where df.join(..., how="cross") fails with SnowparkJoinException: (1112): Unsupported using join type 'Cross'.

Fixed a bug where querying a DataFrame column created from chained function calls used a wrong column name.

Added asc, asc_nulls_first, asc_nulls_last, desc, desc_nulls_first, desc_nulls_last, date_part, and unix_timestamp in functions.

Added the property DataFrame.dtypes to return a list of column name and data type pairs.

Added the following aliases:

• functions.expr() for functions.sql_expr().

• functions.date_format() for functions.to_date().

• functions.monotonically_increasing_id() for functions.seq8().

• functions.from_unixtime() for functions.to_timestamp().

Fixed a bug in SQL simplifier that didn’t handle Column alias and join well in some cases. See https://github.com/snowflakedb/snowpark-python/issues/658 for details.

Fixed a bug in SQL simplifier that generated wrong column names for function calls, NaN and INF.

The session parameter PYTHON_SNOWPARK_USE_SQL_SIMPLIFIER will be True after Snowflake 7.3 is released. In snowpark-python, session.sql_simplifier_enabled reads the value of PYTHON_SNOWPARK_USE_SQL_SIMPLIFIER by default, meaning that the SQL simplfier is enabled by default after the Snowflake 7.3 release. To turn this off, set PYTHON_SNOWPARK_USE_SQL_SIMPLIFIER in Snowflake to False or run session.sql_simplifier_enabled = False from Snowpark. It is recommended to use the SQL simplifier because it helps to generate more concise SQL.

Added Session.generator() to create a new DataFrame using the GENERATOR table function.

Added the SECURE parameter to the functions that create a secure UDF or UDTF.

Added new APIs for async job:

• Session.create_async_job() to create an AsyncJob instance from a query id.

• AsyncJob.result() now accepts the argument result_type to return the results in different formats.

• AsyncJob.to_df() returns a DataFrame built from the result of this asynchronous job.

• AsyncJob.query() returns the SQL text of the executed query.

DataFrame.agg() and RelationalGroupedDataFrame.agg() now accept variable-length arguments.

Added parameters lsuffix and rsuffix to DataFrame.join() and DataFrame.cross_join() to conveniently rename overlapping columns.

Added Table.drop_table() so you can drop the temp table after calling DataFrame.cache_result(). Table is also a context manager, so you can use the with statement to drop the cache temp table after use.

Added Session.use_secondary_roles().

Added functions first_value() and last_value(). (contributed by @chasleslr)

Added on as an alias for using_columns and how as an alias for join_type in DataFrame.join().

Fixed a bug in Session.create_dataframe() that raised an error when schema names had special characters.

Fixed a bug in which options set in Session.read.option() were not passed to DataFrame.copy_into_table() as default values.

Fixed a bug in which DataFrame.copy_into_table() raised an error when a copy option had single quotes in the value.

Session.add_packages() now raises a ValueError when the version of a package cannot be found in Snowflake Anaconda channel. Previously, Session.add_packages() succeeded and a SnowparkSQLException exception was raised later in the UDF or stored procedure registration step.

Added method FileOperation.get_stream() to support downloading stage files as a stream.

Added support in functions.ntiles() to accept an int argument.

Added the following aliases:

• functions.call_function() for functions.call_builtin().

• functions.function() for functions.builtin().

• DataFrame.order_by() for DataFrame.sort()

• DataFrame.orderBy() for DataFrame.sort()

Improved DataFrame.cache_result() to return a more accurate Table class instead of a DataFrame class.

Added support to allow session as the first argument when calling StoredProcedure.

Improved nested query generation by flattening queries when applicable. This improvement can be enabled by setting Session.sql_simplifier_enabled = True. DataFrame.select(), DataFrame.with_column(), DataFrame.drop() and other select-related APIs have more flattened SQL now. DataFrame.union(), DataFrame.union_all(), DataFrame.except_(), DataFrame.intersect(), DataFrame.union_by_name() have flattened SQL generated when multiple set operators are chained.

Improved type annotations for async job APIs.

Fixed a bug in which Table.update(), Table.delete(), Table.merge() tried to reference a temp table that did not exist.

Added experimental APIs for evaluating Snowpark dataframes with asynchronous queries:

• Added keyword argument block to the following action APIs on Snowpark dataframes (which execute queries) to allow asynchronous evaluations:

  • DataFrame.collect(), DataFrame.to_local_iterator(), DataFrame.to_pandas(), DataFrame.to_pandas_batches(), DataFrame.count(), DataFrame.first(), DataFrameWriter.save_as_table(), DataFrameWriter.copy_into_location(), Table.delete(), Table.update(), Table.merge().

• Added method DataFrame.collect_nowait() to allow asynchronous evaluations.

• Added class AsyncJob to retrieve results from asynchronously executed queries and check their status.

Added support for table_type in Session.write_pandas(). You can now choose from these table_type options: temporary, temp, and transient.

Added support for using Python structured data (list, tuple and dict) as literal values in Snowpark.

Added keyword argument execute_as to functions.sproc() and session.sproc.register() to allow registering a stored procedure as a caller or owner.

Added support for specifying a pre-configured file format when reading files from a stage in Snowflake.

Added support for displaying details of a Snowpark session.

Fixed a bug in which DataFrame.copy_into_table() and DataFrameWriter.save_as_table() mistakenly created a new table if the table name was fully qualified, and the table already existed.

Deprecated keyword argument create_temp_table in Session.write_pandas().

Deprecated invoking UDFs using arguments wrapped in a Python list or tuple. You can use variable-length arguments without a list or tuple.

Updated snowflake-connector-python to 2.7.12.

Added support for displaying source code as comments in the generated scripts when registering UDFs. This feature is turned on by default. To turn it off, pass the new keyword argument source_code_display as False when calling register() or @udf().

Added support for calling table functions from DataFrame.select(), DataFrame.with_column(), and DataFrame.with_columns() which now take parameters of type table_function.TableFunctionCall for columns.

Added keyword argument overwrite to session.write_pandas() to allow you to overwrite contents of a Snowflake table with that of a Pandas DataFrame.

Added keyword argument column_order to df.write.save_as_table() to specify the matching rules when inserting data into a table in append mode.

Added method FileOperation.put_stream() to upload local files to a stage via a file stream.

Added methods TableFunctionCall.alias() and TableFunctionCall.as_() to allow aliasing the names of columns that come from the output of table function joins.

Added function get_active_session() in module snowflake.snowpark.context to get the current active Snowpark session.

Improved the function function.uniform() to infer the types of inputs max_ and min_ and cast the limits to IntegerType or FloatType, respectively.

Fixed a bug in which batch insert should not raise an error when statement_params is not passed to the function.

Fixed a bug in which column names should be quoted when session.create_dataframe() is called with dicts and a given schema.

Fixed a bug in which creation of a table should be skipped if the table already exists and is in append mode when calling df.write.save_as_table().

Fixed a bug in which third-party packages with underscores cannot be added when registering UDFs.

Added keyword only argument statement_params to the following methods to allow for specifying statement level parameters:

• collect, to_local_iterator, to_pandas, to_pandas_batches, count, copy_into_table, show, create_or_replace_view, create_or_replace_temp_view, first, cache_result and random_split on class snowflake.snowpark.Dateframe.

• update, delete and merge on class snowflake.snowpark.Table.

• save_as_table and copy_into_location on class snowflake.snowpark.DataFrameWriter.

• approx_quantile, statement_params, cov and crosstab on class snowflake.snowpark.DataFrameStatFunctions.

• register and register_from_file on class snowflake.snowpark.udf.UDFRegistration.

• register and register_from_file on class snowflake.snowpark.udtf.UDTFRegistration.

• register and register_from_file on class snowflake.snowpark.stored_procedure.StoredProcedureRegistration.

• udf, udtf and sproc in snowflake.snowpark.functions.

Added support for Column as an input argument to session.call().

Added support for table_type in df.write.save_as_table(). You can now choose from these table_type options: temporary, temp, and transient.

Added validation of object name in session.use_* methods.

Updated the query tag in SQL to escape it when it contains special characters.

Added a check to see if Anaconda terms are acknowledged when adding missing packages.

Fixed the limited length of the string column in session.create_dataframe().

Fixed a bug in which session.create_dataframe() mistakenly converted 0 and False to None when the input data was only a list.

Fixed a bug in which calling session.create_dataframe() using a large local dataset sometimes created a temp table twice.

Aligned the definition of function.trim() with the SQL function definition.

Fixed an issue where snowpark-python would hang when using the Python system-defined (built-in function) sum vs. the Snowpark function.sum().

Added support for user-defined table functions (UDTFs).

• Use function snowflake.snowpark.functions.udtf() to register a UDTF, or use it as a decorator to register the UDTF.

• You can also use Session.udtf.register() to register a UDTF.

• Use Session.udtf.register_from_file() to register a UDTF from a Python file.

Updated APIs to query a table function, including both Snowflake built-in table functions and UDTFs.

• Use function snowflake.snowpark.functions.table_function() to create a callable representing a table function and use it to call the table function in a query.

• Alternatively, use function snowflake.snowpark.functions.call_table_function() to call a table function.

• Added support for the over clause, which specifies partition by and order by when lateral joining a table function.

• Updated Session.table_function() and DataFrame.join_table_function() to accept TableFunctionCall instances.

When creating a function with functions.udf() and functions.sproc(), you can now specify an empty list for the imports or packages argument to indicate that no import or package is used for this UDF or stored procedure. Previously, specifying an empty list meant that the function would use session-level imports or packages.

Improved the __repr__ implementation of data types in types.py. The unused type_name property has been removed.

Added a Snowpark-specific exception class for SQL errors. This replaces the previous ProgrammingError from the Python connector.

Added a lock to a UDF or UDTF when it is called for the first time per thread.

Improved the error message for pickling errors that occurred during UDF creation.

Included the query ID when logging the failed query.

Fixed a bug in which non-integral data (such as timestamps) was occasionally converted to integer when calling DataFrame.to_pandas().

Fixed a bug in which DataFrameReader.parquet() failed to read a parquet file when its column contained spaces.

Fixed a bug in which DataFrame.copy_into_table() failed when the dataframe is created by reading a file with inferred schemas.

Session.flatten() and DataFrame.flatten().

Restricted the version of cloudpickle <= 2.0.0.

Added support for the vectorized UDFs via Python UDF Batch API. The Python UDF batch API enables defining Python functions that receive batches of input rows as Pandas DataFrames and return batches of results as Pandas arrays or Series. This can improve the performance of UDFs in Snowpark.

Added support for inferring the schema of a DataFrame by default when it is created by reading a Parquet, Avro, or ORC file in the stage.

Added functions current_session(), current_statement(), current_user(), current_version(), current_warehouse(), date_from_parts(), date_trunc(), dayname(), dayofmonth(), dayofweek(), dayofyear(), grouping(), grouping_id(), hour(), last_day(), minute(), next_day(), previous_day(), second(), month(), monthname(), quarter(), year(), current_database(), current_role(), current_schema(), current_schemas(), current_region(), current_avaliable_roles(), add_months(), any_value(), bitnot(), bitshiftleft(), bitshiftright(), convert_timezone(), uniform(), strtok_to_array(), sysdate(), time_from_parts(), timestamp_from_parts(), timestamp_ltz_from_parts(), timestamp_ntz_from_parts(), timestamp_tz_from_parts(), weekofyear(), percentile_cont() to snowflake.snowflake.functions.

Added support for creating an empty DataFrame with a specific schema using the Session.create_dataframe() method.

Changed the logging level from INFO to DEBUG for several logs (e.g., the executed query) when evaluating a dataframe.

Improved the error message when failing to create a UDF due to pickle errors.

Removed the following APIs that were deprecated in 0.4.0: DataFrame.groupByGroupingSets(), DataFrame.naturalJoin(), DataFrame.joinTableFunction, DataFrame.withColumns(), Session.getImports(), Session.addImport(), Session.removeImport(), Session.clearImports(), Session.getSessionStage(), Session.getDefaultDatabase(), Session.getDefaultSchema(), Session.getCurrentDatabase(), Session.getCurrentSchema(), Session.getFullyQualifiedCurrentSchema().

Added typing-extension as a new dependency with the version >= 4.1.0.

Removed pandas hard dependencies in the Session.create_dataframe() method.

Added stored procedures API.

Added Session.sproc property and sproc() to snowflake.snowpark.functions, so you can register stored procedures.

Added Session.call to call stored procedures by name.

Added UDFRegistration.register_from_file() to allow registering UDFs from Python source files or zip files directly.

Added UDFRegistration.describe() to describe a UDF.

Added DataFrame.random_split() to provide a way to randomly split a dataframe.

Added functions md5(), sha1(), sha2(), ascii(), initcap(), length(), lower(), lpad(), ltrim(), rpad(), rtrim(), repeat(), soundex(), regexp_count(), replace(), charindex(), collate(), collation(), insert(), left(), right(), endswith() to snowflake.snowpark.functions.

The call_udf() function now also accepts literal values.

Provided a distinct keyword in array_agg().

Fixed an issue that caused DataFrame.to_pandas() to have a string column if Column.cast(IntegerType()) was used.

Fixed a bug in DataFrame.describe() when there is more than one string column.

You can now specify which Anaconda packages to use when defining UDFs.

Added add_packages(), get_packages(), clear_packages(), and remove_package() to class Session.

Added add_requirements() to Session so you can use a requirements file to specify which packages this session will use.

Added parameter packages to function snowflake.snowpark.functions.udf() and method UserDefinedFunction.register() to indicate UDF-level Anaconda package dependencies when creating a UDF.

Added parameter imports to snowflake.snowpark.functions.udf() and UserDefinedFunction.register() to specify UDF-level code imports.

Added a parameter session to function udf() and UserDefinedFunction.register() so you can specify which session to use to create a UDF if you have multiple sessions.

Added types Geography and Variant to snowflake.snowpark.types to be used as type hints for Geography and Variant data when defining a UDF.

Added support for Geography geoJSON data.

Added Table, a subclass of DataFrame for table operations.

Methods update and delete update and delete rows of a table in Snowflake.

Method merge merges data from a DataFrame to a Table.

Overrided method DataFrame.sample() with an additional parameter seed, which works on tables but not on views and sub-queries.

Added DataFrame.to_local_iterator() and DataFrame.to_pandas_batches() to allow getting results from an iterator when the result set returned from the Snowflake database is too large.

Added DataFrame.cache_result() for caching the operations performed on a DataFrame in a temporary table. Subsequent operations on the original DataFrame have no effect on the cached result DataFrame.

Added property DataFrame.queries to get SQL queries that will be executed to evaluate the DataFrame.

Added Session.query_history() as a context manager to track SQL queries executed on a session, including all SQL queries to evaluate DataFrames created from a session. Both query ID and query text are recorded.

You can now create a Session instance from an existing established snowflake.connector.SnowflakeConnection. Use parameter connection in Session.builder.configs().

Added use_database(), use_schema(), use_warehouse(), and use_role() to class Session to switch database/schema/warehouse/role after a session is created.

Added DataFrameWriter.copy_into_table() to unload a DataFrame to stage files.

Added DataFrame.unpivot().

Added Column.within_group() for sorting the rows by columns with some aggregation functions.

Added functions listagg(), mode(), div0(), acos(), asin(), atan(), atan2(), cos(), cosh(), sin(), sinh(), tan(), tanh(), degrees(), radians(), round(), trunc(), and factorial() to snowflake.snowpark.functions.

Added an optional argument ignore_nulls in function lead() and lag().

The condition parameter of function when() and iff() now accepts SQL expressions.

All function and method names have been renamed to use the snake case naming style, which is more Pythonic. For convenience, some camel case names are kept as aliases to the snake case APIs. It is recommended to use the snake case APIs.

Deprecated these methods on class Session and replaced them with their snake case equivalents: getImports(), addImports(), removeImport(), clearImports(), getSessionStage(), getDefaultSchema(), getDefaultSchema(), getCurrentDatabase(), and getFullyQualifiedCurrentSchema().

Deprecated these methods on class DataFrame and replaced them with their snake case equivalents: groupingByGroupingSets(), naturalJoin(), withColumns(), and joinTableFunction().

Property DataFrame.columns is now consistent with DataFrame.schema.names and the Snowflake database identifier requirements.

Column.__bool__() now raises a TypeError. This will ban the use of logical operators and, or, not on Column object. For example, col("a") > 1 and col("b") > 2 will raise a TypeError. Use (col("a") > 1) & (col("b") > 2) instead.

Changed PutResult and GetResult to subclass NamedTuple.

Fixed a bug which raised an error when the local path or stage location has a space or other special characters.

Changed DataFrame.describe() so that non-numeric and non-string columns are ignored instead of raising an exception.

Updated snowflake-connector-python to 2.7.4.

Added Column.isin() with an alias Column.in_().

Added Column.try_cast(), which is a special version of cast(). It tries to cast a string expression to other types and returns null if the cast is not possible.

Added Column.startswith() and Column.substr() to process string columns.

Column.cast() now also accepts a str value to indicate the cast type in addition to a DataType instance.

Added DataFrame.describe() to summarize the stats of a DataFrame.

Added DataFrame.explain() to print the query plan of a DataFrame.

DataFrame.filter() and DataFrame.select_expr() now accept a SQL expression.

Added a new bool parameter called create_temp_table to methods DataFrame.saveAsTable() and Session.write_pandas() to optionally create a temp table.

Added DataFrame.minus() and DataFrame.subtract() as aliases to DataFrame.except_().

Added regexp_replace(), concat(), concat_ws(), to_char(), current_timestamp(), current_date(), current_time(), months_between(), cast(), try_cast(), greatest(), least(), and hash() to the snowflake.snowpark.functions module.

Fixed an issue where Session.createDataFrame(pandas_df) and Session.write_pandas(pandas_df) raised an exception when the Pandas DataFrame had spaces in the column name.

Fixed an issue where DataFrame.copy_into_table() sometimes erroneously printed an error level log entry.

Fixed an API documentation issue where some DataFrame APIs were missing from the documentation.

Updated snowflake-connector-python to 2.7.2, which upgrades the pyarrow dependency to 6.0.x. Refer to the Python connector 2.7.2 release notes for more information.

Added the createDataFrame() method for creating a DataFrame from a Pandas DataFrame.

Added the write_pandas() method for writing a Pandas DataFrame to a table in Snowflake and getting a Snowpark DataFrame object back.

Added new classes and methods for calling window functions.

Added the new functions cume_dist(), to find the cumulative distribution of a value with regard to other values within a window partition, and row_number(), which returns a unique row number for each row within a window partition.

Added functions for computing statistics for DataFrames in the DataFrameStatFunctions class.

Added functions for handling missing values in a DataFrame in the DataFrameNaFunctions class.

Added new methods: rollup(), cube(), and pivot() to the DataFrame class.

Added the GroupingSets class, which you can use with the DataFrame groupByGroupingSets method to perform a SQL GROUP BY GROUPING SETS.

Added the new FileOperation(session) class that you can use to upload and download files to and from a stage.

Added the copy_into_table() method for loading data from files in a stage into a table.

In CASE expressions, the functions when and otherwise now accept Python types in addition to Column objects.

When you register a UDF you can now optionally set the replace parameter to True to overwrite an existing UDF with the same name.

UDFs are now compressed before they are uploaded to the server. This makes them about 10 times smaller, which can help when you are using large ML model files.

When the size of a UDF is less than 8196 bytes, it will be uploaded as in-line code instead of uploaded to a stage.

Fixed an issue where the statement df.select(when(col("a") == 1, 4).otherwise(col("a"))), [Row(4), Row(2), Row(3)] raised an exception.

Fixed an issue where df.toPandas() raised an exception when a DataFrame was created from large local data.